Open Collections

UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Analysis of the management of uncertainty in long-term planning for electric utilities Irvine, Laura Jean 2017

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Item Metadata


24-ubc_2017_february_irvine_laura.pdf [ 1.03MB ]
JSON: 24-1.0340723.json
JSON-LD: 24-1.0340723-ld.json
RDF/XML (Pretty): 24-1.0340723-rdf.xml
RDF/JSON: 24-1.0340723-rdf.json
Turtle: 24-1.0340723-turtle.txt
N-Triples: 24-1.0340723-rdf-ntriples.txt
Original Record: 24-1.0340723-source.json
Full Text

Full Text

Analysis of the management of uncertainty inlong-term planning for electric utilitiesbyLaura Jean IrvineA thesis submitted in partial fulfilment of the requirements forthe degree ofMaster of Applied ScienceinThe Faculty of Graduate and Postdoctoral Studies(Civil Engineering)The University of British Columbia(Vancouver)January 2017c© Laura Jean Irvine, 2017AbstractElectric utilities engaging in integrated resource planning face a variety ofuncertainties which complicate the development of robust plans. These un-certainties occur in variables such as demand growth, energy price, greenhouse gas regulations, and water inflows for hydroelectric-dominated utili-ties, just to name a few. This study examines the current planning methodsin use among (largely North American) utilities with a particular focus on thefeatures of each method that manage or mitigate uncertainty. The two mostcommon planning methods (portfolio-based and scenario-based planning) areanalysed and their advantages, disadvantages, potential alterations, and cir-cumstances of best application are evaluated. These findings are then appliedto the case of BC Hydro, one of the largest electric utilities in Canada, withrecommendations for changes to their current planning process.iiPrefaceThis dissertation is work carried out by the author Laura J. Irvine under thesupervision of Prof. Ziad Shawwash. Chapter 1 and Chapter 2 were pre-pared as an internal report for BC Hydro’s Energy Planning and GenerationResource Management groups. A version of Chapter 1 is also intended forpresentation at the 2017 HydroVision Conference in Denver, USA. Chapter3 is prepared as a manuscript for submission to a journal in energy planningand policy. As Chapter 3 draws together all the findings and implicationsfrom Chapters 1 and 2, some sections in Chapter 3 are modifications of pre-vious sections.Prof. Shawwash contributed to editing of all chapters and preparation ofChapter 3 as a potential journal paper. Sanjaya de Zoysa from BC Hydro’sEnergy Planning group provided guidance on utility selection, BC Hydro’sIRP process, and manuscript editing for Chapter 1. Wun Kin Cheng andDoug Robinson from BC Hydro’s Generation Resource Management groupexplained the structure of the HYSIM and GOM models and their use in theIRP, which was incorporated in Chapter 2.iiiTable of ContentsAbstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iiPreface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iiiTable of Contents . . . . . . . . . . . . . . . . . . . . . . . . . . . . ivList of Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . viiList of Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ixList of Abbreviations . . . . . . . . . . . . . . . . . . . . . . . . . . xiAcknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . xiiIntroduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11 Literature review on current uncertainty management inthe energy industry . . . . . . . . . . . . . . . . . . . . . . . . . 31.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31.2 Utilities reviewed . . . . . . . . . . . . . . . . . . . . . . . . . 41.2.1 Arizona Public Service . . . . . . . . . . . . . . . . . . 41.2.2 Idaho Power . . . . . . . . . . . . . . . . . . . . . . . . 81.2.3 Los Angeles Department of Water and Power . . . . . 111.2.4 Northwest Power and ConservationCouncil . . . . . . . . . . . . . . . . . . . . . . . . . . 13iv1.2.5 PacifiCorp . . . . . . . . . . . . . . . . . . . . . . . . . 161.2.6 Public Service Company of Colorado . . . . . . . . . . 191.2.7 Tacoma Power . . . . . . . . . . . . . . . . . . . . . . . 231.2.8 Tennessee Valley Authority . . . . . . . . . . . . . . . 271.2.9 Californian IOUs and Energy Context . . . . . . . . . 301.2.10 International utilities . . . . . . . . . . . . . . . . . . . 361.3 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 382 BC Hydro’s approach to uncertainty in the 2013 IRP . . . 392.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 392.2 Summary of recommendations . . . . . . . . . . . . . . . . . . 402.3 Portfolio development . . . . . . . . . . . . . . . . . . . . . . . 412.3.1 Optimisation model . . . . . . . . . . . . . . . . . . . . 422.4 Alternatives for portfolio construction . . . . . . . . . . . . . . 492.5 Portfolio testing . . . . . . . . . . . . . . . . . . . . . . . . . . 522.6 Evaluation of model outputs . . . . . . . . . . . . . . . . . . . 552.6.1 Recommended metrics . . . . . . . . . . . . . . . . . . 572.6.2 Scorecards . . . . . . . . . . . . . . . . . . . . . . . . . 672.6.3 Trade-off analysis . . . . . . . . . . . . . . . . . . . . . 692.7 Summary and conclusion . . . . . . . . . . . . . . . . . . . . . 703 Practical methods of considering uncertainty in integratedresource planning for hydropower systems . . . . . . . . . . . 723.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 723.1.1 Structure of paper . . . . . . . . . . . . . . . . . . . . 733.2 Literature review . . . . . . . . . . . . . . . . . . . . . . . . . 743.2.1 Common uncertainties faced by utilities . . . . . . . . 753.2.2 Utility approaches to planning under uncertainty . . . 763.2.3 Initial steps for both planning methods . . . . . . . . . 783.2.4 Modelling . . . . . . . . . . . . . . . . . . . . . . . . . 793.2.5 Portfolio-based planning . . . . . . . . . . . . . . . . . 81v3.2.6 Scenario-based planning . . . . . . . . . . . . . . . . . 843.2.7 Assessment criteria for portfolios . . . . . . . . . . . . 873.3 Application to the BC Hydro system . . . . . . . . . . . . . . 923.3.1 Recommendations for BC Hydro’s IRP process basedon the results of this study . . . . . . . . . . . . . . . . 953.4 Conclusions and policy implications . . . . . . . . . . . . . . . 954 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103viList of Tables1.1 APS’s energy mix . . . . . . . . . . . . . . . . . . . . . . . . . 61.2 Idaho Power’s energy mix . . . . . . . . . . . . . . . . . . . . 91.3 LADWP’s energy mix . . . . . . . . . . . . . . . . . . . . . . 121.4 PacifiCorp’s energy mix . . . . . . . . . . . . . . . . . . . . . 171.5 Public Service Company of Colorado’s energy mix . . . . . . . 191.6 Tacoma Power’s energy mix . . . . . . . . . . . . . . . . . . . 241.7 TVA’s energy mix . . . . . . . . . . . . . . . . . . . . . . . . . 271.8 PG&E’s energy mix . . . . . . . . . . . . . . . . . . . . . . . . 331.9 SCE’s energy sources . . . . . . . . . . . . . . . . . . . . . . . 352.1 The portfolio construction scheme for dynamic programming . 502.2 Capacity of each resource in a portfolio multiplied by the ap-propriate multiplier for use in flexibility calculations . . . . . . 642.3 Summary of recommended metrics for portfolio comparison . . 652.4 Example of a scorecard for comparing portfolios using metricsand calculations . . . . . . . . . . . . . . . . . . . . . . . . . . 682.5 Example of a scorecard for comparing portfolios using metrics,with the metric rankings instead of the calculated numbers . . 682.6 Example of a score card using weights to reflect utility priori-ties among metrics . . . . . . . . . . . . . . . . . . . . . . . . 683.1 Uncertainties considered by the utilities . . . . . . . . . . . . . 773.2 Choice of planning process for studied utilities . . . . . . . . . 78vii3.3 Example of a scorecard for comparing portfolios using metricsand calculations . . . . . . . . . . . . . . . . . . . . . . . . . . 913.4 Example of a scorecard for comparing portfolios using metrics,with the metric rankings instead of the calculated numbers . . 913.5 Example of a score card using weights to reflect utility priori-ties among metrics . . . . . . . . . . . . . . . . . . . . . . . . 913.6 Example of a BC Hydro resource portfolio (BC Hydro 2013IRP) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97viiiList of Figures1.1 A general IRP process . . . . . . . . . . . . . . . . . . . . . . 51.2 APS’s IRP process . . . . . . . . . . . . . . . . . . . . . . . . 71.3 Idaho Power’s IRP process . . . . . . . . . . . . . . . . . . . . 101.4 NWPCC’s IRP process . . . . . . . . . . . . . . . . . . . . . . 131.5 PacifiCorp’s IRP process . . . . . . . . . . . . . . . . . . . . . 201.6 Public Service Company of Colorado’s IRP process . . . . . . 211.7 Tacoma Power’s 2015 IRP process . . . . . . . . . . . . . . . . 251.8 TVA’s IRP process . . . . . . . . . . . . . . . . . . . . . . . . 292.1 Scenario variables and their potential values in BC Hydro’s2013 IRP . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 442.2 BC Hydro’s current portfolio development process . . . . . . . 482.3 The IRP process using a dynamic programming model in placeof System Optimizer . . . . . . . . . . . . . . . . . . . . . . . 512.4 An example of dynamic programming selecting between vari-ous portfolios and optimising the overall resource selection . . 512.5 The inputs and outputs of a HYSIM run . . . . . . . . . . . . 552.6 The inputs and outputs of a GOM run . . . . . . . . . . . . . 562.7 A BC Hydro portfolio . . . . . . . . . . . . . . . . . . . . . . 612.8 Total capacity for each resource in the IRP over the planningperiod . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 612.9 Inputs and models required to calculate the recommendedmetrics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 662.10 Updated portfolio development and assessment process . . . . 71ix3.1 Overview of portfolio-based planning process . . . . . . . . . . 823.2 Overview of scenario-based planning process . . . . . . . . . . 853.3 BC Hydro’s generation and transmission system (BC Hydro) . 943.4 BC Hydro’s current portfolio development process . . . . . . . 963.5 Suggested portfolio development and assessment process . . . 96xList of AbbreviationsArizona Public Service (APS)Demand-side management (DSM)Duke Energy Indiana (DEI)Energy efficiency (EE)Greenhouse gas (GHG)Los Angeles Department of Water and Power (LADWP)Northwest Power and Conservation Council (NWPCC)NorthWestern Energy (NWE)Pacific Gas & Electric (PG&E)Public Service Company of Colorado (PSCC)San Diego Gas & Electric (SDG&E)Southern California Edison (SCE)Tacoma Public Utility / Tacoma Power (TPU)Tennessee Valley Authority (TVA)xiAcknowledgementsI would like first to express my deep gratitude to my research supervisor,Prof. Ziad Shawwash, for his guidance and support throughout my studiesand for the opportunity to gain a broader picture of this interesting field.I am appreciative of the funding of my research from grants provided toProf. Shawwash by BC Hydro. I am also grateful to those at BC Hydro whoshared their expertise. In particular I would like to thank Sanjaya de Zoysa,who shared his insights on BC Hydro’s IRP process and provided guidanceon the development of what would become chapter one of this thesis. Iwould also like to thank Doug Robinson for his explanations of the GOMmodel, and Wun Kin Cheng, who was endlessly patient in his assistancewith understanding and running the HYSIM model.xiiIntroductionBackgroundElectricity planning, as any planning, involves making decisions with uncer-tain information about the future. Projections based on historical data arenot completely accurate representations of the future and can fail to capturethe emergence of new factors. Because of the essential nature of electricityservices, utilities have to provide reliable service to their customers in spiteof their own uncertainty.Reliability of supply comes from accurate capacity planning on the part ofutilities. This requires knowledge of future energy demand as well as planningto ensure supply keeps pace with this load. To match capacity expansionto load growth, utilities are required to make judgements about the futurevalues of variables that affect load growth and resource acquisition. Thesevariables are factors like population growth and economic growth (which gointo determining the load), and energy price, GHG taxes and prices, uptakeof demand-side management, and regulation on particular fuel sources, all ofwhich affect which potential resources are best for meeting new load. Utilitieshandle this uncertainty in their planning in a variety of ways. This thesiswill investigate current practices for managing uncertainty in electric utilityplanning, examine how the planning steps mitigate or manage uncertainty,and develop a formal framework for evaluating the robustness of a capacityexpansion plan in the face of uncertainty. This framework will then be applied1to the specific case of BC Hydro. The goals for this research are: fullerunderstanding of current uncertainty management in practice; identificationof both good practice and shortfalls in current planning; and assessment ofBC Hydro’s planning method, with the additional aim of improving theircurrent uncertainty management in the IRP.Structure of thesisThis thesis is structured in the following manner. Chapter 1 presents a litera-ture review of the current state of IRP practice in North America, examiningthe published IRPs of fifteen utilities. Each utility’s practice is explained,any unique features are highlighted, and uncertainty management in eachmethod is identified. Chapter 2, building on the literature review, analysesin depth BC Hydro’s IRP process, looking at their optimisation models, sim-ulation models, and sensitivity analysis, and recommending changes to theprocess to better consider the effect of uncertainty on the portfolio develop-ment. Chapter 3 then presents conclusions on the content of both precedingchapters, identifying the underlying drivers for choice of planning method,the explicit consideration of uncertainty in each step of planning, the prosand cons of each method for uncertainty management, and a brief summaryof recommendations for BC Hydro.2Chapter 1Literature review on currentuncertainty management in theenergy industryThis section reviews the current practice of uncertainty management amonga number of North American utilities and was originally prepared as a reportfor BC Hydro. The references cited in this section have been included in thethesis bibliography.1.1 IntroductionBC Hydro carries out an integrated resource plan (IRP) for their long termcapacity expansion every five years. This research was commissioned by BCHydro to identify the most common practices in utility resource planning, inparticular management of uncertainty, with a view to providing BC Hydrowith more methods for managing their uncertainty in their planning process.While other jurisdictions do practice IRP (South Africa[1], Queensland [2],among others) Canada and particularly the USA had the most examples ofIRP. As a result, the report largely focuses on utilities in these countries.3Across North America different jurisdictions have different requirements forIRP planning, with some states requiring it and some having abolished it.The individual utilities reviewed were chosen for a variety of reasons. PublicService Company of Colorado, PacifiCorp, and Arizona Public Service werechosen for review based on a paper on best practice published by the Reg-ulatory Assistance Project [3]. Tennessee Valley Authority was reviewed asan example of a public utility that is wholly owned by the United States gov-ernment. Tacoma Public Utility and Idaho Power are smaller utilities thathave somewhat different planning processes. Finally, the Californian utilitiesPG&E, SCE, and LADWP were included to give a perspective on planningprocess in a partially deregulated energy market.The terminology used throughout to describe planning techniques is de-rived from Hirst and Schweitzer[4]. Scenario analysis is identified as being atechnique in which alternative versions of the future are developed, combina-tions of resources that perform best in each future are selected by a model,and the best options are combined into a complete plan. Portfolio analysisis defined as planning in which multiple future resource portfolios are devel-oped, each corresponding to a specific company objective, and then modelled,analysed, and assessed against potential futures. A general overview of anIRP process is illustrated in Figure Utilities reviewedThe following sections summarise the main features of the planning processesof the reviewed utilities.1.2.1 Arizona Public ServiceArizona Public Service (APS) serves 1.2 million customers in the state ofArizona [5]. As of 2014, the company had 8,124 MW of generating capacityheavily based on fossil fuels and nuclear power (Table 1.1). APS files their4Figure 1.1: A general IRP processState or governmentrequirement for IRPDetermineadditional capacityrequiredBuild resource portfoliosSensitivityanalysisDecision toolsMetricsPoliciesRegulationsPreferred portfolioLoad growthExisting capacityGas priceEnergy priceScenariosResource optionsAdditionalscenariosAdditional variablesMonte Carlosimulation5Table 1.1: APS’s energy mixEnergy source PercentageNuclear 28Coal 38Natural gas 24Renewables 5EE/DSM 5Total 100EE: Energy efficiencyDSM: Demand side managementIRP, which has a planning horizon of 15 years, with the Arizona CorporationCommission every two years. The IRP also contains the 2014-2018 ActionPlan, which outlines the steps to be taken in the near-term to implementthe IRP recommendations over the full planning period. Figure 1.2 outlinesAPS’s IRP process.APS forecasts peak load growth of 3 percent per year over the fifteen yearplanning horizon. Weather, population growth, economic trends, and energyconsumption patterns are used to create the load forecast using PROVIEW,a module of the Strategist model from Ventyx/ABB. The IRP considers threeplanning forecasts: a current path with 3 percent average growth per year; alow load growth path with 1.6 percent average growth per year; and a highload growth scenario with growth of 4.2 percent per year.APS handles most of its uncertainty by using multiple scenarios and arange of deterministic forecasts. Examples include scenarios such as retiringcoal or setting a higher renewable energy goal than currently required bylaw. The utility uses a total of six scenarios in their optimisation model,PROVIEW, to develop their resource portfolios. The variables that changebetween scenarios include load forecast, gas prices, power prices, inflation,renewable energy regulations, carbon prices, and tax incentives for technol-ogy. The process is traditional deterministic optimisation; PROVIEW is6Figure 1.2: APS’s IRP processArizona PublicUtilities CommissionIRP requirementAssess needsBuild resource portfoliosSensitivityanalysisDecision toolsMetricsPoliciesRegulationsPreferred portfolioShort term action planExisting capacityPlanned butnot-yet-completedgenerationLoad, gas/energy prices,DSM forecasts5 scenariosResource optionsincluding DSMBase-caseassumptionsAdditionalscenariosAdditional variablesMonte Carlosimulation7given the existing resources and a scenario with various constraints and al-lowed to compile an optimal portfolio from a range of potential resourceswith the aim of minimising cost. Once the six portfolios have been created,the four lowest-cost ones are chosen for sensitivity analysis with the PRO-MOD model. APS used several key metrics to select a preferred portfoliofrom those generated by the model. These were fuel diversity, portfolio cost(both net present value of revenue requirements as well as average systemgeneration cost), cumulative capital expenditures, natural gas used, carbondioxide emissions, and water use. These criteria were useful for evaluatingportfolio performance in terms of concerns other than cost. The eventualpreferred portfolio can be chosen based on its stable performance across allthe metrics, even if it does not outperform every other portfolio on everymetric.1.2.2 Idaho PowerIdaho Power operates in the states of Idaho and Oregon [6]. The investor-owned utility services about 515,000 customers in southern Idaho and easternOregon and a generation capacity of 3,954 MW [7] (see Table 1.2 for a break-down of sources). The utility is required to file an IRP with both the OregonPublic Utility Commission and the Idaho Public Utility Commission everytwo years.Idaho Power has historically been a summer-peaking utility due to de-mand from irrigation pumps and air conditioning. Their load forecast forthe IRP is developed by Moodys Analytics, Inc., and is based on regionaland national economic activity, population forecasts, employment levels, andhistorical energy consumption patterns. Because of uncertainty, Idaho Poweruses an expected case (median) forecast as well as two additional forecasts(70th percentile and 90th percentile) to capture most of the expected vari-ability.Idaho Power does not use an optimisation model to construct resource8Table 1.2: Idaho Power’s energy mixEnergy source PercentageHydro 36Coal 34Natural gas and diesel 7Market purchases 8Power Purchase Agreements (PPAs)–Wind 9.8–Biomass 0.6–Hydro 2.4–Natural gas 0.5–Waste 0.3–Geothermal 1.4Total 100portfolios (see Figure 1.3 for overview of process). Instead, their portfoliosare constructed manually to meet the supply-demand gap. This is carriedout in discussion with stakeholders and guided by the company’s planningobjectives such as reduced use of coal. In total, 23 portfolios were created andanalysed. Idaho Power is currently most concerned about potential carbonprices and regulatory changes, so the most recent IRP portfolios all featuresome level of coal retirement. Manual selection of portfolios allows IdahoPower to focus on what they perceive to be their greatest risks, allowing directmanagement of uncertainty. The costs of the portfolios were simulated overthe 20-year planning period using AuroraXMP with base case assumptions.Sensitivity analysis of each portfolio was carried out by varying one ofthree variables: natural gas price, load, and hydroelectric variability. Thesevariables were given log-normal or normal distributions and run with theportfolio over 100 iterations. Also included in the sensitivity analysis waslevel of compliance with Section 111(d) of the Clean Air Act (CAA(, whichregulates carbon dioxide emissions. Based on the results, the 11 lowest cost9Figure 1.3: Idaho Power’s IRP processIdaho Public Utilities CommissionOregon Public Utilities CommissionAssess needsBuild resource portfoliosSensitivityanalysisDecision toolsMetricsPoliciesRegulationsPreferred portfolioExisting capacityLoad growth forecastPrice forecastsResource screeningResource optionsBase-caseassumptionsAdditionalscenariosLevels of compliancewith the CAAMonte Carlosimulation10portfolios were chosen for further scrutiny. This analysis focused on thestandard deviations of the cost of each portfolio over their 100 iterations.The portfolios with the lowest standard deviation changes over the 100 it-erations were considered to be least susceptible to large year-to-year swingsand therefore were deemed to be more robust choices. Tipping-point analysiswas carried out for several of the best pairs of scenarios to see how capitalprice changes would affect the choice of one portfolio over the other. Basedon these results, a preferred portfolio was eventually chosen.1.2.3 Los Angeles Department of Water and PowerThe LADWP is the largest municipal electricity utility in the United States[8]. The utility has 7,640 MW of owned capacity largely focused on coal(Table 1.3), and provides power to about 1.5 million customers in Los An-geles and the Owens Valley [9]. Unlike investor owned utilities (IOUs) likeSCE and PGE, LADWP remains vertically integrated, owning and operatingthe bulk of its generation, transmission, and distribution systems. Becauseof this, LADWP follows an IRP process rather than the long-term procure-ment planning (LTPP) process mandated for IOUs, and their planning con-sequently has much in common with utilities in non-regulated jurisdictions.LADWP divides customers into service categories when developing theirload forecast. Econometric models are used for load forecasting for residen-tial, commercial and industrial customer classes, while trend models are usedfor intradepartmental, street lighting, and Owen Valley customer classes.The utility also considers how electric vehicles may affect their load, anduses the California Energy Commission’s (CEC) forecast for this customerclass. The forecasts from these methods are modified by LADWP to re-flect their programs in EE and DSM. A combined load forecast is developedfrom this input and used for the rest of the planning process. LADWP buildsportfolios manually and then runs simulations of scenarios with the Planningand Risk model from Ventyx. Five portfolios were constructed for the 201411Table 1.3: LADWP’s energy mixEnergy source PercentageEligible renewables 23–Biomass and waste 6–Geothermal 1–Small hydro 1–Solar 1–Wind 14Coal 42Large hydro 4Nuclear 10Naturgal gas 17Unspecified 4Total 100IRP; four of them reflecting LADWPs commitment to increasing their RPSgeneration, and one base case. Each of the RPS scenarios is built arounda different GHG reduction strategy. The utility acknowledges the flexibil-ity needed to integrate solar and wind by including pumped storage hydroand natural gas in all portfolios. The California Public Utility Commission’s(CPUC) required loading order results in DSM and EE being considered assupply-side resources rather than modifications of the load, but they do notcompete directly against other resources because of the manual portfolio se-lection. Sensitivity analysis is carried out using deterministic high and lowforecasts for coal, natural gas, and CO2 prices, using the Planning and Riskmodel. For coal and natural gas prices, the high and low forecasts are 10 per-cent above and 5 percent below the base case, respectively. Once portfolioshave been simulated, the results are compared against each other using a de-tailed scorecard. LADWPs assessment criteria are reliability, environmentalstewardship, and economic (cost) considerations. Once a preferred portfoliois identified, it may be modified further to reflect these criteria or to comply12Figure 1.4: NWPCC’s IRP processwith new policy directives. The IRP also includes a short term action plan,outlining the procurement actions to be taken in the first four years of theplanning period.1.2.4 Northwest Power and ConservationCouncilThe NWPCC is a regional organisation created in 1980 with the passing ofthe Pacific Northwest Electric Power Planning and Conservation Act. Themain role of the organisation is to develop the 20-year power plan for the Pa-cific North West (Washington, Oregon, Idaho, and Montana), updated everyfive years[10]. The latest iteration is the sixth power plan, to be supplantedby the seventh in October 2015. The plan covers regional energy planningbut does not guarantee that local capacity needs will be met. The optimisa-tion model used by the NWPCC is MS Excel-based[11] and therefore is easilyaccessible to a majority of programmers. The model is called the RegionalPortfolio Model, or RPM. Another model, GENESYS, is used for assessingthe reliability of the plans produced, simulating loss of load probability andother reliability measures. HydSim, from the Bonneville Power Adminis-tration, is also used, to simulate hourly hydroelectric generation based onregional hydrological data.The NWPCC states in their sixth plan that the plan should be explo-rative rather than predictive, and that the chosen plan should be robust in13a wide range of potential conditions because of the uncertainty of foresight.To accommodate this, the plan includes decision criteria to evaluate risk inaddition to the optimisation. In terms of scenario selection, NWPCC allowsprices, load, and other variables to vary beyond historical levels, thereby in-cluding scenarios which are unlikely but possible. This is in recognition ofsome of the unprecedented changes that have occurred in the energy industryand the economy in the recent past.Three load forecasts are developed by the NWPCC using their regionalproduction-cost model, each forecast corresponding to certain economic drivers.Individual classes of customer are assigned separate growth rates based oneconomic and demographic trends, and the results are combined to form abase-case load forecast. The eventual base-case forecast was for 1.2 percentgrowth in demand per year over the duration of the planning period. In ad-dition, two other forecasts were developed to represent high and low demandconditions. The low demand forecast reflected slow recovery from the recenteconomic recession and therefore low power demand, with growth of 0.8 per-cent per year. The high forecast was used to demonstrate robust recovery,using a growth rate of 1.5 percent per year.The Excel-based RPM simulates each manually constructed resource port-folio against 750 scenarios, with quarterly time steps over a 20 year planninghorizon. The model records the net present value of the costs for the portfolioin a given scenario, and repeats this for all scenarios, building a distributionof costs for that portfolio. The introduction of increasing quantities of windpower has increased the need for load-following capacity, and this is reflectedin the choice of resources for the portfolios. Wind is also given a flexibilitypenalty between $6 dollars/MWh and $12/MWh to reflect the need for thisancillary capacity. The NWPCC uses TailVaR90 as their measure of portfoliorisk. This is the average of the highest ten percent of the net present valuecost outcomes associated with a given portfolio across all 750 scenarios. Apreferred portfolio would likely be among the portfolios with both low ex-14pected costs as well as low TailVaR90 values and hence a low risk of highcosts. This captures not just the probability of an undesired outcome, butalso the magnitude. With each portfolio then having an average net presentcost and TailVaR90 value, they can be plotted to determine the feasibilityspace and the efficient frontier.The model times resource additions not by in-service date but by earliestconstruction date. This is also a risk management strategy, as inaccurateforecasts can lead to both over- and under-construction. The earliest con-struction date is the point beyond which it is not possible to change the choiceof resource. Decisions on resource addition are made at each time step basedsolely on trends and information available up to that point; the model has noforeknowledge and may have to correct at a later time-step a decision thatwas taken earlier. This gives a realistic view of how decisions are made andaltered over time, allowing eventual identification of a portfolio that is lesssensitive to wide-ranging uncertainty rather than simply the least-cost for agiven scenario. Sensitivity analysis is carried out in the GENESYS modelby varying seasonal prices, hydro conditions, and load. The model gener-ates random profiles for these variables on an hourly basis. To preserve thecorrelations between the variables, the random profiles are generated with acorrelation of 0.95 between them. Transmission constraints are also studiedbecause Idaho is summer-peaking while Oregon and Washington are winter-peaking areas. Inadequate transmission capacity could therefore hamper theability of utilities in neighbouring regions to trade energy.The NWPCC also comments on the transparency of the modelling pro-cess, stating that stakeholders and the public should feel that a model is nota black box and should be able to see how their concerns translate into themodel. This is addressed in the first part of NWPCC’s IRP process (see theyellow tabs in Figure 1.4), where preliminary analysis and education includesdiscussion of how the model works and uses input data. Initial portfolio con-struction also involves stakeholders and may be carried out immediately after15preliminary analysis and education to ensure that understanding of the for-mer carries over to the latter. Later stages also include further stakeholderinput on the initial results and seek data for sensitivity analysis. One optionthat was identified for increasing confidence in the model is to run the sameinputs in a different model, such as AuroraXMP . Wildly dissimilar resultswould point to underlying differences in assumptions and model structure,and could be used to highlight whether the assumptions and structures arevalid.1.2.5 PacifiCorpPacifiCorp operates 10,400 MW of capacity and serves 1.8 million customersin Oregon, Washington, California, Utah, Idaho, and Wyoming [12]. Fiveof these states have IRP requirements, and while Wyoming does not, it re-quires utilities that file IRPs outside the state to also file these with its stateutility commission [3]. PacifiCorp therefore faces the challenge of imple-menting system-wide planning that fulfils the requirements of six differentjurisdictions. The IRP is updated every two years, with a planning horizonof 20 years (see Figure 1.5 for overview of process). As of 2013, PacifiCorp’sgeneration was dominated by coal, followed by natural gas (Table 1.4).PacifiCorp uses econometric models to develop load forecasts based onhistorical usage, weather, economic growth and customer behaviour/changes.Different forecasts are generated for different user groups, with groups thathave similar usage patterns combined together. These forecasts are thencombined to give the overall system forecast. PacifiCorp assumes an averageannual energy growth rate of 0.85 percent for its single base-case forecast.PacifiCorp considers DSM as a supply-side resource in the portfolio-building process, rather than as a load modifier. This allows DSM to competedirectly with other resources and can lead to a lower overall portfolio cost.Considering DSM as a resource can have beneficial effects on uncertainty, asnew generation can be delayed until larger uncertainties are resolved. The16Table 1.4: PacifiCorp’s energy mixEnergy source PercentageCoal 62Natural gas 17.4Hydro 6.3Wind 8.1Biomass 0.5Geothermal 0.4Solar 0.03Unspecified 5.3Total 100utility focuses on resource acquisition in the first ten years of the IRP. Toassist this process, PacifiCorp also develops an action plan for the first twoto four years of this period, focusing on potential regulatory changes andeconomic triggers that could dramatically change the resource portfolio.PacifiCorp’s main approach to uncertainty is to analyse a large number ofscenarios. Scenario development occurred in consultation with stakeholders.19 input scenarios (core cases) were chosen, each with varying assumptionsabout five key variables: prices and timing of CO2 regulations; natural gasand electricity prices; assumptions about RPS policies and federal tax incen-tives; policy assumptions around coal-fired plants and retrofitting to meetregional haze regulations; and kick-in times and ramp rates for DSM andefficiency resources. In addition to the 19 scenarios, PacifiCorp also hadfive different scenarios around the construction of new transmission capac-ity; when applied to each of the 19 core cases, this created 94 scenarios.Sensitivity scenarios were also used to look at alternative load forecasts andat resource-specific assumptions. Twelve of these sensitivity scenarios wereconsidered, each typically paired to a core case. In total 106 scenarios weredeveloped and modelled. The company also attempted to address the un-certainty around an economic level of renewable resources. PacifiCorp ran17two rounds of portfolio optimisation using System Optimizer. The first setincluded only scenarios in which no RPS requirements were included. Thisallowed all resources to compete for allocation in the portfolio based on low-est cost, including renewables. From these results, the company obtained anidea of the level of renewable resources that was economic for the simulatedscenario. The second set of scenarios had equivalents of the first set butwith RPS requirements of varying levels. The renewable resources chosenfrom the first run for each scenario were forced into the portfolio for thissecond run of the scenario as a minimum level of renewables. Any gap be-tween this level of renewables and the RPS requirement was filled with newrenewables chosen through RPS Scenario Maker, an optimisation model forrenewable resources. Once these resources were input into System Optimizeras fixed requirements, System Optimizer was run again to compile the restof the portfolio, which may or may not add renewable resources in additionto those forced into the portfolio. Sensitivity analysis was then carried outfor all portfolios developed through System Optimizer. The Planning andRisk module of System Optimizer was used for this. The uncertainty herewas largely handled by having a range of projections for the most importantvariables affecting portfolio cost. Three different carbon prices per short ton(zero CO2 price, medium with $16 in 2022 rising to $26 in 2032, and highwith $14 in 2022 rising to $75 in 2032) were input as projections into PaR.The model also used Monte Carlo simulation to get varied projections of load,gas prices, electricity prices, hydro energy availability, and thermal unit avail-ability. For each portfolio, 100 simulations were run with these variations inunderlying variables, giving 100 different cost estimates for each portfolio.The top-performing portfolios from the PaR model were chosen based on thefrequency with which they were below the mean values and in the upper tail(i.e. having one of the five highest values after the Monte Carlo simulation)of the simulated variables. The primary criteria used for judgement were:risk-adjusted cost, CO2 emissions, and supply reliability. One portfolio was18Table 1.5: Public Service Company of Colorado’s energy mixEnergy source PercentageCoal 53Natural gas 25Wind 19Hydro 2Solar 1Total 100then selected as the preliminary preferred portfolio, based on the companysassessment of the portfolio’s risk-adjusted PVRR, carbon dioxide emissions,and supply reliability (specifically measured as average annual Energy NotServed).1.2.6 Public Service Company of ColoradoThe Public Service Company of Colorado (Public Service) operates as part ofXcel Energy Inc. in Colorado [13]. The company serves 1.4 million customersin Colorado with a system capacity about 7,600 MW. This capacity countsfor about two-thirds of the load, and Public Service relies on power purchasesto augment their generation. Public Service relies heavily on coal and naturalgas with significant levels of wind energy as well (Table 1.5).Electric resource planning in Colorado occurs in two phases. During phase1, the utility compiles information on their existing generation fleet, assessesneed for additional resources, and plans how to acquire those resources. Thisphase leads to the development of the “Electric Resource Plan” or IRP, and aResource Acquisition Plan (RAP). The IRP is filed with the Colorado PublicUtilities Commission and updated every four years. Once the IRP is approvedby the Commission, phase 2 begins, in which the company implements theIRP and the RAP. The overall process is illustrated in Figure 1.6.Public Service is facing significant changes in its operating environment,19Figure 1.5: PacifiCorp’s IRP processOregon, Washington, California,Utah, Idaho, andWyoming Public Utilities CommissionsAssess needsBuild resource portfolioswithout renewablesBuild portfolios with renewablesSensitivityanalysisDecision toolsMetricsPoliciesRegulationsPreferred portfolioExisting capacityLoad growthGas price forecastsRenewable resource “floor”Scenario developmentResource optionsDSM as supply-sideresourceMonte Carlo simulationof PVRR and carbon costs20Figure 1.6: Public Service Company of Colorado’s IRP processColorado PublicUtilities CommissionIRP requirementAssess needsBuild resource portfoliosSensitivityanalysisDecision toolsMetricsPoliciesRegulationsPreferred portfolioShort term action planExisting capacityPlanned butnot-yet-completedgenerationLoad, gas/energy prices,DSM forecastsResource screeningIncreasing inclusionof renewablesBase-caseassumptionsAdditionalscenarios21as outlined in their 2011 IRP. In common with other utilities, the recent re-cession has affected demand for electricity and economic growth forecasts arestill uncertain. Public Services electric load growth forecast over the RAPperiod (2011-2018) is 292 MW, compared to the 2007 forecast of 1,000 MWover the same period. Uncertainties also surround the potential withdrawalof large customers such as the City of Boulder; if this occurs, the 2018 de-mand forecast could drop to 5 MW. To comply with the Clean Air-CleanJobs Act (CACJA) passed in 2010, the company is retiring 600 MW of coal,fuel switching from coal to natural gas on another 450 MW of generation,and installing emissions controls on another three coal-fired units over thenext six years. Public Service uses a planning period of 40 years, extendingfrom 2011 until 2050. DSM is included as a modification of the load fore-cast, not as a separate resource that competes against supply-side resources.The load modelled in their optimisation model, Strategist, consists of theprojected load plus the planning reserve margin, in this case 16.3 percent.Public Service prepares both a base-case (median) load forecast and highand low forecasts for sensitivity. The base-case load forecast is for growth of0.3 percent per year over the planning period, while the low forecast is fora reduction of 0.6 percent per year and the high forecast predicts growth of1.1 percent per year. The forecasts are based on economic projections fromIHS Global Insight, Inc. Monte Carlo simulation was used to develop thealternative high and low forecasts, which are the borders of a confidence en-velope of 70 percent about the median forecast. The IRP stated that basedon the load forecast, Public Service intends to delay most new generationconstruction and instead fill the gap with power purchases. This is a riskmanagement strategy whereby the company avoids high capital costs andwaits for uncertainty to resolve before committing to new generation. Therisk incurred instead is that power prices may increase higher than expectedover the short term. Uncertainty was primarily dealt with by running mul-tiple scenarios. The concern appears to be around renewable resources and22whether regulations will force a certain level of renewables to be used. Byinputting increasing amounts of forced renewable resources into the least-cost baseline case and re-optimising through Strategist, Public Service cansee the range of costs that could occur. Sensitivity analysis on the nine port-folios, undertaken with the Planning and Risk model, gives an estimation ofhow robust the portfolios are if their underlying assumptions change. Forthe sensitivity analysis, inputs such as CO2 price, tax credits, gas prices,and sales were varied to reflect a different future. Several features of thecompanys 2011 IRP are noteworthy. One of these is the inclusion of a clearcontingency plan; the events most likely to cause a capacity shortfall areidentified and clear actions to mitigate each event are listed. These rangefrom near-term events like a PPA falling through to more distant events likeslow construction of new generation. Public Service also takes an interestingapproach to integrating wind energy. The company intends to increase theirtotal wind generation, acquisition of which began under the previous RAPin 2004, to 2,100 MW by the end of 2012. This represents a sizeable portionof generation capacity that is intermittent, and the company plans for thisuncertainty by also selecting other resources that can be dispatched withina 30-minute time frame to manage this fluctuation. As utilities sometimesreject increased renewables for precisely this uncertainty, Public Services ap-proach offers a contrasting example of how renewables can be integrated.1.2.7 Tacoma PowerTacoma Power is a division of Tacoma Public Utilities, operating in theTacoma area of the state of Washington [14]. Tacoma Power is a publicenergy company, serving 169,000 customers and running a system that isdominated by hydro-power generation (Table 1.6). However, owned gener-ation provides only about fifty percent of Tacoma Power’s energy, as powerpurchase agreements with Bonneville Power Authority constitute over half ofits energy. The utility is required to file a full IRP with the Washington State23Table 1.6: Tacoma Power’s energy mixEnergy source PercentageHydro 90.6Nuclear 6.1Coal 1.2Natural gas 0.5Other(biomass, petroleum, waste, wind) 1.7Total 100Department of Commerce every four years, with updates required every twoyears. The IRP has a planning horizon of 15 years. Figure 1.7 presents anoverview of the most recent planning process.Uncertainty in rainfall and hydrology is the largest contributor to TacomaPowers overall uncertainty. Tacoma Power handles this high variability inrainfall by planning to the lowest historical stream flow (since the 1930 wa-ter year). This means that in average years the utility runs a surplus, whichit sells largely to the Bonneville Power Authority. Tacoma Power managesload uncertainty by dividing customers into different consumer categoriesand using different projection methods for each. For example, contract in-dustrial customer load is forecast by analysis of historical trends and directconsultation with the customer. Customers like lighting services, which growmore predictably, have loads forecast by extrapolation of historical trends.Residential customer load is forecast by regression analysis based on demo-graphics, weather data, and economic trends. Price forecasts are developedfor Tacoma Power by Wood Mackenzie using AuroraXMP , which TacomaPower then modifies with a risk adder for gas price uncertainty and for car-bon prices. Two price forecasts are considered in the planning, a high (75thpercentile) and low (10th percentile) forecast.In their 2013 IRP, Tacoma Power modelled their hydro operations withVISTA DDS (VISTA), a model from Hatch Ltd that optimises the opera-24Figure 1.7: Tacoma Power’s 2015 IRP processWashington StateDepartment of CommerceAssess needsBuild resource portfoliosSensitivityanalysisDecision toolsMetricsPoliciesRegulationsPreferred portfolioExisting capacityLoad growth forecastPrice forecastsResource screeningResource optionsBase-caseassumptionsAdditionalscenariosAdditional resourceoptionsSimulation25tion of their generating units. DSM was not considered to be a supply-sideresource in Tacoma Powers planning instead it was modelled as a modifierof the load. One of Tacoma Powers aims in the 2013 IRP was to delay theneed for new generation through DSM and increased efficiency. As hydro isthe largest component of Tacoma Powers system, existing and planned hydroresources were simulated in VISTA as base load generation. The uncertaintythat other utilities consider by multiple scenarios was considered by simulat-ing the hydro portfolio with historical water years. The model counted thefrequency of power deficits and surpluses with their chosen hydro portfolioover the course of the simulations and made an estimate of how often theyare in surplus given the historical water conditions. The planning scenariosused the critical water year, with runs taking the average flow for the year astests for how much surplus was expected under normal conditions. After thehydro portfolio had been run, other resources, such as combined cycle gasturbines (CCGTs(, wind, biomass, solar, and pumped storage, are then con-sidered as add-ons to the hydro base portfolio and added to the simulation.As Tacoma Power considered increased demand to be their greatest risk,several scenarios with higher than expected load were run to see the abilityof the portfolios to meet that demand. Sensitivity analysis was carried outusing Crystal Ball. As load and water year are independent of each other,the model varied these separately. Load variation is assumed to be ±15aMW with a triangular distribution. The model then selected a load withinthis distribution, chose a random water year and ran the simulation. Theprocess was iterative, with VISTA revising the shape of Tacoma Powers loadforecasts and inputs into Crystal Ball, and Crystal Ball revising the resourceportfolio. When the utility was satisfied that the process had produced areliable portfolio, it was selected as the preferred portfolio.Tacoma Power released their 2015 IRP in November 2015, using a dif-ferent process to the 2013 IRP[15]. The utility now uses Plexos to simulatethe operation of their portfolios over the planning period with four different26Table 1.7: TVA’s energy mixEnergy source PercentageCoal 34Natural gas 27Nuclear 18Hydro–Conventional 12–Pumped storage 4Renewables–Wind 4–Solar/biomass <1DSM 3Total 100scenarios. As Tacoma Power is able to meet expected load growth with DSMand EE, the IRP showed no need for additional generation resources. How-ever, in the interests of having analysis to fall back on if conditions changedrastically, the utility screened potential resource additions and selected themost suitable ones for their system. The effect of adding these resources tothe system was then tested by running simulations in Plexos, where each newresource was added to a particular portfolio as a block of 50 MW annually,and run against the four scenarios.1.2.8 Tennessee Valley AuthorityTVA is a federally-owned corporation covering most of Tennessee and partsof Alabama, Mississippi, Kentucky, Georgia, North Carolina, and Virginia[16]. TVA operates the largest public power system in the United Stateswith 36,520 MW of capacity, serving 9 million customers. As of the 2015IRP, coal and gas accounted for about 60 percent of TVAs generation (Table1.7).As a federal agency, TVA is obliged to file an IRP and environmental im-27pact statement (EIS) under the National Environmental Policy Act of 1970(see Figure 1.8 for outline of the full process). The 2015 iteration of the IRPis an update of the 2011 IRP, brought about by significant changes in theunderlying assumptions of the 2011 IRP. The IRP is produced in conjunc-tion with an environmental impact statement (EIS(, both of which must besubmitted to the Federal Energy Regulatory Commission (FERC) every fouryears. The goal of TVAs IRP process is to identify an integrated resourceplan that performs well under a variety of potential conditions, so robustnessis favoured more highly than absolute least-cost. TVA uses statistical andmathematical models to develop their load forecast. The utility looks at keydrivers of electricity sales - economic activity and growth, electricity prices,customer retention, and the price of competing energy sources - and esti-mates the load based on these variables. Historical records of power use arealso included. These are combined to develop a single load forecast in con-sultation with stakeholders and directly-served customers. TVA constructstheir overall forecasts from county-level forecasts.TVAs planning process is scenario-based, and uses System Optimizer.Five scenarios are chosen, representing potential futures over which the com-pany has no control, and based on what the company perceives the greatestuncertainties to be (e.g. gas price, carbon pricing, coal prices, economicgrowth, etc.). TVA then develops five resource planning strategies. Thesestrategies represent decisions that are within the companys control, such asasset additions or change of fuel type. In the 2015 draft, there are five ofthese: traditional base case least cost; emissions reduction; focus on long-term market supply; energy efficiency; and maximise renewable energy ca-pacity. Each strategy is then run against each scenario, modelled by SystemOptimizer, to generate a resource portfolio for each intersection. DSM isincluded in the model as a supply-side resource. Sensitivity analysis wascarried out for each resource portfolio using the MIDAS model. This modeluses a form of Monte Carlo simulation to create distributions of the under-28Figure 1.8: TVA’s IRP processFederal EnergyRegulatory Commisssion(FERC)Assess needsBuild resourceportfolios forintersection of strategyand scenarioSensitivityanalysisDecision toolsMetricsPoliciesRegulationsPreferred strategy andfive resource plansExisting capacityLoad forecastPrice forecastsSelection of viableresource optionsStakeholder discussionFive scenariosFive strategiesDSM as a supply-sideresourceMonte Carlo simulationof variables29lying variables. In this IRP, 72 Monte Carlo runs iterations were carried outfor each portfolio. The sensitivity scenarios centred on addition of nuclearcapacity, energy efficient and DSM effectiveness, pricing and performance ofrenewables, and high/low scenarios for power price, fuel price, carbon price,and load. In conjunction with their stakeholders and based on their strategicaims, TVA selected five broad metrics by which a portfolio would be assessed.These were cost, financial risk, stewardship, Valley economics, and flexibility.Within these categories, specific targets or measures were defined to evalu-ate each strategy for every scenario. These were combined into a scorecard,under which each strategy could be scored for the chosen measures over allfive scenarios. In total five scorecards would be made, one for each strategy.TVAs planning process is particularly transparent. The optimisation modelis used to compile resource portfolios, but it is actually the strategy thatis being evaluated. The eventual preferred strategy will have five resourceportfolios, corresponding to the five scenarios, all scored against the compa-nys long-term aims. This gives the company a broad picture of the mix ofresources that will mesh well with the chosen strategy. Planning uncertaintyis handled by having several scenarios, but also by not using probability toselect what futures are most likely. Instead, issues that concern the companyand stakeholders are developed into scenarios with less focus on whether it isa most likely future. Because these scenarios are not based on extrapolationusing historical data, they have the possibility of capturing behaviour thathas not happened before. Given current debate about stationarity in weatherpatterns and climate, this is a valuable consideration.1.2.9 Californian IOUs and Energy ContextCalifornia is different from most of the other North American jurisdictionsstudied because of its deregulated energy market. Instead of the type ofIRP process typically followed by vertically integrated utilities in regulatedmarkets, a deregulated market is dominated by competition for short term30power purchase contracts [17]. Separate companies handle generation, trans-mission, and distribution. The role of a utility is generally that of distribu-tion, not generation, although some utilities will have generation capacity toserve customers that do not opt for unbundled service. New generation isprocured by competitive generation companies in response to market con-ditions, so there is little incentive for long-term resource planning on thepart of utilities. The IRP process was replaced with a planning frameworkcalled an LTPP, which focused on procurement through power purchases.The CPUC began requiring utilities to file an LTPP in 2004, with updatesevery two years. Between 2004 and the present, however, the LTPP has splitinto two streams; one dealing with short- to mid-term procurement throughpurchases (bundled procurement plan), and one dealing with long-term sys-tem reliability and capacity expansion (system resource plan). The returnto longer-term planning is partly due to the increased volume of renewableresources imposed by the state. California is pursuing an ambitious policytowards reduction of GHG emissions and as part of this requires that utili-ties procure at least 33 percent of power from renewable resources by 2020[18]. As renewable resources frequently have more intensive capital costsand higher energy costs than traditional generation [17], investors have beenreluctant to commit to these resources without guarantee of cost recovery,and longer-term planning has helped to alleviate this issue. For both plans,the load forecast is developed by the CPUC in conjunction with the CECand California Independent System Operator (CAISO), and provided to theutilities along with a standardised set of planning assumptions and scenarios[19]. The CPUC updates this planning information every two years. CAISO,which has oversight of the transmission planning process (TPP), uses similarassumptions, and in 2014 the two organisations decided to coordinate anduse the same assumptions and scenarios for both processes[20]. The utilitiesconstruct portfolios that will meet the load forecast at least cost to the cus-tomer and that accord with the required state loading order (EE→ DSM→31renewables→ efficient fossil fuels) and renewables targets[18]. Hydroelectricresources with nameplate capacity greater than 40 MW per unit operatedare not eligible to be counted as renewable resources [8], so a distinction ismade between small hydro and large hydro in the loading order. Because ofthese constraints, the IOUs typically compile portfolios without the use ofmodels and therefore have fewer scenarios than utilities that use models forportfolio development [21][22]. The utility submits its LTPP to the CPUCfor approval and, if the plan is approved, begins to send out requests foroffers (RFOs) for construction of new generation.Pacific Gas and ElectricPG&E operates in the state of California, supplying electricity and gas to thecentral and northern areas of the state. With a customer base of about 5.4million electricity customers and owned capacity of 7,677 MW , the companyis one of the largest load serving entities in the WECC [24, 23]. Table 1.8provides the breakdown of PG&E’s energy sources as of 2014 [9]. Prior to2004, PG&E submitted a short-term procurement plan (STPP) every yearto the CPUC. From 2004 onwards, the company has filed an long-term pro-curement plan (LTPP) every two years.PG&E obtains their load forecasts from the CEC. The utility then canuse this as a base-case but may also modify their forecast and include otherforecasts if they can show good reason for doing so. PG&E uses a low fore-cast based on the CECs low forecast; a base forecast based on the CECs highforecast; and a high forecast based on the CECs high forecast with additional0.3 percent growth per year. As load in deregulated jurisdictions dependsheavily on changes in prices and the market, planning for procurement hasa relatively short planning period of generally six to 12 months. Planning inthis case occurs in consultation with CAISO. Longer term planning dealingwith system capacity, typically with a horizon of ten years, occurs through32Table 1.8: PG&E’s energy mixEnergy source PercentageEligible renewables 22–Biomass and waste 4–Geothermal 5–Small hydro 2–Solar 5–Wind 6Large hydro 10Nuclear 22Naturgal gas 28Unspecified 18Total 100the CPUC. Portfolios are constructed manually by PG&E to meet the loadidentified in their needs assessment, and run against the scenarios provided.The performance of each portfolio in each scenario is evaluated with the useof metrics that reflect the most important concerns of the utility. From thisevaluation, a preferred portfolio is identified and submitted to the CPUC forapproval[25]. Uncertainty is included largely through scenario modelling. Inthe latest iteration (2014), the scenarios were built by collaboration betweenthe CPUC, the CEC, and CAISO, and given to all utilities as a standard setof futures [19]. These three agencies also develop and maintain the Excel-based tool used by several Californian utilities to simulate their portfoliosand construct additional scenarios of their own. The Plexos platform isused for modelling in PG&Es LTPP process. Uncertainty is included byhaving a range of deterministic projections for a variable or Monte Carlosimulation to sample from a distribution of a variable [26]. PG&E explic-itly considers three types of uncertainty in the LTPP[21]. The first is shortterm cyclical uncertainty, such as weather, hydro conditions or forced out-ages. These are often partially covered by reserve margins, and are handled33by assigning probabilities and distributions to the variables. The second islong-term structural uncertainties, which are not covered by reserve margins,and include such variables as long-term load growth, potential movement ofcustomers to community choice aggregator schemes (governmental entitiesto serve local residential and business energy needs), and changes in regu-lations that govern resource adequacy. The third is long-term commercialuncertainties, also not covered by reserve margins, and includes risks likedelay in completion of new generation facilities, problems in obtaining per-mits, and delay in approval for new projects. Each general uncertainty isbroken down into specific events or concerns and built into a scenario. Riskis identified by running Monte Carlo simulations for each portfolio and gen-erating a distribution of portfolio cost. The to-expiration value-at-risk foreach portfolio is then used as a measure of the risk of a portfolio having highcosts[25]. Because of the separation of planning into two streams dealingwith procurement and system capacity, most of PG&Es risk management istypical of financial management (e.g. hedging in accordance with regulations,etc.). The eventually identified preferred portfolio of purchases is submittedto the CPUC and CAISO for approval. All purchases and transfers are thenconducted through CAISO.Southern California EdisonSCE serves 14 million electricity customers in central and southern areasof California. The company is the largest subsidiary of the public utilitycompany Edison International. SCE generates about 16 percent of the powerit supplies, with the remainder coming from market purchases [27]. Theutilitys owned generation relies on natural gas and a selection of renewableresources (Table 1.9). SCE still owns its transmission system although it wasforced to sell some of its generation assets as part of the deregulation of theCalifornia energy market in the late 1990s. To comply with the states RPSregulations, SCE sold its share of the Four Corners coal-fired plant in 2012.34In addition, the San Onofre Nuclear Generating Station, of which SCE helda 78 percent ownership stake, was closed in 2013.Table 1.9: SCE’s energy sourcesEnergy source PercentageEligible renewables 22–Biomass and waste 1–Geothermal 9–Small hydro 1–Solar 1–Wind 10Coal 6Large hydro 4Nuclear 6Naturgal gas 28Unspecified 34Total 100SCE also obtains their load forecast directly from the CPUC/CEC. SCEuses CAISOs Load and Resources Analysis (L&R) tool to determine its spe-cific service area needs[22]. The L&R tool identifies shortages in variousload regions by subtracting the expected load for the area from the avail-able generating capacity in that area. The user is able to choose from severalstandardised planning assumptions in the tool, for example choosing a higheror lower load forecast. The utility also uses an additional forecast developedthrough econometric modelling by HIS Global Insight for sensitivity pur-poses.Portfolios are compiled manually by the utility in discussion with stake-holders. These are then run against the standard scenarios defined by CPUC,CEC, and CAISO. The best performing portfolio, according to the companyspre-chosen metrics, is chosen as the preferred portfolio. The least-cost port-folio is not necessarily the preferred portfolio if other portfolios are of similar35cost but offer better outcomes on other valued criteria such as reliability orRPS compliance. Sensitivity is considered by running additional scenariosthat deal with shifts in market prices, and by using Monte Carlo simulationfor distributions for fuel and power prices. SCEs process in general resemblesthat of PG&E and other investor owned utilities (IOUs) in California.1.2.10 International utilitiesEskom - South AfricaEskom is one of the 20 largest energy producers in the world by generation ca-pacity, with owned generation of 41,194 MW. Eskom was founded in 1923 asthe Electric Utilities Commission and converted in 2002 to a public company.The company is wholly owned by the government of South Africa, generating95 percent of the electricity used in the country and 45 percent of the elec-tricity used in Africa [28]. The companys latest IRP (Integrated ResourcePlan for Electricity 2010-2030) was filed in March 2011 with the Departmentof Energy. The plan is expected to be updated at least every two years[1]. The company relies overwhelmingly on coal-fired generation (90 percentin 2011) with contributions from hydro (5 percent) and nuclear (5 percent).The hydro, however, is imported overland from Mozambique. Natural gas forpeaking generation accounts for less than 1 percent and contributions fromrenewables are insignificant. Eskoms IRP process begins within government,initiated by the Department of Energy (DoE). A first round of consultationtakes place with the public and other stakeholders to identify concerns andopportunities. Five scenarios are then developed by working groups in theDoE and Eskom, representing different policy directions. These scenarios areinput into Plexos, which uses optimisation to create least cost resource port-folios. The portfolios are analysed and modified by various working groupsto build a balanced portfolio that addressed governments risk concerns andobjectives. In the current IRP, these concerns were:361. Reduce carbon emissions2. New technology uncertainties (cost, lead time, operability, learningrates)3. Water usage4. Localisation and job creation5. Southern African regional development and integration6. Security of supplyThe portfolio chosen from this process is designated the Revised BalancedScenario (RBS). The DoE then commences the second round of consultationwith the public, industry, and other stakeholders. Results from this consul-tation process (in the 2011 IRP, issues such as changes for costing of nuclearplants, learning rates, and disaggregation of solar technologies were included)are taken and added to the scenarios in the second round of optimisation.Again, the resulting portfolios are assessed for fit with government policy,and then a final resource portfolio and plan is chosen, designated the Policy-Adjusted IRP. Uncertainty in this process is dealt with by having several sce-narios, informed by industry, government, and the public. The two rounds ofconsultation in this process allowed input into what original scenarios weredeveloped and into the assumptions used for the modelling in the first run.Eskom also recognises the risk of relying heavily on a single fuel source (coal)and to buffer this has chosen to move towards a more diversified portfolio.This IRP process relies less heavily on optimisation and more on decisionanalysis. The aim is not so much least-cost as reliability and stability. Riskin each portfolio would ideally be monetised and added to the cost of theportfolio for full analysis. However some risks are not easily monetised, sothe second best approach would be to assign probability distributions to eachrisk and use the standard deviation as a measure of the risk. This was also37not done, due to lack of time and discussion about the most appropriatedistribution for each risk. The third option, used in the IRP, was for simpleassignment of risk by expert opinion. Each aspect of risk for a particulartechnology as given a risk value, and the combined weighted risk values wereassigned to the technology, and then to the portfolio containing that technol-ogy. This allowed working groups to make decisions on which portfolios weremost robust. Contingency planning was also part of managing risk in thisIRP. For each technology and planned capacity expansion, decision trees wereused to outline the decisions that should be taken to maintain adequate sup-ply if particular events occurred. This method seems to incorporate RobustDecision Making (RDM) techniques, in which an action plan is developedbackwards, based on avoiding or mitigating events that could cause the planto fail.1.3 ConclusionThis report has presented the IRP process for a sample of electric utilitiesacross North America. An effort has been made to provide a broad pic-ture of IRP-using utilities, with diversity in location, generation, and marketstructure. Our research suggests that two main methods of planning areused among the surveyed utilities: a method based on manually constructedportfolios; and a method based on development of scenarios and use of anoptimisation model to construct and test portfolios. BC Hydro’s currentplanning process falls into the scenario-planning group and is largely typicalof that class of planning.38Chapter 2BC Hydro’s approach touncertainty in the 2013 IRPThis section was prepared as a report for BC Hydro’s Energy Planning group,recommending a modified approach to their uncertainty management. Someinformation about specific models and model operation has been removed.References for this report appear in the bibliography section of the thesis.2.1 IntroductionBC Hydro’s long term planning process involves the production of an IRPevery five years. The plan covers a period of 30 years from the year of planpublication and outlines the capacity expansion strategy for the utility. Aswith all forms of planning, BC Hydro must work with limited and change-able data to develop a strategy that will be robust in an uncertain future.This report outlines BC Hydro’s present modelling set-up and recommendsa revised framework to deal with uncertainties typically encountered in thedevelopment of long term capacity expansion plans.392.2 Summary of recommendationsA modified IRP process is laid out in the following steps:1. Determine what are the primary objectives and performance measuresfor the IRP process and portfolio analysis and assessment,2. Develop specific metrics to assess the performance measures,3. Use an optimisation model to develop optimal portfolios,4. Screen and select a set of portfolios to analyse more rigorously,5. Use HYSIM/GOM to simulate developed portfolios under a range ofconditions and observe their behaviour,6. Compare the portfolio performances with the use of clearly definedmetrics and trade-off analysis,7. Select a portfolio that performed well across all metrics.This process would require the implementation of several recommenda-tions to change the current process. These are:• Use HYSIM/GOM for further sensitivity analysis of individual portfo-lios, expanding to include simulating with alternative loads, gas prices,and energy prices as well as the current alternative water years. HYSIM/GOMis capable of this analysis, although minor changes will be needed tostreamline the process and automate for multiple input scenarios.• Develop metrics and scorecards and/or efficient frontier analysis forassessing portfolio performance;• Build a model or interface capable of changing the inputs to HYSIMautomatically for multiple runs, to streamline the process of runningthe increased number of simulations;402.3 Portfolio developmentThe utilities studied in a comparison of energy planning methods conductedfor BC Hydro [29] used two methods for developing portfolios for IRP. Theportfolios were either chosen manually, with the utility deciding on the inclu-sion of individual resources (portfolio-based planning), or were selected by amodel using some form of optimisation (scenario-planning).Manual portfolio development option has the advantage of directness be-cause the utility directly selects the resources needed to satisfy their load.The disadvantage is that for a large load-resource gap, a large number ofresource combinations can fill the gap and it can be difficult to select thebest combination manually. This method appears to work best when theoptions are constrained or the load-resource gap is small and can be easilyfilled by one or two resources. Manual portfolio development occurred amongthe Californian investor owned utilities (IOUs) because of their constraintsunder the state loading order, which prevents them from using particularclasses of resources until all affordable alternatives of a more favoured classof resources have been exhausted[17][19]. Tacoma Power also used manualportfolio development because they had no need of new resources and simplychose to test an addition to their system of 50 annual megawatts (aMW) ofenergy in different forms[15]. BC Hydro shares some similarities with theCalifornian IOUs because of constraints imposed by the British ColumbiaUtilities Commission (BCUC) and the British Columbia Clean Energy Actof 2010. These include a requirement for BC Hydro to achieve electricity self-sufficiency by 2016, meaning that BC Hydro must be able to meet electricitysupply obligations by 2016 and each year thereafter with energy generated inBritish Columbia; minimum targets on the percentage of energy that mustcome from renewable sources (93 per cent); target reductions in green housegases (GHG); and demand reduction of at least 66 per cent by 2020 [30].However, BC Hydro has many options to fill their load-resource gap, evenunder these constraints, and the gap between generation capacity and load41is such that a combination of one or two resources will not be sufficient tomake up the deficit.The other method used was scenario-based planning where the utility var-ied underlying variables to produce an array of scenarios and then used anoptimisation model to build portfolios of resources that were optimal for eachscenario. This method takes some of the ambiguity out of resource selection,as the optimisation model chooses resources based on defined objective func-tions and constraints, making the choice easier to justify than direct manualutility selection. This is important for a government-owned utility as it mustbe transparent in its planning processes. Models also tend to be more effi-cient at calculating the costs and benefits of including a particular resourcein a portfolio than a human, especially when there are many resource op-tions available and the load-resource gap requires combinations of many ofresources. The disadvantage of this method is the difficulty in formulating anall encompassing objective function that captures several non-commensurateobjective function terms. Size also appears to play a role, as the utilitiesreview [29] highlighted that larger utilities tend to use scenario-based plan-ning rather than portfolio-based planning, partly because the difficulties inselecting resources manually increase with utility size as more resource com-binations are required. Of the utilities reviewed, none with installed capacitygreater than 4,000 MW used portfolio-based planning. BC Hydro’s currentsituation, with a sizeable future load-resource potential gap and a variety ofoptions to meet this gap, suggests that a scenario-based planning methodmay be more practical than a portfolio-based planning method.2.3.1 Optimisation modelThe recommendation to use a scenario-based planning method requires theuse of a model for portfolio compilation. The model currently used by BCHydro, System Optimizer, has been successfully used for the IRP processby Duke Energy [31], PacifiCorp [12], and Tennessee Valley Authority [16],42among others. BC Hydro’s current implementation of portfolio development,where up to 4,000 scenarios are used, is more comprehensive than the ma-jority of scenario planning utilities reviewed, where the maximum number ofscenarios studied was 106 by PacifiCorp [12]. There is no urgent recommen-dation to change models, as System Optimizer is used by many utilities insimilar circumstances to BC Hydro, particularly Tennessee Valley Authority,which also has significant heritage assets and is government-owned. How-ever, another program used for the same purpose is Strategist, from the sameprovider as System Optimizer. Where System Optimizer uses mixed-integerprogramming for its optimisation, Strategist uses dynamic programming, andmay yield slightly different results. However, both System Optimizer andStrategist are deterministic models. System Optimizer solves a mixed inte-ger optimisation problem and it can potentially be formulated as a stochasticmixed integer problem to address some of the uncertainties in the planningproblem, but the problem becomes very difficult, if not even impossible, tosolve. Strategist is a deterministic dynamic programming model which canpotentially be extended to solve the stochastic optimisation problem.We recommend that BC Hydro investigates the potential use of dynamicprogramming to solve the optimisation problem, as it could potentially beextended to address some of the uncertainties involved in long term capacityexpansion problems. The inputs to System Optimizer for developing theportfolios are a price forecast, a load forecast, a resource inventory, and aninflow sequence. The development of these inputs is discussed further in thefollowing sections.Each System Optimizer “run” uses a single scenario made up of a selec-tion of components. The full complement of options used by BC Hydro canbe seen in Figure 2.1, where the highlighted boxes are the selections for aparticular scenario run. In each run, BC Hydro can select and build scenar-ios under three main categories: uncertainties considered; resource choices;and modelling assumptions and parameters. Under the uncertainties consid-43ered, BC Hydro varies market price, load forecast, DSM deliverability, andadditional load from LNG development in the north of the province. Underresource choices, they can vary usage of the 7% non-clean threshold, DSMoptions, and Site C (all units in) timing. Under modelling assumptions andparameters they can vary BCH/IPP cost of capital, use of pumped storageas a resource option, Site C capital cost, capital cost for alternatives to SiteC, and wind integration cost in dollars per megawatt hour.Figure 2.1: Scenario variables and their potential values in BC Hydro’s 2013IRPFor each scenario, the System Optimizer model selects resources to min-imise the overall cost of the resource in that set of conditions. This producesone portfolio of resources with a cost value. The uncertainty in this anal-ysis lies in how accurately the scenarios reflect reality over time. Ideally autility could run as many different scenarios as they required to capture allaspects of uncertainty, but in practical terms the maximum is usually in the44thousands because of time and computing constraints. A portfolio outputfrom System Optimizer contains: list of resources; in-service date; resourcetype; resource location; installed and dependable capacity; firm and total en-ergy; net present value of generation and transmission resource costs, traderevenue, DSM option, and total portfolio cost; transmission expansion; andsimulated generation and load.Load forecastLoad forecast is obtained by aggregation of residential, commercial, and in-dustrial loads. Residential and commercial loads are obtained from Sta-tistically Adjusted End-Use (SAE) models using both economic variables(disposable income, population, retail sales, employment) and non-economicvariables (weather, average stock efficiency of various end uses of electricity).Industrial loads are either developed for specific sub-sectors — for examplepulp mills or mining — in consultation with the major customers in theseindustries, or developed from GDP growth projections.Load forecasts contain uncertainties related to the variables used in theirderivation. For example, if economic growth stagnates, energy demand tendsto decrease, while a boom in an energy-intensive industry like liquid naturalgas could significantly increase energy demand. Because of this, utilitiesoften run simulations with multiple load forecasts, developing these throughhaving different assumptions about economic and population growth, fuelavailability, energy prices, etc. BC Hydro plans to the average load forecast,as per BCUC-approved policy [30].Energy price forecastThe energy price forecast is used to assess trade revenues and benefits andis developed using several factors:• Cost of new resources45• Gas price• Modelling of WECC loads and resources• Forward marketOf these four inputs, gas price and forward market are external forecasts,not developed by BC Hydro. The gas price is obtained from the New YorkMercantile Exchange (NYMEX)forecasts, while PowerEx provides the for-ward market prices and forecasts. Modelling of WECC loads and resourcesis carried out by the Price Forecast team using a production costing model[30]. These are models that capture the operational costs of a generationfleet and minimise costs while dispatching the system under various con-straints [32]. This model takes as inputs the plans of the WECC utilitiesfor new generation and each jurisdiction’s load forecast, and dispatches theplanned resources to minimise the cost of energy. The price forecast obtainedfrom this analysis is then used for developing projected market forecasts andscenarios. For example, different energy prices might be obtained from theproduction cost model by running low, medium, and high gas price forecasts.Resource inventoryThe resource inventory consists of all existing resources in the current systemplus any potential resources that the utility is considering including in aparticular portfolio. For each resource, the information included is capacity,energy, average price of energy in dollars per megawatt hour, and location ofresource. The resource inventory does not change between scenarios, unlikethe other inputs, which can be altered to produce different scenario-portfoliopairs.InflowsBC Hydro uses 60 years of historic inflow data to calculate an average systemoperation for planning purposes. This takes into account plant constraints,46non-power constraints such as environmental releases, and seasonal varia-tions. While using real data means that the average is a good representationof the previous 60 years of inflow conditions, it also implies an assumptionof stationarity that may not hold over future planning periods due to factorssuch as climate change. Firm energy is determined by dispatching the sys-tem in the lowest inflow years on record and calculating the energy producedunder such critical conditions.47Figure 2.2: BC Hydro’s current portfolio development processPrice ForecastReview alternativesfor generationProjectsResource optionsVariable Cost EvaluationFixed Cost EvaluationEvaluate Reliabilityand Non-Power ImportsLoad ForecastGeneratedportfolioHYSIM/GOM runsfor selected portfolios(see Figure 2.5)System Optimizer482.4 Alternatives for portfolio constructionDynamic programming (DP) is an alternative way to optimise portfolio se-lection. This modelling method works backwards from a given end state todetermine the optimal intermediate states and thus the overall optimal path,using Bellman’s principle of optimality, which states “An optimal policy hasthe property that whatever the initial state and initial decision are, the re-maining decisions must constitute an optimal policy with regard to the stateresulting from the first decision”[33].DP is usually called a multi-stage decision-making process. Instead ofdeciding on all decision variables in one single optimization procedure, theDP procedure dynamically divides the problem into many smaller decisionproblems (e.g., optimal portfolio choice), one for each possible discrete statein each stage in a planning process and the problem is iteratively and sequen-tially solved to find the optimal solution or in the case of capacity expansionproblems, the optimal investment strategy [33].A general outline of a planning method using dynamic programming in-stead of mixed integer programming is shown in Figure 2.3. Resource com-binations are constructed to cover all potential combinations of resources.For example, if a utility had five resource options that it wanted to opti-mise, the five portfolios constructed would be as shown in Table 2.1. Theexample is not intended to show all possible combinations, merely how adynamic programming model would move between states (i.e. portfolios) inits optimisation.Each of these portfolios would be initially simulated using HYSIM/GOMto obtain the costs and benefits of portfolio. Dynamic programming wouldthen be used to choose, at each time step in the planning period, whichstate – i.e. portfolio – would be optimal. Figure 2.4 provides an exampleof the set up for a dynamic programming problem, in which each state is aparticular portfolio, each stage is a time step between the present and theplanning horizon, and the nodes represent decisions to implement a particular49Table 2.1: The portfolio construction scheme for dynamic programmingPortfolio Resources1 A2 A, B3 B, C4 A, B, C, D5 A, C, D, E6 etc...portfolio at that stage.The advantages of using dynamic programming are that the benefits ofeach portfolio, not just the per unit cost of energy and of construction, areincorporated into the optimisation and resource selection without the needfor multiple feedback loops through HYSIM and GOM. The disadvantageis significant simulation time because of the HYSIM/GOM runs needed toobtain operational costs and benefits for each portfolio with its particularresource combination. In addition, BC Hydro would be required to obtaina different model for this type of optimisation, as System Optimizer is amixed-integer model and incapable of dynamic programming. Two dynamicprogramming models that have been used recently by utilities are Strate-gist (Public Service Company of Colorado [13], Arizona Public Service[5])and PowerSimm (NorthWestern Energy[34]). The technical details of thesemodels are not published and therefore their advantages and disadvantagescannot be fully assessed.50Figure 2.3: The IRP process using a dynamic programming model in placeof System OptimizerResourcecombinationsPortfoliosHYSIM/GOMsimulationCost and benefitsfrom simulationDynamic programming modelOptimal portfolio ofresources and timingFigure 2.4: An example of dynamic programming selecting between variousportfolios and optimising the overall resource selectionStage 1P1,1P2,1P3,1P4,1P5,1Stage 2P1,2P2,2P3,2P4,2P5,2Stage 3P1,3P2,3P3,3P4,3P5,3Stage 4P1,4P2,4P3,4P4,4P5,4Stage 5P1,5P2,5P3,5P4,5P5,5Portfolio 1Portfolio 2Portfolio 3Portfolio 4Portfolio 551It is recommended that investigation of a prototype model of the capacityexpansion problem for the BC Hydro system is considered. This could beformulated and solved using DP and Approximate DP in two phases. Usingthe BC Hydro System Optimizer inputs, Phase I would formulate and solve adeterministic DP capacity expansion problem for the BC Hydro system andthe results will be compared to the currently used System Optimizer results.Phase II would formulate and solve the stochastic optimization problem usingthe data assembled in Phase I on capacity expansion portfolios and theirstochastic state transitions given GOM run results for different scenarios ofhistoric inflow sequences, market price and load forecasts.2.5 Portfolio testingOnce a portfolio is built, a utility can use it “as is” or carry out furthertesting on the portfolio. It should be remembered that a portfolio developedthrough the aforementioned process is a portfolio optimised for a particularscenario. While this scenario may accurately reflect a particular future, itdoes not guarantee that the portfolio is optimal in a different future. All ofthe utilities studied in the previous review that used scenario-based planningconducted some further analysis on their most promising portfolios. Thisreport recommends an approach in which a portfolio is subjected to a rangeof scenarios to simulate its performance rather than ending the analysis af-ter the portfolio generation. This is often called sensitivity testing and wasfound to be part of the IRP processes of Arizona Public Service [5], AvistaCorp[35], Duke Energy Indiana[31], PacifiCorp[12], Public Service Companyof Colorado[13], and Tennessee Valley Authority[16]. This is somewhat akinto the analysis that occurs in portfolio-based planning once a portfolio hasbeen selected. Idaho Power provides an excellent example of this process,where, once their portfolios have been manually compiled, the utility usesAuroraXMP to simulate the operation of the portfolio with different values of52three underlying variables: natural gas price, load, and hydroelectric vari-ablity. Probability distributions of these variables (normal or log-normal)were derived and Monte Carlo simulation used to randomly sample fromthese values for 100 iterations, resulting in 100 different costs for a particularportfolio. These values give an indication of the spread of the portfolio costsand therefore of the vulnerability of the portfolio to changes in underlyingconditions (a proxy measure for risk).BC Hydro’s current process simulates a portfolio’s operation using theHYSIM and GOM models. Both HYSIM and GOM are deterministic mod-els. Once System Optimizer has produced a portfolio, the information aboutresources and cost is entered into HYSIM. HYSIM then simulates the op-eration of the portfolio under each of sixty years of inflow data, attemptingto avoid both shortages and spills and to maximise the value of BC Hydroresources. The end result of this analysis is a range of system operationsand costs and benefits for the portfolio, corresponding to the sixty wateryears. In effect, HYSIM expands the single inflow forecast given to SystemOptimizer into sixty different forecasts and assesses the performance of theportfolio in each. HYSIM is currently run with an Excel-based spreadsheetinterface. Inputs to HYSIM can be seen in Figure 2.5. The model runs ontime steps of a month, or twelve time steps per year. Inputs to HYSIM canbe seen in Figure 2.5.BC Hydro then uses GOM for more detailed simulation of portfolios.GOM is an optimisation model programmed in the AMPL language. Themodel is run from a GUI linked to the shared HYSIM-GOM database, collec-tively called the Study Manager. GOM is used for outage cost studies, plantconfiguration studies, to test the costs of certain constraints on the system,and for “what if” studies with flows, costs, etc. As GOM is an optimizer,it solves all time steps simultaneously with perfect foresight, which may beoverly optimistic. The combination of GOM using HYSIM outputs mitigatesthis tendency. GOM takes as inputs the outputs of HYSIM for monthly en-53ergy over all years in the planning period and the information from SystemOptimizer about the new resources. GOM optimises the timing and amountof importing and exporting energy, the dispatch of thermal resources, andwhere/when/how much water to store or use from BC Hydro’s reservoirs. AsHYSIM and GOM are used together, effectively working as a single step inthe IRP process, the arguments for using HYSIM rather than a commerciallyavailable model apply equally to GOM.The outputs from a GOM run are the feasible operational generation andreservoir pool schedule, and system and individual plant incremental costs,as well as the benefits that accrue to the system such as trade benefits frompower import/export, shaping benefits, energy shift benefits, and flexibiltybenefits. This gives a detailed picture of the actual operation of a portfoliogenerated by System Optimizer. This information can be used to refine aportfolio, and the process can loop back through System Optimizer, HYSIM,and GOM if needed.For these runs with HYSIM and GOM, corporate market price forecastsare inflated/deflated to account for dry/wet years’ inpacts on the Mid C mar-ket. This results in a range of costs for the different water years, giving anindication of the spread of the portfolio costs. However, inflows are not theonly variable that is uncertain, and variables like load forecast and energyprice can also be varied to assess the performance of the portfolio, based onthe inputs to System Optimizer. The methods used in creating these forecastsare detailed in §2.3.1–§2.3.1. To generate alternative forecasts for sensitivitypurposes, utilities may re-run their forecasting models with different assump-tions, such as higher economic growth leading to increased demand, or withnew sources of gas affecting market energy prices. Variation in inflows of-ten comes from historical records of such data, and a utility may choose torun simulations with all years of a water record or with the highest, lowest,and average flows. Utilities may focus on scenarios that are probabilisticallylikely, looking for instance at the mean and the 30th and 70th percentiles of a54Figure 2.5: The inputs and outputs of a HYSIM runSystemOptimizerPortfolio HYSIMGas PriceHeavy and Light Load HourImports/ExportsWater Years (Database)Resource typeand characteristicsCost of portfolio operationfor each of 60 yearsTrade benefitsShaping benefitsEnergy shift benefitsvariable, or may consider the worst-case scenarios if they are particularly riskaverse. BC Hydro’s HYSIM/GOM model is capable of running this sort ofsimulation, and already does this for the different water years. Other mod-els that could be used are AuroraXMP [6][35] or PowerSimm[34]. However,HYSIM is uniquely suited to BC Hydro’s operations because of its capacityto model the Columbia River Treaty operations and how it values water instorage, particularly with BC Hydro’s large reservoir facilities on the Peaceand Columbia rivers. HYSIM is also an in-house model, whereas AuroraXMPand PowerSimm are commercial models requiring significant modificationsand changes to model the BC Hydro system and will require the purchase oflicenses and thus have higher costs.2.6 Evaluation of model outputsIf portfolios are being simulated and tested to provide greater information toa utility’s decision makers, some method of presenting this information in an55Figure 2.6: The inputs and outputs of a GOM runHYSIM output GOMFeasible operationalgeneration scheduleReservoir poolscheduleSystem and individualplant costsBenefits-Trade-Shaping-Energy shift-Flexibility-Reserves56easily useful form is required. The presentation of this information shouldespecially facilitate and assist with comparison among portfolios. This re-quires the development, early in the planning process, of performance met-rics. These depend on a utility’s policies and objectives as well as theiroperating environment. Metrics for assessing portfolio performance can varygreatly depending on the objectives of the utility. The utilities studied in[29] used a variety of metrics to assess portfolio performance; observed were• Mean cost [5][10][12][16]• Standard deviation of costs [6] “Financial risk”[16]• Tail VaR(85, 90, 95) [10][12]• Fuel diversity [5]• Water use [5]• CO2 emissions [5][12]• Flexibility [16][13].Each metric addresses a particular aspect of uncertainty. A summary ofthe recommended metrics is presented in Table 2.3 and discussed further inthe following section.2.6.1 Recommended metricsThe recommended metrics cover a range of uncertainties and provide a utilitywith a broader picture of the performance of their portfolio. Not all metricsare calculated from the same data. Mean portfolio cost, standard deviation ofcosts, and tail Value-at-Risk would all be calculated from the results of mul-tiple HYSIM and GOM runs, where the portfolio is simulated with differentvalues of underlying variables and a distribution of outcomes is generated.57CO2 emissions would also be calculated from these simulation results, usingequations to estimate the carbon emissions based on the capacity of thermalresources in the portfolio. Avoided carbon emissions can also be calculatedbased on the equivalent thermal resources that renewables displace. Portfolioresource diversity would be calculated using the direct outputs of System Op-timizer which indicate the capacity and energy of each proposed resource andtheir percent contribution to the overall portfolio. Flexibility likewise wouldbe calculated from the System Optimizer outputs of portfolio make-up, usingdifferent data depending on the choice of flexibility metric. The process ofcalculating the metrics is illustrated graphically in Figure 2.9, showing theinputs to the models, the models used, the outputs of the model, and theformulas needed to calculate the metrics from the data.Expected value of costMean cost is a common metric used to show the average cost of implementinga particular portfolio. This is calculated asµ =∑n1 Portfolio costn, (2.1)where n is the number of portfolio HYSIM/GOM simulation runs.Standard deviation of costStandard deviation of costs shows how greatly the cost may vary and there-fore what are reasonable contingencies to put in place. A smaller standarddeviation of costs would indicate a portfolio that is stable across a wide va-riety of futures and therefore has a lower risk of exceeding cost thresholds.Standard deviation is calculated asσ =√∑(x− µ)2N, (2.2)58where x represents each value in the population (in this case, portfolio cost),µ is the mean value of the population, and N is the number of values in thepopulation.Tail value-at-risk of portfolioTail Value-at-Risk (TVaR) is calculated, like mean cost and standard de-viation of cost, from the distribution generated by multiple HYSIM/GOMsimulations of the portfolio. TVaR demonstrates the expected value of a lossgiven that an event outside a given probability level has occurred. Portfolioswith lower TVaR values are therefore less risky. TVaR for a given percentileis calculated asTV aRα (X) = E [−X|X ≤ −V aRα (X)] = E [−X|X ≤ xα] , (2.3)where X is the variable being considered (in this case, cost), xα is the upperα-quantile given by xα = inf{x ∈ R : Pr(X ≤ x) ≥ α}, | is the mathematicalexpression for “given”, E is the symbol for expected value or average, andVaRα(X) is the the value-at-risk for a particular variable at a particular valueof α.Portfolio resource diversityResource diversity or resource mix suggests how vulnerable a portfolio willbe to changes in fuel prices, as a system heavily dependent on one mainresource will be significantly more vulnerable to changes in the price of thatresource than a portfolio with a variety of generation options. A portfoliowith a greater diversity of fuel sources would therefore be considered to havelower risk than one that relies heavily on a single resource. The variance ofa portfolio, in the statistical sense of the term, could be a useful measure ofthe spread of the portfolio’s resource distribution, and is calculated by59σ2 =1nn∑i=1(xi − µ)2 (2.4)where µ is the expected value of the capacity of all resources in the portfolio(see Equation 2.1) and xi is the capacity of an individual resource. As thevariance indicates the average of the spread of the variables about the mean,a low value of variance indicates that the energy generation is spread rela-tively evenly among the various generation options. A high value of varianceindicates that a few of the resources are dominating the resource mix. Thefollowing example uses a portfolio from BC Hydro’s 2013 IRP to illustratethe calculations (see Figure 2.7 and Figure 2.8).The total firm capacity for each resource is used for this variance cal-culation. Any of the totals, either installed/firm capacity/energy could beused.The variance calculations would be repeated for multiple portfolios and theresults compared to determine the most diverse portfolio.Water consumptionWater use was only a metric for Arizona Public Service, which operates ina region of scarce water resources and therefore has an interest in choosinggeneration that does not rely on heavy water usage. This metric could becalculated by historical water usage of a similar sized plant and by interpo-lation. A high water use would be undesirable. This metric is not of greatvalue for BC Hydro given that their calculations already consider the value ofwater stored in their dams and optimise to use this as efficiently as possible.Carbon emissionsCO2 emissions was used as a metric by many utilities concerned about newregulation that would put a price on carbon emissions, affecting the dollar perMWh ratio of a high-CO2 emitting resource and therefore portfolio make-up.60Figure 2.7: A BC Hydro portfolioFigure 2.8: Total capacity for each resource in the IRP over the planningperiod61Emissions avoided could also be calculated, based on the carbon emissionsfrom a thermal resource of equivalent capacity. If utilities are concernedabout the price on carbon, a portfolio with a lower level of carbon emissionwill be of lower risk. The volume of carbon emissions would be an outputof the portfolio simulation process, and direct comparison between portfolioswould be possible. If there are a range of carbon emissions for each portfoliodue to the simulation process, then measures like the mean and the standarddeviation can also be use for this analysis.FlexibilityFlexibility was used as a metric by utilities that were interested in integratinghigher levels of renewable energy from intermittent resources such as wind.As an illustration, the Public Service Company of Colorado was consideringthe addition of 1,200 MW of wind energy to their system in their 2011 IRPand therefore was interested in adding resources that could manage withvariability, e.g. natural gas. Higher levels of flexible resources in a portfoliowould indicate lower risk of the utility having shortfalls in capacity. Severalmetrics for measuring flexibility have been proposed in literature [36]:• Percent of GW of installed capacity capable of load-following relativeto peak demand,• Systems where power area size, grid strength, interconnection (trans-mission), and number of power markets are given scores and combinedinto an overall flexibility score,• Maximum upward or downward change in load that the system is capa-ble of managing in a given time period from a given initial operationalstate, and• Expected percentage of incidents in a given time period where the sys-tem cannot cope with the changes in net load.62The complexity of the calculations increases going down the list, and asBC Hydro already has significant ability to buffer renewables due to theirlarge hydro resources, perhaps a relatively simple metric such as the first onelisted would be sufficient.An example of a flexibility calculation follows, using the portfolio in Fig-ure 2.7 and Figure 2.8. In this case, all resources that do not contributeto flexibility, such as wind and run-of-river which actually reduce flexibility,are assigned a negative value equivalent to 5% of the installed capacity. Forthe resources that do contribute to flexibility, pumped storage is assigneda contribution of twice its installed capacity because of its ability to bufferwith both capacity and pumping speed, and Site C is assumed to be able tocontribute 25% of its installed capacity to flexibility. The flexibility valuesfor each resource in this portfolio are therefore illustrated in Table 2.2. Thepeak energy demand expected in the 2013 IRP was 14,500 MW, thus usingthe first of the flexibility formulas listed above, the flexibility of this portfolioisFlexibility =Installed capacity (MW)×multiplierPeak demand (MW)=4, 889.414, 500= 0.33763Table 2.2: Capacity of each resource in a portfolio multiplied by the appro-priate multiplier for use in flexibility calculationsResource Installed Capacity (MW) Flexibility ContributionSite C 1100 275GMS units 1-5 Cap Increase 220 220MSW2 LM 25 -1.25Revelstoke Unit 6 500 500Pumped Storage LM 1000 2000Wind PC21 99 -4.95Wind PC28 153 -7.65Wind PC13 135 -6.75Wind PC16 99 -4.95Wind PC19 117 -5.85Wind PC10 297 -14.85MSW1 VI 12 -0.6Biomass VI 30 -1.5Run of River LM 80 100 62 -3.1Wind PC09 207 -10.35Wind PC15 108 -5.4Biomass PR 28 -1.4Biomass LM 30 -1.5Wind PC14 144 -7.2Wind PC20 159 -7.95Pumped Storage LM 1000 2000Wind PC11 126 -6.3Wind PC41 45 -2.25Wind PC42 63 -3.15Wind PC18 138 -6.9Wind VI14 35 -1.75Total 4889.464Table 2.3: Summary of recommended metrics for portfolio comparisonMetric Logic for inclusion MeasureMeanProvides the most likely cost, GHGemission level, water use, etc. of a port-folioArithmetic meanStatistical dispersionShows how stretched or squeezed thedistribution of a chosen portfolio char-acteristic is, and therefore is a measureof riskStandard deviation, variance, in-terquartile rangeExpected value of loss (or gain) ifan event outside a given level ofprobability occursShows average cost of the highest 10percent of cases, measure of financialriskTail Value at Risk or Value at RiskDiversity of portfolio resourcesIndicates reduced vulnerability of port-folio to swings in single fuel priceVariance, possibly calculated from re-source capacity in MWCarbon emissionsFor regulatory purposes due to theprovince’s green/clean/renewable en-ergy mandatesTons of CO2 per MWhFlexibilityGives an indication of how easily thesystem can deal with fluctuations fromintermittent resourcesPercent of total capacity able to bufferintermittent load relative to peak de-mand65Figure 2.9: Inputs and models required to calculate the recommended metricsInflow recordsEnergy pricesLoad forecastsHYSIM/GOMPortfolioCO2 emissionsResourcemixFuel diversityFlexibilityScenario 1...Scenario nPortfolio cost 1...Portfolio cost nExpected cost of portfolioStandard deviation of portfolio costsTail Value-at-Risk of the portfolio662.6.2 ScorecardsMetrics can be combined into scorecards for comparing portfolios. For ex-ample, each portfolio could be ranked from n to 1, where n is the number ofportfolios being tested, for each separate metric. This would result in eachportfolio having a score reflecting their relative merit among the portfoliosas a whole.Consider the case of a utility comparing three portfolios: A, B, and C.Each of these will be compared with four different metrics: expected cost,standard deviation of costs, fuel diversity, and tons of CO2 emitted. Basedon the values calculated for each of the metrics in Table 2.4, each portfoliocan be given a ranking out of three for each metric (Table 2.5). This analysiswould show that portfolio A performed most strongly when considered acrossall the metrics.67Table 2.4: Example of a scorecard for comparing portfolios using metrics andcalculationsPortfolio Expected cost Standard deviation Fuel diversity CO2 emissionsUnit Millions Millions Statistical variance Total tons CO2A 30 5 189 100B 25 4 589 500C 40 3 322 400Table 2.5: Example of a scorecard for comparing portfolios using metrics,with the metric rankings instead of the calculated numbersPortfolio Mean cost Standard deviation Fuel diversity CO2 emissions TotalA 2 3 1 1 7B 1 2 3 3 9C 3 1 2 2 8Table 2.6: Example of a score card using weights to reflect utility prioritiesamong metricsWeights 40% 30% 10% 20%Portfolio Mean cost Standard deviation Fuel diversity CO2 emissions TotalA 0.8 0.9 0.1 0.2 2.0B 0.4 0.6 0.3 0.6 1.9C 1.2 0.3 0.2 0.4 2.168If a utility was very concerned about a particular metric, they could applya multiplier to ensure that metric carries greater weight in the analysis. Forexample, the utility could rank their metrics in order of importance – 1) meancost, 2) CO4 emissions, 3) standard deviation of costs, and 4) fuel diversity– and give each of these a weight such that the total adds up to 100 percent(see Table 2.6). This can result in significantly different relative rankings forportfolios, as in this case portfolio B is the best performing.This report recommends the introduction of a system of metrics andscorecards for BC Hydro’s uncertainty management and portfolio testing.Of the various metrics identified in the jurisdictional review, several could berelevant to BC Hydro and are detailed in Table 2.3. These provide a broadpicture of a portfolio’s characteristics. The disadvantage of this approachis that more work is necessary to carry out the analysis and interpret thedata. The advantages are easy comparison of portfolios and a transparentprocess for ranking, assisting with both decision-making and justification ofdecisions to the public and regulatory bodies. The alternative is to have asingle criterion for choosing a portfolio, such as mean cost. However, use of asingle criterion does not take into consideration the risk of a portfolio beingmore costly, its environmental impacts, or any other factors that may affectthe success of the portfolio’s implementation.2.6.3 Trade-off analysisMetrics can also be useful for direct trade-offs between two performancecharacteristics of one portfolio at a time. If for instance a utility is inter-ested comparing portfolio mean cost to level of risk, they can plot mean costagainst a measure of risk (say standard deviation of costs or TVaR) and findthe lowest-cost portfolio for a given level of risk i.e. the “efficient frontier”,trading off between the two metrics. In addition, this sort of plot would al-low a utility to observe incremental cost changes to obtain a lower or higherlevel of risk. This is the approach advocated by the Northwest Power and69Conservation Council in their sixth regional power plan [10]. This reportrecommends the use of this method for BC Hydro in cases where two met-rics appear to be inversely related, such as fuel diversity and mean cost. Acomparison between mean cost and TVaR or standard deviation can also beconsidered. While the disadvantage of this approach is increased manipula-tion of the data, the advantages for decision-making are significant: directcomparison between contradictory metrics and ease of identification of the“best” portfolios under consideration.2.7 Summary and conclusion• Continue to use the current capacity expansion model (System Opti-mizer) for developing portfolios from multiple scenarios• Investigate the potential use of dynamic programming for portfolio se-lection and compare its output with that of System Optimizer• Use HYSIM/GOM for further sensitivity analysis of individual port-folios, expanding beyond the 60 water years and expanding to includealternative loads, gas prices, and energy prices• Develop metrics and scorecards and/or efficient frontier analysis forassessing portfolio performance• Develop a GUI for running HYSIM to streamline the process of runningthe increased number of simulations. This can be done by enhancingthe existing GOM GUI.70Figure 2.10: Updated portfolio development and assessment processPrice ForecastReview alternativesfor generationResource optionsVariable Cost EvaluationFixed Cost EvaluationEvaluate Reliabilityand Non-Power ImportsOther metricsLoad ForecastPortfoliosPresecreening tominimise portfoliosHYSIMGOMPerformancecharacteristicsMetrics, scorecards,trade-off analysisRecommendedportfolioImplement portfolioSystem Optimizer71Chapter 3Practical methods ofconsidering uncertainty inintegrated resource planningfor hydropower systemsThis chapter was written as a manuscript for publication in a journal relatedto electric planning or energy policy. References in cited in this chapter canbe found in the Bibliography section of the thesis. Some sections within thedocument have been moved for clarity. Section 3.2.7 is a modified version ofsection 2.6.1 from Chapter 2.3.1 IntroductionElectric energy utilities face a variety uncertainties when engaging in longterm capacity expansion planning. These include uncertainties such as futuredemand, prices for fuels such as gas and coal, new regulations restricting orprohibiting fossil fuel generation, regulations governing carbon pricing andgreen house gas emissions, level of subscription to demand reduction scheme72and effectiveness of energy efficiency programs, flexibility required to inte-grate higher levels of intermittent renewable resources, resource availability(gas, water, coal, etc.) and so on. Some of these uncertainties have becomemore prominent in the last few decades, particularly those around renewableresource supply and pricing. Utilities desiring to reduce their susceptibilityto risk therefore have to incorporate in their planning some method of assess-ing the impacts and mitigating the effects of these uncertainties. This paperinvestigates the current practice of electric energy planning among utilitiesin the United States and Canada. This work focused on achieving four ob-jectives: (1) examine the various planning methods used by utilities in NorthAmerica, (2) investigate how the planning processes handled uncertainty, (3)assess how circumstances prompted use of a particular planning method, and(4) develop and recommend a conceptual framework for treating uncertaintyin IRP processes for large scale hydroelectric systems.3.1.1 Structure of paperThis paper is organised in the following manner. Section 3.2 presents theuncertainties facing utilities, the utility approaches to planning under un-certainty, commonalities and differences in the planning methods identified,the modelling methods and programs used, and the introduction of metricsand scorecards as assessment criteria for rating portfolio performance underuncertainty. Section 3.3 describes BC Hydro’s system and operating environ-ment and its current planning method, which is contrasted with the findingsfrom Section 3.2. The development of a new framework for BC Hydro isoutlined and recommendations for progression from the current regime tothe new framework are presented. The conclusions and policy implicationsof preceding sections are then discussed in Section 3.4.733.2 Literature reviewIRP, defined as electricity planning that considers both supply-side anddemand-side resources for inclusion in capacity expansion resource portfo-lios [37, 4], is practised by relatively few jurisdictions. Much of the UnitedStates either practices IRP or is returning to IRP processes after failure ofcompetitive generation in deregulated markets [23, 3, 38]. Canada has avariety of planning systems, ranging from the deregulated Alberta ElectricSystem Operator that carries out load forecasts and then sends out requestsfor offers, to the more vertically integrated British Columbia system whereBC Hydro submits an IRP every five years to the British Columbia UtilitiesCommission [39, 30]. South Africa’s national energy company, Eskom, insti-tuted IRP planning in 2010 with the publication of their Integrated ResourcePlan for Electricity 2010-2030 [1]. The state of Queensland in Australia canarguably be said to practise IRP because of its emphasis on load reduction –through demand-reduction and energy efficiency – to avoid extensive capitalworks necessary to supply a widely dispersed population [2]. Brazil’s largeelectric system is beginning to consider energy efficiency (EE) but does notdirectly pit EE or demand-side management (DSM) as competitors againstnew generation and therefore cannot be said to carry out IRP [40]. Thisstudy therefore focused on the United States, where examples of IRP areplentiful, and on Canada.The information necessary for this study was obtained from each utility’spublished integrated resource plan. These were available from the utility di-rectly or from a state or province’s public utilities commission. The utilitieswere chosen to provide a good coverage of resource mixes and regulatoryprocesses in North America. Both east and west coast utilities were selectedto allow for differences in climate and fuel mix used. The east coast, for ex-ample, had a higher reliance on nuclear energy [31, 16]and less capacity forsolar than the west coast [5, 8, 12]. Differences in market structure and own-ership (public or private) also influenced utility selection, such as PG&E as74a private investor-owned utility providing contrast to the municipally ownedLADWP. Each utility’s plan was examined to identify the planning methodused and the strategies used for managing uncertainties. Planning methodsand uncertainty strategies that recurred frequently among the utilities wereanalysed to discover the combination of factors that lead to the choice ofplanning method used.3.2.1 Common uncertainties faced by utilitiesFrom analysis of the utilities’ published IRPs, some common uncertaintieswere discovered across the study. The main uncertainties were load growth,energy prices, and gas prices (as a significant driver of energy prices dueto gas-fired generation acting as rapidly available generation in the event ofa shortfall in capacity). Other common uncertainties were regulations andlevel of carbon pricing, effectiveness of demand-side management, and costsfor renewable resources. Table 3.1 provides a breakdown of the uncertaintiesexplicitly considered by each utility in the study. Some of these variableshave historical data that allow a utility to make educated assumptions abouttheir variabilities and potential values. Others like carbon pricing and effec-tiveness of demand-side management are more difficult to predict because oflack of historical precedents. However, historical precedents do not guaran-tee continuation of these trends in the future, as the changes in gas priceswith the development of shale gas in North America has illustrated. Recentresearch considers methods of quantifying the economic value of DSM toreduce this uncertainty [41].The common approach to uncertainty management among the utilities in-volved building scenarios with differing values of the uncertainties considered.This translates into utilities with high concern about gas prices developingscenarios with five different gas prices. Utilities concerned about changes inload would likewise produce scenarios with several different values of loadand simulate the performance of their system under each of these scenarios.75Some utilities chose to use aggressive values of a variable to produce worst-case scenarios and plan and prepare accordingly. Others decided to eliminatetheir exposure to a particular risk completely, for example by divesting fromall coal-fired generation in an attempt to remove uncertainty around carbonprices and emission regulations.3.2.2 Utility approaches to planning under uncertaintyUtilities appeared to use one of two main methods for managing their plan-ning under uncertainty. In one method, which we will call portfolio-basedplanning or portfolio planning, the utility builds a portfolio which they thentest under uncertainty. In the second method, which we will call scenario-based planning or scenario planning, the utility builds scenarios that covera range of uncertain futures and then build portfolios expected to performwell in these futures. Table 3.2Utilities also considered uncertainty in the length of their planning hori-zon and the frequency of plan updating. A plan to be executed over a shortperiod would be more flexible and therefore more sensitive to uncertainty, butthere is a trade off with stability and the length of time required for com-missioning new resources. Most companies opted for a 20-year horizon, withexceptions among the Californian investor owned utilities (“IOUs”) whichchose 10 years. The frequency of plan updating also differed, ranging fromtwo years to five years, with two years being most common. The utilitieswith updating periods longer than two years were BC Hydro, Tennessee Val-ley Authority, and the Northwest Power and Conservation Council [30, 16,10]. Both BC Hydro and the Tennessee Valley Authority are governmentowned utilities (by the province of British Columbia and by the U.S. federalgovernment, respectively) with significant owned generation assets, whichperhaps provides the stability for updating less regularly. The NorthwestPower and Conservation Council is an advisory body which considers thePacific Northwest region as a whole and does not own any generation, thus it76Table 3.1: Uncertainties considered by the utilitiesUtility LoadFuelprices(gas,coal)EnergypricesRetirements(coal,nuclear,hydro)GHGpricesandpenaltiesRenewableenergyregulationHydroelectricavailabilityPVorCHPuptakeDSM/EEuptakeInterestrateTaxincentivesElectricvehiclesImportsTransmissionPlantconstructioncostsStorageadditionsAPS • • • • • • • • •Avista • • • • • •BC Hydro • • • • • • •DEI • • • • • •Idaho Power • • • • • •LADWP • • • • •NWPCC • • • • • • • •NWEPG&E • • • • • •Pacificorp • • • • • •PSCC • • • • •SDG&E • • • • • •SCE • • • • • •TPU • • • • • • • • •TVA • • • • • •77Table 3.2: Choice of planning process for studied utilitiesUtility Method Planning horizon Update period Plan vintagePortfolio Scenario (years) (years)APS × 15 2 2014Avista × 20 20 2013BC Hydro × 20 5 2013DEI × 20 2 2013Idado Power × 20 2 2015LADWP × 20 2 2014NWPCC × 20 5 2010NWE × 20 2 2013PG&E × 10 2 2011PacifiCorp × 20 2 2013PSCC × 40 4 2011SDG&E × 10 2 2014SCE × 10 2 2011TPU × 15 2 2015TVA × 20 4 2015has less urgency to update a plan for implementation than most operationalutilities[10]. In addition, more frequent plans might not capture longer-termtrends over the Pacific Northwest. The update periods chosen by utilitiesseem to be a balance between the need for stability in a long-term plan andthe need for flexibility to meet uncertain futures.3.2.3 Initial steps for both planning methodsA necessary initial step, even before determining the need for new resources,is the choice of performance metrics against which the portfolio performanceswill be measured. This assists in guiding the overall process, as it clarifiesthe utility’s objectives. The assessment criteria can range from somethingas simple as picking the lowest-cost portfolio to more complex criteria bal-ancing cost, risk, the regulatory environment, company direction, etc. Tosome extent, the choice of criteria reflects a utility’s attitude to uncertainty,with more conservative utilities, typified by government-owned or crown cor-78porations [16, 30], choosing to consider a wider range of criteria to mitigatea broad spectrum of risks, while private utilities may weight their analysismore heavily towards risks in costs.To determine need for new resources, the utility must have a forecastof future load. Uncertainty in this aspect is high, as load growth is influ-enced by variables such as economic growth and population trends [17, 35],weather [35, 31], etc., all of which introduce uncertainty into the forecast.The utilities studied largely choose to use mean values of their underlyingfactors and produce a base-case most-likely forecast, mitigating uncertaintyby also developing forecasts for higher and lower load levels [5, 31, 12] toshow the potential variation in their load. All the utilities except TacomaPower applied reserve margins to this forecast, ranging from an additional10% up to 15%.Once a estimate of future need has been determined, utilities may screenpotential resource options for inclusion in portfolios. Uncertainty manage-ment in this step consists of avoiding resources with significant risks or requir-ing the inclusion of particular resources for risk management reasons. Thismay mean rejecting new/retiring old coal generation over concern about newregulatory controls on emissions, or requiring new gas generation for balanc-ing the inclusion of wind due to renewable energy regulations.3.2.4 ModellingBoth portfolio-based planning and scenario-based planning require the useof models to assist in simulating portfolio performance or assist in com-piling portfolios for given scenarios. The most popular models used byportfolio-planning utilities in this study were the PowerSimm model and theAuroraXMP model. NorthWestern Energy used the PowerSimm model fortheir 2013 IRP modelling [34]. The model works in two general stages. First,PowerSimm builds simulations of future prices using regression relationshipsof energy supply and observed price patterns. The model attempts to keep79relationships between weather, load, wind, hydro, market prices intact, tomore accurately simulate future conditions. Distributions are assigned toeach variable, and Monte Carlo simulation is carried out to produce randomscenarios. The model then simulates the operation of the portfolios generat-ing units over a particular scenario. The projections from the first step arefed into the operational module, which simulates hourly generation costs forthe portfolios. The model then optimises the operation of the portfolio overthe planning horizon by trying to minimise generation costs. Both Avista andIdaho Power used the AuroraXMP model, but for different purposes. Avistaused the model to generate 500 scenarios by random sampling from sets ofgas prices, loads, water years, thermal outages, and wind penetration lev-els [35]. The utility then used their internally developed PRiSM model tobuild optimal portfolios for each of the scenarios. Idaho Power instead usedAuroraXMP for simulating the operation of their manually-compiled portfolios[6], in the same manner as NorthWestern Energy used PowerSimm.For scenario-planning methods, once the scenarios are developed, theyare input into the utility’s optimisation model. The model will compile aportfolio, using the screened resource options, that is optimal according tothe constraints given in the scenario. The utility must be very clear abouttheir definition of “optimal”, which follows from the development of the per-formance metrics identified earlier in the process. In some cases this is thelowest cost portfolio, while in others it is the least risky or the most stableportfolio. The models most commonly used were System Optimizer (“SO”)and Strategist. Both models select optimal resource combinations for a par-ticular input scenario, either using mixed integer or dynamic programmingoptimisation methods [30, 43]. An objective function describing the rela-tionship between variables and the value to be optimised is developed, andthe program run to minimise or maximise the objective function. PacifiCorp,Tennessee Valley Authority, Duke Energy, and BC Hydro all used SO in theirIRP planning [12, 31, 16, 30], while Strategist was used by Arizona Public80Service and Public Service Company of Colorado [5, 13].3.2.5 Portfolio-based planningPortfolio planning is so called because a utility first develops their portfolios,manually, and then simulates the performance of the portfolios. The processis illustrated in Figure 3.1. There are advantages to this method, as a utilityhas direct control over which resources are selected and in what quantities.This works particularly well for utilities that have limited options for addingnew generation and therefore have little difficulty in selecting combinationsof resources. An illustration of this is available in the three largest IOUsin California: Pacific Gas & Electric, Southern California Edison, and SanDiego Gas & Electric. Californian regulations mandate that utilities mustfollow a particular “loading order” for meeting shortfalls in power supply,namely that shortfalls must be first met by energy efficiency and demand-side management, then by renewables, and lastly by efficient fossil fuels [19,17]. Thus portfolio construction for these utilities is trivial.The portfolio planning method is also effective for smaller utilities thatdo not need to select a large number of resources to fill their load-resourcegap. An example of this was provided by Tacoma Power’s 2015 IRP, inwhich the utility determined that all future growth in their planning pe-riod could be met by their current capacity and by energy efficiency anddemand-side management measures [15]. However, the utility decided toconduct analysis on new resources in the event of unexpected changes toload and experimented with adding an additional 50 annual MW (aMW)of energy efficiency/demand-side management, wind, solar, combined cyclegas, Columbia River hydro purchase, or run-of-river hydro power purchase totheir existing portfolio. This method was effective because Tacoma Power’slow load growth allowed the energy gap to be filled by a single resource andmade the portfolio comparison relatively simple.Following on from the shared initial steps in Section 3.2.3, utility con-81Figure 3.1: Overview of portfolio-based planning processSelect metrics forportfolio comparisonDetermine load-supply gapand need for new resourcesScreen potentialresource optionsBuild portfoliosSimulate portfoliosAnalyse portfolio performanceIs portfolio performancesatisfactory?Review byUtilities Commissionand stakeholdersFinal portfolioNoYes82structs their portfolio, manually choosing a combination of resources thatsatisfies the requirements for load. Uncertainty management at this step usu-ally involves the utility constructing several different portfolios, with resourcecombinations that the utility is interested in or concerned about. Idaho Powerfor example developed 23 portfolios for their 2015 IRP, all featuring variouslevels of coal retirement because of uncertainty over Section 111(d) of theClean Air Act [6]. Stakeholder consultation may be part of the portfoliodevelopment process, mitigating the risk of later disagreement over resourcechoices.The utility then sets up the scenarios for simulating and optimising thedispatch of the portfolio. As the aim of the analysis is to stress underly-ing variables and observe the changes in portfolio performance, the utilitiesspecify a range for each of their key variables and then use Monte Carlosimulation to select from these distributions and develop a random scenario.Idaho Power selected natural gas price, customer load, and hydroelectric vari-ability as their stochastic variables, assigned distributions to each variable,and created 100 different scenarios consisting of random draws from the threedistributions, resulting in a distribution of costs for each portfolio.Finally, the utility uses the results from the simulation/optimisation todecide on a preferred portfolio, based on the performance criteria defined atthe beginning of the process and in consultation regulators and stakehold-ers. If cost was the main criterion, then the portfolio that had the lowestaverage cost should be selected as the preferred portfolio. If minimising therange of net present value of total portfolio cost is the aim, then the util-ity would choose the portfolio with the lowest spread of costs. Idaho Powertook the results of their 100 iterations and created graphs of portfolio costversus exceedance probability, comparing the 95%, 50% and 5% exceedanceprobabilities for all portfolios, and graphs of standard deviation versus ex-ceedance probability for all portfolios. Idaho Power also conducted tippingpoint analysis for two portfolios, one with a high penetration of PV solar and83one with 300 MW of pumped hydro storage. The utility wanted to investi-gate the effect of variation in capital cost on the overall cost of the portfolios.By varying only the capital cost of the solar and of the pumped hydro, theutility could determine their preferred portfolio (from among the two in thetipping analysis) if the capital costs are known. Eventually, this combinedanalysis led to the choice of a preferred portfolio.3.2.6 Scenario-based planningScenario planning involves a utility constructing combinations of futures andusing an optimisation model to compile portfolios that are optimal in eachscenario. An illustration would be a utility concerned about variation in gasprices and load in the future, which then constructs scenarios with combina-tions of high, medium, and low gas price, and high, medium, and low load,resulting in nine different scenarios for use over the planning horizon. Usingan optimisation model, the utility would then set up their objective func-tion to reflect the factors to be optimised, and would run the model to buildportfolios for each scenario. The resulting portfolios are then assessed againstperformance metrics and the best performing one chosen as a preferred port-folio. Scenario planning has the advantage of not requiring manual portfoliocompilation, which can be a difficult task in the case of large utilities withsignificant load-resource gaps. For example, Tennessee Valley Authority pro-jected the energy gap in their 2015 IRP to range from 10,000 to 50,000 GWhover the planning horizon [16]. As it is unlikely a single resource could bridgean energy gap of this size, it becomes necessary to consider combinations ofresources, thus making the analysis more complex.Scenario planning shares several initial steps with portfolio planning,namely selecting metrics for portfolio assessment, determining the load-resourcegap, and screening potential new resource additions. The process divergesfrom this point as shown in Figure 3.2.Scenario development can occur at any point before modelling but com-84Figure 3.2: Overview of scenario-based planning processSelect metrics forportfolio comparisonDetermine load-supply gapand need for new resourcesScreen potentialresource optionsBuild scenariosCompile optimal portfoliofor each scenarioAnalyse portfolio performanceIs portfolio performancesatisfactory?Consultation withregulators andstakeholdersSelection asfinal portfolioNoYes85monly occurs near the beginning of the process. The number of scenariosused can range from several [5, 31, 16]) to hundreds [12] or even thousands[30]. Utilities using fewer scenarios tended to be more deliberate in theirscenario development, selecting disparate variable combinations to broadenthe conditions covered. Utilities using tens or hundreds of scenarios had amore continuous spectrum of scenarios with fewer differences between com-binations. In choosing the variables and values to go into the scenarios (e.g.high/low gas prices, carbon prices, energy prices, etc.), the utility is mak-ing decisions about the uncertainties of most concern to them and managingthese by ensuring they are included in the simulation.Once portfolios have been constructed for all scenarios, using an optimi-sation model as mentioned in Section 3.2.4, the utility will assess portfolioperformance and decide if further testing is required. Using the metrics de-veloped earlier, utilities can rate the performance of each portfolio and decidewhich portfolios are most promising. The choice of preferred portfolio canbe made at this stage, or else a subset of the portfolios may be selectedfor further analysis such as sensitivity testing or Monte Carlo simulation torandomly sample from distributions for the underlying variables. Utilitiescan also look for trends in the resources chosen in the portfolios. If, forexample, a resource option is selected in many portfolios, this may indicatea particularly stable/robust resource across a range of futures, and thus agood candidate for inclusion in the eventual preferred portfolio. To generatealternative forecasts for sensitivity purposes, utilities may re-run their fore-casting models with different assumptions, such as higher economic growthleading to increased demand, or with new sources of gas affecting market en-ergy prices. Variation in inflows often comes from historical records of suchdata, and a utility may choose to run simulations with all years of a waterrecord or with the highest, lowest, and average flows. Utilities may focus onscenarios that are probabilistically likely, looking for instance at the meanand the 30th and 70th percentiles of a variable, or may consider the worst-case86scenarios if they are particularly risk averse.3.2.7 Assessment criteria for portfoliosOnce a portfolio is built, a utility can use it “as is” or carry out further testingon the portfolio. It should be noted that a portfolio developed through thescenario planning process is a portfolio optimised for that particular scenarioonly and is not optimised over all potential realisations of the stochasticvariables. Thus it is important to note that while this scenario may accuratelyreflect a particular future, it does not guarantee that the portfolio is optimalin a different future. In addition, the scenario is assumed to be constant forthe entire planning horizon, thus the process is essentially the developmentof scenario trees for which optimal portfolios are developed. There is noability to switch between different branches of the tree at any stage, as indynamic programming. All of the utilities in the study that used scenario-based planning conducted some further analysis on their most promisingportfolios. This is somewhat akin to the analysis that occurs in portfolio-based planning once a portfolio has been selected. Idaho Power providesan excellent example of this process, where, once their portfolios have beenmanually compiled, the utility uses AuroraXMP to simulate the operation ofthe portfolio with different values of three underlying variables: natural gasprice, load, and hydroelectric variability. Probability distributions of thesevariables (normal or log-normal) were derived and Monte Carlo simulationused to randomly sample from these values for 100 iterations, resulting in100 different costs for a particular portfolio. These values give an indicationof the spread of the portfolio costs and therefore of the vulnerability of theportfolio to changes in underlying conditions (a proxy measure for risk).Assessment of portfolio performance requires the development of perfor-mance metrics. These depend on a utility’s policies and objectives as wellas their operating environment. Metrics can vary greatly depending on theobjectives of the utility. The utilities studied in the jurisdictional review used87a variety of metrics to assess portfolio performance; observed were• Mean cost [5, 10, 12, 16]• Standard deviation of costs [6] or “Financial risk”[16]• Tail VaR(85, 90, 95) [10, 12]• Fuel diversity [5]• Water use [5]• CO2 emissions [5, 12]• Flexibility [16, 13].Each metric addresses a particular aspect of uncertainty and provides theutility with slightly different information about their portfolio. Mean cost orexpected value of cost is a common metric used to show the most likely costof implementing a particular portfolio.Standard deviation of costs shows how greatly the cost may vary andtherefore what are reasonable contingencies to put in place. A smaller stan-dard deviation of costs would indicate a portfolio that is stable across a widevariety of futures and therefore has a lower risk of exceeding cost thresholds.TVaR demonstrates the average of the most extreme values for a givenpercentile of the distribution, giving an indication of a worst case scenario.Portfolios with lower TVaR values are therefore less risky.Fuel diversity or resource mix suggests how vulnerable a portfolio willbe to changes in fuel prices, as a system heavily dependent on one mainresource will be significantly more vulnerable to changes in the price of thatresource than a portfolio with a variety of generation options. A portfoliowith a greater diversity of fuel sources would therefore be considered to havelower risk than one that relies heavily on a single resource. The variance ofa portfolio, in the statistical sense of the term, could be a useful measure88of the fuel diversity and spread of the portfolio’s resource distribution. Asthe variance indicates the average of the spread of the variables about themean, a low value of variance indicates that the energy generation is spreadrelatively evenly among the various generation options. A high value ofvariance indicates that a few of the resources are dominating the resourcemix.Water use was only a metric for Arizona Public Service, which operatesin a region of scarce water resources and therefore has an interest in choosinggeneration that does not rely on heavy water usage. This metric could becalculated by historical water usage of a similar sized plant and by interpo-lating for the scale of a new plant. A high water use would be undesirablefor a resource.CO2 and other green house gas emissions were used as a metric by manyutilities concerned about new regulation that would impose a price on car-bon emissions, affecting the dollar per MWh ratio of a high-CO2 emittingresource and therefore portfolio make-up. If utilities are concerned about theprice on carbon, a portfolio with a lower level of carbon emission will be oflower risk. The volume of carbon emissions would be an output of the port-folio simulation process, and direct comparison between portfolios would bepossible. If there are a range of carbon emissions for each portfolio due to thesimulation process, then measures like the mean and the standard deviationcan also be use for this analysis.Flexibility was used as a metric by utilities that were interested in in-tegrating higher levels of renewable energy from intermittent resources suchas wind. As an illustration, the Public Service Company of Colorado wasconsidering the addition of 1,200 MW of wind energy to their system in their2011 IRP and therefore was interested in adding resources that could man-age the variability of wind, e.g. using natural gas peakers. Higher levelsof flexibility resources in a portfolio would indicate lower risk of the utilityhaving shortfalls in capacity. Several metrics for measuring flexibility have89been proposed in literature by a recent NREL report [36]. These included:• Percent of GW of installed capacity capable of load-following relativeto peak demand• GIVAR III flexibility scoring framework, where power area size, gridstrength, interconnection (transmission), and number of power marketsare combined into an overall flexibility score• Maximum upward or downward change in load that the system is capa-ble of managing in a given time period from a given initial operationalstate, and• Expected percentage of incidents in a given time period where the sys-tem cannot cope with the changes in net load [36]Metrics can be combined into scorecards for comparing portfolios. Forexample, each portfolio could be ranked from n to 1, where n is the numberof portfolios being tested, for each separate metric. This would result in eachportfolio having a score reflecting their relative merit among the portfoliosas a whole.For example, consider the case of a utility comparing three portfolios:A, B, and C. Each of these will be compared with four different metrics:expected cost, standard deviation of costs, fuel diversity, and tons of CO2emitted. Based on the values calculated for each of the metrics in Table 3.3,each portfolio can be given a ranking out of three for each metric (Table3.4). This analysis would show that portfolio A performed most stronglywhen considered across all the metrics.If a utility was very concerned about a particular metric, they could applya multiplier to ensure that metric carries greater weight in the analysis. Forexample, the utility could rank their metrics in order of importance – 1) meancost, 2) CO4 emissions, 3) standard deviation of costs, and 4) fuel diversity– and give each of these a weight such that the total adds up to 100 percent90Table 3.3: Example of a scorecard for comparing portfolios using metrics andcalculationsPortfolio Expected cost Standard deviation Fuel diversity CO2 emissionsUnit Millions Millions Statistical variance Total tons CO2A 30 5 189 100B 25 4 589 500C 40 3 322 400Table 3.4: Example of a scorecard for comparing portfolios using metrics,with the metric rankings instead of the calculated numbersPortfolio Mean cost Standard deviation Fuel diversity CO2 emissions TotalA 2 3 1 1 7B 1 2 3 3 9C 3 1 2 2 8Table 3.5: Example of a score card using weights to reflect utility prioritiesamong metricsWeights 40% 30% 10% 20%Portfolio Mean cost Standard deviation Fuel diversity CO2 emissions TotalA 0.8 0.9 0.1 0.2 2.0B 0.4 0.6 0.3 0.6 1.9C 1.2 0.3 0.2 0.4 2.191(see Table 3.5). This can result in significantly different relative rankings forportfolios, as in this case portfolio B is the best performing.Metrics can also be useful for direct trade-offs between two performancecharacteristics of one portfolio at a time. For example if a utility is interestedcomparing portfolio mean cost to level of risk, they can plot mean cost againsta measure of risk (say standard deviation of costs or TVaR) and find thelowest-cost portfolio for a given level of risk i.e. the “efficient frontier”,trading off between the two metrics. In addition, this sort of plot wouldallow a utility to observe incremental cost changes to obtain a lower or higherlevel of risk. This is the approach advocated by the Northwest Power andConservation Council in their sixth regional power plan [10]. While thedisadvantage of this approach is increased manipulation of the data, theadvantages for decision-making are significant: direct comparison betweencontradictory metrics and ease of identification of the “best” portfolios underconsideration.3.3 Application to the BC Hydro systemBritish Columbia Hydro and Power Authority (BC Hydro) is the main elec-tricity utility in British Columbia, and fourth largest in the WECC [23].The utility operates 31 hydroelectric generating stations and two thermalstations, providing over 43,000 GWh to a customer base of 1.9 million cus-tomers [44, 30]. The vast majority of BC Hydro’s energy comes from theirhydro-power generation assets on the Peace and Columbia Rivers (see Figure3.3 from [45].BC Hydro, as a crown corporation, adheres to the Clean Energy Act of2010. This includes the province’s energy objectives (Section 1 of the CEA),of which the most pertinent to BC Hydro’s IRP process are: (a) to achieveelectricity self-sufficiency; (b) to take demand-side measures and to conserveenergy, including the objective of the authority reducing its expected increase92in demand for electricity by the year 2020 by at least 66%; (c) to generateat least 93% of the electricity of British Columbia from clean or renewableresources and to build the infrastructure necessary to transmit that electricity[46].The utility conducts an IRP process every five years [30] and files thisdocument with the Minister, British Columbia Ministry of Energy and Mines.BC Hydro’s IRP process uses a scenario-based method with the SO modelto build resource portfolios (see Figure 3.4). The SO model’s main objectiveis to minimise the present value of costs net of trade revenue and to ensurethat a number of constraints are met. As a large number of portfolios aregenerated using this method, BC Hydro selects a subset of these portfoliosto simulate further. Currently each of the subset of portfolios is simulated inthe utility’s HYSIM model and Generalized Optimization Model (GOM) toobtain the feasible operational generation and reservoir pool schedule, andsystem and individual plant incremental costs, as well as the benefits thataccrue to the system such as trade benefits from power import/export andshaping benefits. This gives a detailed picture of the actual operation of aportfolio generated by System Optimizer. An example of a finalised portfolio,as presented in the appendices for chapter 6 of the 2013 IRP [30], is shownin Figure 3.6, where each resource has dependable and installed capacity andenergy, and an in-service date.93Figure 3.3: BC Hydro’s generation and transmission system (BC Hydro)94BC Hydro’s current process simulates a portfolio’s operation over 60 dif-ferent water years using the HYSIM and GOM models. For these runs,market price forecasts are inflated/deflated to account for dry/wet years’impacts on the Mid C market. This results in a range of costs for the dif-ferent water years, giving an indication of the spread of the portfolio costs.However, inflows are not the only variable that is uncertain, and variableslike load and gas and power price forecasts can also be varied to assess theperformance of the portfolio, based on the inputs to System Optimizer.3.3.1 Recommendations for BC Hydro’s IRP processbased on the results of this studyOne issue with the current process lies with the performance assessment cri-teria used to choose the preferred portfolio. Currently the portfolio with thelowest net cost (after accounting for trade revenue) is chosen as the preferredportfolio, which neglects uncertainties in many variables. The current useof the HYSIM and GOM models to simulate portfolio performance with 60different water years could be augmented by also simulating with differentload forecasts, energy prices, and gas prices, leading to broader picture ofthe potential variation in portfolio performance. A larger number of metricsthat represent the uncertainties of the portfolios could then be applied to as-sess the data, addressing more uncertainties and helping the utility developa more robust portfolio. A schematic of an updated process is presented inFigure Conclusions and policy implicationsWe have examined the various long-term planning methods used by a numberof utilities in North America and identified portfolio planning and scenarioplanning as two key methods used to manage the uncertainty inherent in long95Figure 3.4: BC Hydro’s current portfolio development processPrice ForecastReview alternativesfor generationProjectsResource optionsVariable Cost EvaluationFixed Cost EvaluationEvaluate Reliabilityand Non-Power ImportsLoad ForecastPortfolioHYSIM/GOM runsPortfolio performance onBC Hydro’s systemRecommended portfolioSystem OptimizerFigure 3.5: Suggested portfolio development and assessment processPrice ForecastReview alternativesfor generationResource optionsVariable Cost EvaluationFixed Cost EvaluationEvaluate Reliabilityand Non-Power ImportsOther metricsLoad ForecastPortfoliosPresecreening tominimise portfoliosHYSIMGOMPerformancecharacteristicsMetrics, scorecards,trade-off analysisRecommendedportfolioSystem Optimizer96Table 3.6: Example of a BC Hydro resource portfolio (BC Hydro 2013 IRP)Year Resource Selected Capacity - MW Energy - GWhInstalled Dependable Firm Total2023 Site C 1,100 1,100 5,100 5,1002028 GMS units 1-5 Cap Increase 220 2202029 MSW2 LM 25 24 208 2082030 Revelstoke Unit 6 500 488 26 262032 Pumped Storage LM 1,000 1,0002033 Wind PC21 99 26 371 3712033 Wind PC28 153 40 591 5912034 Wind PC13 135 35 541 5412034 Wind PC16 99 26 377 3772034 Wind PC19 117 30 441 4412035 Wind PC10 297 77 1,023 1,0232036 MSW1 VI 12 12 100 1002036 Biomass VI 30 30 239 2392036 Run of River LM 80 100 62 10 174 2232037 Wind PC09 207 54 713 7132037 Wind PC15 108 28 382 3822037 Biomass PR 28 28 223 2232037 Biomass LM 30 30 239 2392038 Wind PC14 144 37 527 5272038 Wind PC20 159 41 610 6102038 Pumped Storage LM 1,000 2,0002039 Wind PC11 126 33 473 4732039 Wind PC41 45 12 155 1552039 Wind PC42 63 16 219 2192040 Wind PC18 138 36 486 4862040 Wind VI14 35 9 114 11497term planning. The advantages and disadvantages of both methods have beendiscussed and their explicit treatment of uncertainty examined. Portfolio-based planning has several features that are advantageous in planning underuncertainty. These include: direct selection of portfolio resource combina-tions, which allows utilities to be explicit in avoiding particularly uncertainresources; simulation of a portfolio with randomised scenarios, providing ex-cellent understanding of an individual portfolio’s performance in a wide rangeof futures; and easier decision making given that portfolio-planning requiresmanual construction of resource combinations and therefore rarely resultsin more than tens of portfolios. However, this method appears to be bestsuited to utilities with small load-resource gaps and/or utilities with strictregulations on the type of resources that can be added, as both of these con-ditions simplify the portfolio construction process. Scenario-based planninglikewise has several advantageous features: portfolio construction in the casewhere many resources are required is very difficult to carry out manuallyand is therefore simplified by the use of an optimisation model; use of mod-els for construction of portfolios can be more easily shown to be be unbiasedif stakeholders express concerns; and a large number of scenarios, coveringmany potential futures, yields an equally large number of portfolios that canprovide indications of stable resources across disparate futures. This methodis perhaps best employed when a utility has a large load-resource gap suchthat satisfying projected load requires combinations of tens of resources. Insuch cases it becomes difficult to build resource combinations and test theirperformance without the use of modelling. It is also interesting to note thatthis method essentially requires an extra step compared to portfolio-planning,as sensitivity testing after portfolio development for scenario planning is car-ried out very similarly to simulation of manually constructed portfolios inportfolio-planning (compare Figure 3.1 and Figure 3.2). Our findings discussthe uncertainty and the method of managing that uncertainty in each stepof both planning processes, providing utilities with a summary of methods98and potentially assisting in modification of their planning processes.For both methods, our research has noted several ways of assessing port-folio performance to minimise uncertainty. These are to use Monte Carlosimulation to randomly sample from several underlying variables, therebydeveloping random sensitivity scenarios, and to generate ranges of portfo-lio performance for each variable. These performance ranges can then beassessed by using metrics such as the mean, standard deviation, and tailvalue-at-risk. Performance indicators assessed this way could include portfo-lio net present value, CO2 emissions, water use, and so on. Other potentialassessment metrics could include portfolio resource diversity and flexibil-ity. Assessing a portfolio against several rather than one metric allows fora broader picture of performance and provides high level information fordecision-makers and policy developers. The use of explicit metrics also hasimplications for relationships with stakeholders, as a decision-making pro-cess using quantifiable metrics appears to be more transparent, leading toimproved stakeholder engagement.Based on this study, one main recommendation is made for improving BCHydro’s handling of uncertainty in their IRP process: expansion of portfoliosimulation combined with extension of existing metrics for portfolio compar-ison. This could have value in particular for utilities with large hydroelectricgeneration assets, where the storage and valuation of water is necessary. Fur-ther research could investigate other planning methods for uncertainty, suchas use of dynamic programming, and its application to systems similar toBC Hydro’s.99Chapter 4ConclusionChapter 1 has presented a survey of the IRP processes for a sample of NorthAmerican electric utilities. Our research suggests that two main methodsof planning are used: portfolio-based planning and scenario-based planning.A notable point was that scenario-planning was more popular among largerutilities (i.e. greater than 4,000 MW of generating capacity), while smallerutilities favoured portfolio-planning. This trend was not observed in Califor-nian utilities, due to the nature of the state’s electricity market and regula-tion.Chapter 2 analysed BC Hydro’s current long-term planning process andidentified it as following the practice of scenario-planning. Examination ofthe portfolio simulation after portfolio construction with SO suggested thatwhile water storage and water value were being well simulated, other con-tributing factors to uncertainty such as energy price and load were not con-sidered. Inclusion of these variables in the simulation would allow BC Hydroto assess portfolios over a broader range of characteristics, which is beneficialfor uncertainty management. However, increased simulation for more vari-ables also complicates the assessment of which portfolio performs best. Forthis reason, we recommend the development of performance metrics otherthan lowest cost – examples being width of cost distribution, flexibility, and100fuel diversity, among others – and the formalisation of portfolio performanceanalysis in the form of explicit metrics and scorecards. In addition, there iscurrently no mechanism for considering the benefits brought by a portfolio,such as generation of income by export, in addition to the costs of a port-folio. Consideration of such benefits has the potential to modify a portfoliobuilt by SO. Therefore our research suggested that BC Hydro implement thefollowing recommendations:• Continue to use the current capacity expansion model (System Opti-mizer) for developing portfolios from multiple scenarios• Use HYSIM/GOM for further sensitivity analysis of individual port-folios, expanding beyond the 60 water years and expanding to includealternative loads, gas prices, and energy prices• Develop metrics and scorecards and/or efficient frontier analysis forassessing portfolio performance• Develop a GUI for running HYSIM to streamline the process of runningthe increased number of simulations. This can be done by enhancingthe existing GOM GUI• Investigate the potential use of dynamic programming for portfolio se-lection and compare its output with that of System OptimizerIn Chapter 3 the various long-term planning methods were broken downinto their component steps and each step analysed for its effect on uncertaintymanagement.The advantages and disadvantages of both methods have beendiscussed and their explicit treatment of uncertainty examined. Portfolio-based planning offered advantages in direct selection of portfolio resourcecombinations, which allows utilities to be explicit in avoiding particularlyuncertain resources; simulation of a portfolio with randomised scenarios,providing excellent understanding of an individual portfolio’s performance101in a wide range of futures; and easier decision making given that portfolio-planning requires manual construction of resource combinations and thereforerarely results in more than tens of portfolios. However, the method appearedto be limited in application, being best suited to utilities with small load-resource gaps and/or utilities with strict regulations on the type of resourcesthat can be added, as both of these conditions simplify the portfolio construc-tion process. Scenario-based planning had advantages in simplified portfolioconstruction through the use of an optimisation model for the case wheremany resources are required to fill the goad-resource gap; more easily justi-fied decisions as stakeholders can be shown that portfolio development is bya model; and a large number of scenarios, covering many potential futures,that yields an equally large number of portfolios that can provide indicationsof stable resources across disparate futures. For both methods, our researchhas noted several ways of assessing portfolio performance to minimise uncer-tainty. These are to use Monte Carlo simulation to randomly sample fromseveral underlying variables, thereby developing random sensitivity scenar-ios, and to generate ranges of portfolio performance for each variable. Theseperformance ranges can then be assessed by using metrics such as the mean,standard deviation, and tail value-at-risk. Assessing a portfolio against sev-eral rather than one metric allows for a broader picture of performance andprovides high level information for decision-makers and policy developers.102Bibliography[1] Eskom. Integrated resource plan for electricity 2010-2030. 2011. url:[2] Department of Employment, Economic Development and Innovation.Queensland Energy Management Plan. Report. Queensland Govern-ment, 2011. 24 pp. url: (visited on 10/03/2016).[3] R. Wilson and B. Biewald. Best practices in electric utility integrated re-source planning. Synapse Energy Economics, 2013. url: (visited on 10/17/2016).[4] E. Hirst and C. Goldman. “Creating the future - integrated resourceplanning for electric utilities”. In: Annu. Rev. Energy 16 (1991), pp. 91–121. issn: 1056-3466. doi: 10.1146/[5] Arizona Public Service. 2014 integrated resource plan. Phoenix, 2014.url: https : / / www . aps . com / library / resource % 20alt / 2014 _IntegratedResourcePlan.pdf (visited on 09/23/2016).[6] Idaho Power Company. Integrated resource plan 2015. 2015, pp. 1–164.url: (visited on 09/21/2016).103[7] Idaho Power Company. Energy sources. [Online]. 2015. url:[8] Los Angeles Department of Water and Power. 2014 Power IntegratedResource Plan. Los Angeles Department of Water and Power, 2014.url:;jsessionid=r8jTXvqY9Ln7L5NsVPjYYQ1rSr1kCcdzwKbxjP9W211V1j1MQfpk!213437157?_adf.ctrl-state=11smxe8s1n_4&_afrLoop=1081875976128968&_afrWindowMode=0&_afrWindowId=null#@?_afrWindowId=null&_afrLoop=1081875976128968&_afrWindowMode=0&_adf.ctrl-state=8k32xmgt_4 (visited on 09/21/2016).[9] California Energy Commission. Utility Annual Power Content Labelsfor 2014. [Online]. 2014. url:[10] Northwest Power and Conservation Council. Sixth northwest conserva-tion and electric power plan. Feb. 2010. url: (visited on 09/21/2016).[11] Gary W. Dorris, David K. Bellman, and Douglas M. Logan. Assess-ment of the regional portfolio model. Northwest Power and Conserva-tion Council, 2012. url:[12] PacifiCorp. Integrated resource plan. Portland, 2013. url: (visited on 09/20/2016).[13] Public Service Company of Colorado. 2011 electric resource plan. XcelEnergy, 2011. url: (visited on 11/12/2015).104[14] Tacoma Power. Navigating today’s challenges in search of a bright fu-ture: Tacoma Power’s 2013 integrated resource plan. Tacoma: TacomaPublic Utilities, 2013. url:[15] Tacoma Power. Soaring into the future: 2015 integrated resource plan.Tacoma Public Utilities, 2015. url: (visited on 09/20/2016).[16] Tennessee Valley Authority. Integrated resource plan. 2015. url:[17] Doug Allen, Eric Cutter, Michael King, Amber Malone, William Mor-row, Arne Olson, Ren Orans, C.K. Woo, John Candelaria, Susan Lee,Carl Linvill, and Suzanne Phinney. San Francisco: California Pub-lic Utilities Commission, 2008. url: (visited on 09/21/2016).[18] California Public Utilities Commission. California renewable portfoliostandards (RPS). 2007. url:[19] Planning assumptions and scenarios for use in the CPUC RulemakingR.13-12-010 (the 2014 long-term procurement plan proceeding), andthe CAISO 2014-15 transmission planning process. 2014. url: (visited on09/21/2016).[20] J.P. Pfeifenberger, K. Spees, and S.A. Newell. Resource adequacy inCalifornia: options for improving efficiency and effectiveness. The Brat-tle Group, 2012. url:[21] Pacific Gas & Electric Company. 2006 long-term procurement plan. SanFrancisco: California Public Utilities Commission, 2006. url:[22] Testimony of Southern California Edison Company on Track I Issues.Rosemead, California, 2011. url: http : / / www3 . sce . com / sscc /law/dis/dbattach3e.nsf/0/2DFDDFB493ED2ABF882578C0005BC329/$FILE/R.10-05-006_Track+I+2010+LTPP+-+SCE-1+Testimony+on+Track+I+Issues.pdf (visited on 09/20/2016).[23] Jordan Wilkerson, Peter Larsen, and Galen Barbose. “Survey of west-ern U.S. electric utility resource plans”. In: J. Energy Policy 66 (2014),pp. 90–103. doi: 10.1016/j.enpol.2013.11.029.[24] Pacific Gas & Electric Company. PG&E overview. 2014. url:[25] Pacific Gas & Electric Company. Bundled procurement plan. Rulemak-ing: 10-05-006. California Public Utilities Commission, 2011. url: (visited on09/21/2016).[26] California Public Utilities Commission. Collaborative review of plan-ning models. California Public Utilities Commission, Apr. 2014, pp. 1–17. url: AAF9- 1F17B7B77BCE/0/CombinedLongTermProcure2014OIR_Report_CollaborativeReview.pdf.[27] Edison International. Fact sheet: Southern California Edison. 2015.url:[28] Eskom. Company information. 2015. url:[29] L.J. Irvine. Comparison and evaluation of planning processes used forlong-term resource planning among North American utilities. Internalreport. 2016.[30] BC Hydro. Integrated Resource Plan. 2013. url: in- bc/planning- for- our- future/irp/current-plan/document-centre/reports/november-2013-irp.html (visitedon 09/21/2016).[31] Duke Energy Indiana. 2013 integrated resource plan. 2013. url: (visitedon 09/23/2016).[32] J.D. McCalley. Production Costing Model Fundamentals. 2016. url:[33] R.E. Bellman. Dynamic Programming. New Jersey: Princeton Univer-sity Press, 1957. isbn: 0486428095.[34] NorthWestern Energy. 2013 Electric Supply Resource Procurement Plan.NorthWestern Energy, 2013. url:[35] Avista. 2013 Electric Integrated Resource Plan. Avista Utilities, 2013.url: (vis-ited on 09/21/2016).107[36] J. Cochran, M. Miller, O. Zinaman, M. Milligan, D. Arent, B. Palmintier,M. O’Malley, S. Mueller, E. Lannoye, A. Tuohy, B. Kujala, M. Sommer,H. Holttinen, J. Kiviluoma, and S.K. Soonee. Flexibility in 21st centurypower systems. Technical report. 2014. url: (visited on 09/21/2016).[37] E. Hirst and M. Schweitzer. “Uncertainty: A Critical Element of In-tegrated Resource Planning”. In: The Electricity Journal 2 (6 1989),pp. 16–27. issn: 1040-6190. doi: 10.1016/1040-6190(89)90022-5.[38] Mark Bolinger and Ryan Wiser. “Utility integrated resource planning:An emerging driver of new renewable generation in the western UnitedStates”. In: Refocus 6.6 (2005), pp. 20–24. issn: 1471-0846. doi: 10.1016/S1471-0846(05)70483-2.[39] Alberta Electric System Operator. Competitive Process. 2016. url: process/ (visited on10/04/2016).[40] Poul Østergaard, Isabel Soares, and Paula Ferreira. “Energy efficiencyand renewable energy systems in Portugal and Brazil”. In: Int. J. Sus-tain. Energy Plan. Manag. 2 (2014), pp. 1–6. issn: 2246-2929. doi:10.5278/ijsepm.2014.2.1.[41] Jonathan A. Schachter, Pierluigi Mancarella, John Moriarty, and RitaShaw. “Flexible investment under uncertainty in smart distributionnetworks with demand side response: Assessment framework and prac-tical implementation”. In: Energy Policy 97 (2016), pp. 439–449. issn:0301-4215. doi: 10.1016/j.enpol.2016.07.038.[42] San Diego Gas & Electric Company (U 902 E) Draft 2014 Long-TermProcurement Plan. California, 2014. url: (visited on 09/21/2016).108[43] Northern States Power Company. 2016-2030 upper midwest resourceplan. Minnesota: Xcel Energy, 2014, pp. 1–45. 45 pp. url: (visited on 09/19/2016).[44] BC Hydro. About BC Hydro. 2016. url: (visited on 09/19/2016).[45] BC Hydro. BC Bulk Transmission System. 2007. url: (visited on 10/25/2016).[46] Clean Energy Act. British Columbia: Legislative Assembly, 2010. url: (visited on 09/20/2016).Bill 17, 39th Parliament, 2nd Session. Assented to Jun. 3, 2010.[47] Avista. Quick Facts. 2015. url:[48] Duke Energy. Fast Facts: Duke Energy Corporation. 2014. url:[49] NorthWestern Energy. At a glance Montana. 2015. url:[50] PacifiCorp. PacifiCorp Fact Sheet. Portland, 2014. url:[51] Riverside Public Utility. 2014 Power Supply Integrated Resource Plan.Riverside Public Utility, 2014. url:[52] M. Howells, H. Rogner, N. Strachan, C. Heaps, H. Huntington, S.Kypreos, A. Hughes, S. Silveira, J. Decarolis, M. Bazilian, and A.Roehrl. “OSeMOSYS: The open source energy modelling system. Anintroduction to its ethos, structure and development”. In: J. Energ.Policy 39 (July 2011), pp. 5850–5870. doi: 10.1016/j.enpol.2011.06.033. Online.[53] M. Howells. Contribution to the OSMOSYS forum Open Source energyMOdeling SYStem. 2009, pp. 1–14. url:[54] System Analysis — 2060 Project. University of Victoria. Oct. 2015.url:[55] Stockholm Environment Institute, ed. Long-range Energy AlternativesPlanning System (LEAP) — Climate Planning. 2015. url:[56] LEAP: User Guide. 2015. url:[57] Charles Heaps. LEAP: Applications. 2015. url:[58] F. Rogan, C.J. Cahill, H.E. Daly, D. Dineen, J.P. Deane, C. Heaps,M. Welsch, M. Howells, M. Bazilian, and B.P. O´ Gallacho´ir. “LEAPsand bounds – an energy demand and constraint optimised model of theIrish energy system”. In: Energy Efficiency 7 (3 Oct. 2013), pp. 441–466. doi: 10.1007/s12053-013-9231-9. Online.[59] Energy Exemplar. USBR Plexos demo. Nov. 2012. url: online.110[60] Energy Exemplar. Plexos applications. Dec. 2014. url:[61] G.K. Woods. Plexos for power systems – advanced simulation top-ics. Jan. 2015. url:[62] Eskom. Integrated resource plan for electricity 2010-2030. IRP 2010.South Africa: Eskom, 2011, pp. 1–73. url:[63] Epis Inc. Power market forecasting. 2015. url:[64] R Pagoaga and J. Peterson. AURORA model overview. Idaho Power,Dec. 2012. url:[65] EPIS chooses Gurobi. May 2014. url:[66] Y. E. Lee and Y. B. Jung. “Challenges of nuclear power for sustainablerole in Korean energy policy”. In: J. Energ. Convers. Manage. 49 (7July 2008), pp. 1951–1959. doi: 10.1016/j.enconman.2007.09.031.[67] T. Nakawiro and S. C. Bhattacharyya. “Electricity capacity expansionin Thailand: an analysis of gas dependence and fuel import reliance”.In: Energy 33 (5 May 2008), pp. 712–723. doi: 10.1016/[68] Planning and Economic Studies Section, ed. Wien Automatic SystemPlanning (WASP) Package. A computer code for power generating sys-tem expansion. Computer manual series 16. Version WASP-IV. Vienna:International Atomic Energy Agency, 2001, pp. 1–32. 284 pp. url:[69] Riverside Public Utility. 2014 power supply integrated resource plan.City of Riverside, 2014, pp. 1–302. url:[70] NorthWestern Energy. 2013 electricity supply resource procurement plan.Montana, 2013. url: environment/2013- electricity- supply-resource-procurement-plan.[71] Ram M. Shrestha and Charles O.P. Marpaung. “Integrated resourceplanning in the power sector and economy-wide changes in environ-mental emissions”. In: Energy Policy 34.18 (2006), pp. 3801–3811. issn:0301-4215. doi: 10.1016/j.enpol.2005.08.017.[72] Statistics Norway. Electricity, 2014. 2015. url: (visitedon 10/03/2016).112


Citation Scheme:


Citations by CSL (citeproc-js)

Usage Statistics



Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            async >
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:


Related Items