Open Collections

Using SARIMAX to forecast electricity demand and consumption in university buildings

University of British Columbia

Forecasting electricity demand and consumption is critical to the optimal and cost-effective operation of buildings. Time-series forecasting methods identify and learn patterns with data sets and then use these patterns to predict future values. However, the traditional methods tend to fall short in working with seasonal data and external variables. As a result, many time-series forecasting methods are not applicable to electricity consumption data. This type of data is seasonal and highly affected by external factors such as outside air temperature or humidity. Seasonal AutoRegressive Integrated Moving Averages with eXogenous regressors (SARIMAX) is a class of time-series forecasting models that explicitly deals with seasonality in data and external variables. This research used SARIMAX to predict daily average electricity consumption and daily peak demand in two university buildings. Electricity consumption data from 2017 to 2019 were split into training and test sets. The data from 2017 and 2018 were used as the training set, while values from 2019 were used as the test set. Daily average temperature and humidity were used as external variables. A grid search was conducted to find the best model for each building. Next, the residuals of the models were checked to see whether they satisfied the modelling assumptions. Afterwards, the models were used to predict values in 2019. The performance of the models was calculated using 2019 as test data. The method was able to successfully use temperature and humidity as external variables and identify weekly patterns. The degree of forecast accuracy was different between the two buildings. The mean absolute percentage error (MAPE) of predicted values in 2019 was 4.1% in the first building and 12.8% in the second building. The models can be used to make informed decisions about the renovation or recommissioning activities in the building, detect abnormalities in consumption trends, and quantify energy and cost-saving measures. They can also be used to identify and quantify the effects of sudden changes or disruptions in the system or the way occupants behave.

Text

2020-05-25

Attribution-NonCommercial-NoDerivatives 4.0 International

http://hdl.handle.net/2429/74570

Civil Engineering

University of British Columbia

UBCV

http://creativecommons.org/licenses/by-nc-nd/4.0/