UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Power management in a sensor network for automated water quality monitoring Shu, Tongxin 2016

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

Item Metadata

Download

Media
24-ubc_2016_may_shu_tongxin.pdf [ 2.08MB ]
Metadata
JSON: 24-1.0300325.json
JSON-LD: 24-1.0300325-ld.json
RDF/XML (Pretty): 24-1.0300325-rdf.xml
RDF/JSON: 24-1.0300325-rdf.json
Turtle: 24-1.0300325-turtle.txt
N-Triples: 24-1.0300325-rdf-ntriples.txt
Original Record: 24-1.0300325-source.json
Full Text
24-1.0300325-fulltext.txt
Citation
24-1.0300325.ris

Full Text

POWER MANAGEMENT IN A SENSOR NETWORK FOR AUTOMATED WATER QUALITY MONITORING by  Tongxin Shu  B.Sc., Xiamen University, 2014  A THESIS SUBMITTED IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF  MASTER OF APPLIED SCIENCE in THE FACULTY OF GRADUATE AND POSTDOCTORAL STUDIES (Mechanical Engineering)  THE UNIVERSITY OF BRITISH COLUMBIA (Vancouver)    April 2016  © Tongxin Shu, 2016 ii  Abstract Power management is crucial in remote environmental monitoring, especially when long-term monitoring is needed. Renewable energy sources such as solar and wind may be harvested for sustaining a monitoring system. Without proper power management, equipment within the monitoring system may become nonfunctional and as a consequence, the data or events captured during the monitoring process will become inaccurate as well. Based on reinforcement learning, this thesis develops and applies an adaptive sampling algorithm and duty cycling for power management in automated water quality monitoring with energy harvesting. The state of the water quality parameters in a water source such as a lake or river may change in an unpredictable manner (e.g., may remain stable or change abruptly) depending on many factors such as climate or environmental changes or those caused by humans (e.g., waste water discharge from factories, construction, farming, and litter). Ideally, the sampling rate that is used for a sensor signal should depend on the rate at which the signal changes. Hence, adaptive sampling scheme using reinforcement learning is used in the present work, for water quality monitoring. The energy consumption for signal acquisition, processing, and transmission all depend on the sampling frequency, either directly or indirectly. Hence, it is desirable for the sensor nodes to dynamically learn how to determine the best sampling frequency for a sensor signal, depending how the signal changes due to the environmental situations, and adjust the sampling rate accordingly. It is found that by dynamically changing the sampling frequency, the battery state can be maintained at an energy-neutral level. Duty cycling also contributes to achieving the same goal by scheduling the working and sleeping time of a sensor node. It is shown that by switching between the work mode and the sleep mode, a satisfactory battery state can be maintained. These two methods have different degrees of advantage and performance in power management, but it is shown that both iii  methods can achieve the energy neutrality while maintaining a high level of accuracy in the acquired data.  iv  Preface This thesis is an intellectual property of the author, Tongxin Shu. The endeavors in this thesis have been made under guidance of Dr. Clarence W. de Silva, which contributes to offer solutions in optimizing power management issues in water quality monitoring. Dr. Clarence W. de Silva proposed and supervised the overall water quality monitoring project, acquired funding and resources for the project, suggested the topic of the thesis, provided concepts and methodologies in addressing problems in the topic, and revised the thesis presentation. My colleagues Teng Li, Jiahong Chen and I collaboratively built up a hardware platform which was used for conducting experiments. The remaining parts of the work reported in thesis, including implementing the methodologies, analyzing data, and presenting the results, were done by me. v  Table of Contents  Abstract .......................................................................................................................................... ii Preface ........................................................................................................................................... iv Table of Contents ...........................................................................................................................v List of Tables .............................................................................................................................. viii List of Figures ............................................................................................................................... ix List of Symbols ............................................................................................................................. xi List of Abbreviations ................................................................................................................. xiii Acknowledgements .................................................................................................................... xiv Dedication .....................................................................................................................................xv Chapter 1: Introduction ................................................................................................................1 1.1 Background ..................................................................................................................... 2 1.2 Technological Trends...................................................................................................... 4 1.3 Research Objective ......................................................................................................... 5 1.4 Thesis Outline ................................................................................................................. 6 Chapter 2: System Architecture ...................................................................................................7 2.1 Water Quality Monitoring System .................................................................................. 7    2.1.1 The Structure of a Regular Node .................................................................................... 9 2.2 The Designed Sensor Node Prototype .......................................................................... 10    2.2.1 Sensor Node Details ...................................................................................................... 11    2.2.2 Levels of Energy Consumption .................................................................................... 13  vi  2.3 Field Testing ................................................................................................................. 15 Chapter 3: Adaptive Sampling ...................................................................................................17 3.1 Energy-neutral Battery State ......................................................................................... 17 3.2 Adaptive Sampling........................................................................................................ 18    3.2.1 Nyquist-Shannon Sampling Theorem ........................................................................... 19    3.2.2 Mean Relative Error ...................................................................................................... 22    3.2.3 Detection of Minimum Frequency Change................................................................... 23 3.3 Reinforcement Learning ............................................................................................... 25    3.3.1 Rewarding rules for adaptive sampling ........................................................................ 27 3.4 Simulation Example ...................................................................................................... 28 3.5 Conclusion .................................................................................................................... 30 Chapter 4: Self-aware Scheduling…………………………………………………….……….31 4.1 Self-aware Scheduling of Sleep-Awake Strategy ......................................................... 31 4.2 Related Work ................................................................................................................ 31 4.3 Sleep-Awake Strategies at Single Node Level ............................................................. 32    4.3.1 Battery State-Aware Sleep and Awake Strategy .......................................................... 33    4.3.2 Sleep-Awake Strategy Based on Data Standard Deviation .......................................... 35    4.3.3 Hybrid Sleep-Awake Strategy ...................................................................................... 39 4.4 Sleep-Awake Strategies at Distributed Sensor Network Level .................................... 39 4.5 Conclusion .................................................................................................................... 41 Chapter 5: Conclusion……………………………………………..…………………………...42 5.1 Developed Methods ...................................................................................................... 42 5.2 Future Work .................................................................................................................. 43 vii   Bibliography .................................................................................................................................44  viii  List of Tables  Table 2.1 Mode and its consumption  ........................................................................................... 13 Table 2.2 Module and its consumption  ........................................................................................ 14 Table 2.3 Measurement and its consumption ............................................................................... 14 Table 3.1 The rewarding rules ...................................................................................................... 28     ix  List of Figures Figure 1.1 Operation of the ICT platform fro water quality monitoring ........................................ 3  Figure 2.1 The proposed system framework for water quality monitoring .................................... 8 Figure 2.2 The flowchart of processing sensed data ....................................................................... 9 Figure 2.3 The schematic framework of a regular node ............................................................... 10 Figure 2.4 The developed prototype of a regular node ................................................................. 10 Figure 2.5 The regular node .......................................................................................................... 12 Figure 2.6 The sockets in the regular node ................................................................................... 12 Figure 2.7 The deployment of sensor nodes ................................................................................. 16 Figure 2.8 Landscape of the Yosef Wosk reflecting pool ............................................................ 17 Figure 3.2 How sampling affects battery state and accuracy ........................................................ 18 Figure 3.3 Original signal of DO data .......................................................................................... 19 Figure 3.4 Frequency response obtained through Fourier transform ............................................ 20 Figure 3.5 Determining the highest frequency of interest ............................................................ 21 Figure 3.6 The process of calculating MRE ................................................................................. 23 Figure 3.7 Detection of a change in the maximum frequency ...................................................... 24 Figure 3.8 The interaction between an agent and its environment…...……………….…………26 Figure 3.9 The systematic diagram for the algorithm……………………………………………28 Figure 3.10 Simulation with 1200 data samples……………….………………………………...30 Figure 4.3 Thresholds for battery state-aware cycle scheduling…………………..…...………...34 Figure 4.4 Eutrophication and its effects on water……………………….……………………...35 Figure 4.5 Sleep strategy for a sensor node based on standard deviation...……………………...37 Figure 4.6 Hybrid sleep and wakeup strategy……..………………….………..………………...38 x  Figure 4.7 Fixed duration of a sleep mode…………………..……….………..………………...38 Figure 4.8 Scheduling among grouped sensor nodes…..……………………..…………..……...39 Figure 4.9 Grouped sensor nodes take turns to switch from Active to Sleep mode…..………....40    xi  List of Symbols totalE                              Total power consumption activeE                             Power consumed while the sensor node is active operateE                           Energy consumed for specific operating sensor tancons tE                         Constant energy moduleE                          Energy consumed in the corresponding module sF                                 Adaptive sampling frequency maxF                             Highest frequency in original signal c                                  Parameter used for calculating adaptive sampling frequency ix                                 ith data in a series of water-related data                                  A constant parameter upT                               Upper threshold used for detecting frequency change downT                             Lower threshold used for detecting frequency change                                 User-design parameter cF                               Current maximum frequency Q                                Accumulated rewards S                                 Battery state A                                Agent’s action                                  Learning rate                                  Discount factor xii  low                             Lower threshold for switching sensor node to sleep mode low                            Low threshold for switching sensor node to active mode high                          Higher threshold for switching sensor node to sleep mode high                          Higher threshold for switching sensor node to active mode curr                          Current battery state nS                             State of a sensor node y                              Mean value of samples ( )dS y                       Standard deviation of samples xiii  List of Abbreviations DO                               Dissolved Oxygen ORP                             Oxygen Reduction Potential VLSI                            Very Large Scale Integration CAU                            Central Assessment Unit RF                                Radio Frequency MRE                            Mean Relative Error xiv  Acknowledgements I wish to offer my enduring appreciation and gratitude to my supervisor, Prof. Clarence W. de Silva, who has always guided and encouraged me to continue my work in my research field. I still recall the scene when I first time attended the group meeting, receiving many valuable words of advice and guidance from Prof. de Silva, which today are still steering me through difficult situations, time and time again. It would not have been possible to be where I am today without Prof. de Silva’s help.  Also, I would like to acknowledge and thank the members of my research committee, Prof. Mu Chiao and Prof. Ryozo Nagamune.  This research has been supported by IC-IMPACTS Center of Excellence, and by research grants from the Natural Sciences and Engineering Research Council (NSERC), Canada Foundation for Innovation (CFI), the British Columbia Knowledge Development Fund (BCKDF), and the Tier 1 Canada Research Chair in Mechatronics and Industrial Automation, all held by Prof. C.W. de Silva.  I am grateful to the current and past members of Industrial Automation Lab, and I especially wish to extend my gratitude to Mr. Yunfei Zhang and Mr. Teng Li, who have always inspired me and offered me help both in academic matters and life.  Particularly, my deepest gratitude goes to my parents, who have always been there, loving and supporting me unconditionally throughout my upbringing. xv            I  dedicate  this  thesis  to  my  beloved  parents  and  grandparents  for  their   endless  love and  priceless  sacrifices. This work would not ever have been possible without their support.1  Chapter 1: Introduction Research in the subject of remote environmental monitoring has been rapidly developing in recent years. One cannot overstate the importance of remote environmental monitoring, since it significantly provides convenience and flexibility for the concerned people to observe environmental conditions from distance, thereby reducing the risks, cost, and the required time and improves the accuracy and efficiency. Furthermore, it is then possible to monitor large areas in real time and even foresee some natural disasters such as volcano eruptions, sandstorms, and earthquakes, and take remedial actions ahead of the catastrophe.   The water quality monitoring project, which is carried out in our Industrial Automation Laboratory, focuses on the fact that people in rural or undeveloped areas are at high risk of exposure to water-related disease nowadays. However, the problem is not limited to such regions, and urban areas in industrialized countries may be equally vulnerable. For example, recently, there were two major water-related crises that seriously affected Flint city, Michigan, USA and White Rock, British Columbia, Canada. These problems could have been avoided and corrective actions could have been taken in a timely manner if a reliable, accurate, and distributed water monitoring system was available in the affected areas that could rapidly provide sufficient information about water contamination.  Generally, in an automated water monitoring system, a wireless sensor network is used[1], consisting of many sensor nodes that are capable of sensing data such as pH value, Dissolved Oxygen (DO), Conductivity, Oxidation-Reduction Potential (ORP), turbidity, temperature, and so on. The deployed sensors are typically low-power, low-cost and able to 2  transmit and receive data in the wireless network. Power management is of great significance not only in water quality monitoring but also in other types of remote environmental monitoring because of the importance of wisely managing the system power in order to sustain and extend proper operation of the system as much as possible [2].   Previous work on this subject has primarily focused on power management through energy harvesting [3]. At the sensor level there are methods for including such schemes as duty cycling, data reduction and aggregation, and mobility design for power management. Then at the network level, such methods as power-aware routing [4], dynamic voltage scaling and the use of energy buffer or super-capacitors have been explored.  The present thesis, while using solar energy for energy harvesting, focuses on the sensor nodes for power management, where the power consumption is managed by dynamically changing the sampling rate with the use of reinforcement learning. Furthermore, a method called self-aware scheduling is implemented. Self-aware scheduling contributes to power management by switching nodes between active mode and sleep mode subject to solar and water conditions.   1.1 Background The present project is sponsored by the Canadian network of centres of excellence IC-IMPACTS, which aims to develop an innovative approach that uses advanced sensing and sensory processing, Information and Communication Technologies (ICT), and knowledge-based decision making to deliver water quality monitoring services in a uniform and optimal manner, both temporally and spatially. Use of the ICT platform, it will be possible to automatically and remotely monitor and 3  assess the water quality in time over a targeted geographical area. Rapidly determining the condition of the water enables us to predict the future water qualities and even help the regulatory bodies make decisions or alert the people in the affected area regarding changes to the water quality. This process is summarized in Figure 1.1.   Figure 1.1: Operation of the ICT platform for water quality monitoring.  In recent years, water quality monitoring has been gaining its significance since the quality of water, especially the quality of drinking water, is highly correlated with the public health. The Flint River crisis (Flint, Michigan, USA) which took place in April 2015, is such an issue, as 4  it particularly relates to the contamination of drinking water. This crisis has seriously affected the health of more than 6000 children in the city of Flint. The problem has originated as the interior coating of the aging pipes of the water distribution system were eroded by the corrosive water, and lead in the pipes had leached into the water. This crisis could have been prevented or at least foreseen had the local government conducted corrosion control or used a simple water quality monitoring system that is able to monitor the pH value or the concentration of the negative chloride ions of water, which would make the water corrosive. It is therefore clear that a long-term water quality monitoring system is quite necessary in areas and communities where the water quality is vulnerable to environmental changes or pollution particularly in a postindustrial city like Flint.  1.2 Technological Trends Water quality monitoring has been proposed for various purposes of dealing with water issues. Some researchers are focusing on the design of a sustainable, long-term water quality monitoring system for urban river monitoring [5], wastewater treatment monitoring [6], drinking water monitoring [7]. In that context it is paramount to consider power management as a relevant issue and develop a monitoring system that is able to sustain itself and prolong its life time as much as possible, especially in environmental monitoring of remote areas.  In power management, from the viewpoint of energy source, studies are conducted on selecting the proper form of energy depending on the specific application. Energy forms such as wind, photovoltaic (solar), piezoelectric, and thermal have been adopted in applications that call for a sustainable and renewable energy source, particularly for environmental monitoring. In each 5  type of energy, work has been done on how to acquire it from the particular environment and turn it into useful electrical energy.  However, power management can also be studied through the application of hardware design. Previously, researchers have incorporated a super-capacitor as an energy buffer in hardware design as a way to store extra energy for power management [8]. Such implementations are typically incorporated into very large scale integration (VLSI) hardware. Moreover, algorithms and methodologies that are compatible with the hardware structures have been developed as well. Dynamical voltage scaling, power-aware routing, and duty cycling are typical methods used in energy conservation.   1.3 Research Objective Since water quality monitoring is progressively playing a crucial role in ensuring the quality of public health, a monitoring process that generates real-time data is highly desired. However, when it comes to continuous monitoring in remote areas, addressing the issue of power management is quite pertinent. At the sensor level, there are four main processes of power consumption: data transmission, data reception, idle mode and sleeping mode. The event-driven method of switching the sensor node’s state by turning it on or off will be discussed in detail. In this thesis, focusing on the system level, the power management is done dynamically by using adaptive sampling with reinforcement learning. Also, a method based on sleep/awake cycling is presented. These two methods are proposed in the present work to generally help a remote environmental monitoring system that incorporates energy harvesting, better manage its power consumption.  6  1.4 Thesis Outline This thesis is organized as follows:  In Chapter 2, the overall system architecture is presented, including the framework of the ICT platform. Also, a field testing activity is provided for gaining a preliminary understanding of the water quality monitoring project.  In Chapter 3, the method of adaptive sampling based on reinforcement learning is developed. By changing the sampling frequency during the process of sensory data sampling, an energy-neutral battery state is achieved, which manages to dynamically balance the relationship between consumed energy and harvested energy.  In Chapter 4, the implementation of duty cycling is developed. By selectively turning on and off the sensor node based on the battery state and the sampling task, the goal of conserving energy is achieved.  Chapter 5 concludes the thesis by summarizing the main contributions of the thesis and the suggesting possible future work. 7  Chapter 2: System Architecture This chapter demonstrates the overall architecture of the water quality monitoring system, including the framework of the ICT platform. Also, a field testing activity is described to help gain a practical understanding of the water quality monitoring project.   2.1  Water Quality Monitoring System Figure 2.1 provides a graphical representation of the overall architecture of the water quality monitoring system, including the basic framework of its ICT platform. In a body of water (e.g., river or a creek) where the water quality needs to be monitored, a local sensor network would be implemented, which consists of several nodes that are interconnected through wireless communication. A node will contain multiple sensors such as PH value, ORP, and DO). The data from these sensors are first transmitted to a mother node. Local data processing would be conducted at the mother node, which may include filtering, removal of redundant data, and data representation in a compact form (data compression, modeling, etc.). By means of GPRS/3G/Radio communication, the acquired and pre-processed data can be uploaded into the Central Assessment Unit (CAU) and later analyzed for decision making, and interactively observed by users remotely through mobile terminals. See Figure 2.2.  8   Figure 2.1: The proposed system framework for water quality monitoring.  Essentially, the ICT platform is characterized by capabilities of easy deployment, rapid  relocation, and suitability for long-term sampling needs. Each regular node is capable of working independently or collaboratively with other nodes, as needed. Weather conditions, geographical conditions and wireless accessibility will affect the communication between the mother node and the gateway, and there are alternative methods such as GPRS, 3G and ZigBee for data transmission.   Central Assessment Unit (CAU)· Data Management System· Early Warning System· Decision Support System· Graphical User InterfaceInternetLocal Sensing FrameworkMobile terminal Laptop Desktop WorkstationInternetReservoirRiverLakeWellTowerTowerMother NodeRegular Node(Dynamic Node)- propelled floating node- mobile ground nodeBase StationGPRS/3G/RadioGPRS/3G/RadioGPRS/3G/RadioGPRS/3G/RadioGPRS/3G/RadioGPRS/3G/RadioLocal Sensor NetworkBluetooth/Radio/ZigbeeCreekRiver(Static Node)9   Figure 2.2: The flowchart of processing sensed data.  2.1.1  The Structure of a Regular Node At present, the issue of power management mainly focuses on a regular node, which is powered by its inner battery and energized by solar energy. The analog-to-digital (A/D) converter of the data acquisition hardware is directly linked to the sensor probes, which is responsible for sampling and transmitting the sensory data to a microcontroller for filtering and preliminary data aggregation. Then a radio frequency (RF) communication link sends the processed data to a mother node. See Figure 2.3. 10   Figure 2.3: The structure of a regular node.  2.2 The Designed Sensor Node Prototype This section presents the prototype of a regular node that has been developed in our laboratory and deployed for field testing. A view of the prototype, with its sensors marked, is shown in Figure 2.4.   Figure 2.4: The developed prototype of a regular node. 11   We have designed and built several different types of sensor nodes. After several trials, we made further improvements and finally designed the prototype, which is both water-proof and adequately robust. The five sensors are mounted inside a black plastic box (casing). This arrangement also reduces magnetic interference between probes, while sensing. The solar panel is mounted on the top box for collecting solar energy in order to energize the sensor node. Thus far, we have implemented the sensors for electrical conductivity, ORP, DO, Temperature, and pH, which represent the standard in water quality assessment. Some others in our lab are working on data aggregation, in which several parameters are chosen to be aggregated for evaluating water quality. We plan to purchase a turbidity sensor as well for inclusion in the node for water quality monitoring, which should improve the accuracy and precision of water quality monitoring.   2.2.1 Sensor Node Details Figures 2.5 and 2.6 are based on the reference provided by Libelium.  12   Figure 2.5: The regular node. (Sensor node with hardware from Libelium company) (www.libelium.com)  The designed sensor node allows sensor probes and antennas to be plugged into the sockets around the main body of the sensor node. See figure 2.6.   Figure 2.6: The sockets in the regular node. (Waspmote plug and sense technical guide) 13   There are up to 6 sockets in the sensor node’s main body, which facilitate the sensor node to sample 6 sensor signals simultaneously. A USB socket is provided as well, enabling preprogramming from a PC for various purposes of signal processing.  2.2.2 Levels of Energy Consumption The sensor nodes developed by us have 3 different power modes: The On mode means the sensor node is simply turned on, when the current level is 9 mA. The Sleep mode stands for the state where the sensor node is standing by for interruptions from the timer and the sensors while all the other main programs are terminated. The Hibernate mode indicates that the microcontroller and the modules are completely shut down. Then the current level is 0.7 μA, which is rather negligible. Table 2.2.2 presents the current levels for various modes of operation.  Table 2.1: Mode and its consumption. Mode Current Level On 9 mA Sleep 62 μA Hibernate 0.7 μA   14  All the modes operate using 5 V from the sensor board. Hence the power consumption is proportional to the current level. We use two main modules while the sensor node is turned on, and the current levels of various modules are given in Table 2.2.3.   Table 2.2: Module and its consumption. Module Current Level GPRS 10-400 mA ZigBee 37-64 mA   For different types of sampling, the energy consumption varies as the desired measurement changes. Furthermore, there exists some fixed level of consumption when a sensor is operating. The corresponding current levels of the sensors are given in Table 2.2.4.   Table 2.3: Measurement and its consumption. Sensor Current Level Temperature Sensor 3.5 mA Conductivity Sensor 2.5 mA Dissolved Oxygen Sensor 160 μA pH Sensor 170 μA 15  ORP Sensor 170 μA Constant 1.6 mA  In summary, the total power consumption within a certain period may be expressed as:                  total tan modactive operate cons t uleE E E E E                    (2.2.2)  where totalE  denotes the total power consumption, which is the sum of activeE , the consumption of the node in the On mode, plus operateE , the energy consumed for a specific operating sensor, plus tancons tE , the constant energy consumption. If a specific module is chosen, we should add the energy consumed in the corresponding module, which is denoted by moduleE .  2.3 Field Testing We conducted preliminary field testing for water quality monitoring at the Yosef Wosk Reflecting Pool on the university campus (see Figure 2.8). Currently, we have developed five sensor nodes which could be deployed in a distributed manner for monitoring the water quality within a relatively large area (see Figure 2.7). The landscape of the area of the monitored pool is shown in Figure 2.8. In the future we will develop and implement dynamic nodes, which are able to propel to a desired location according a command or autonomously according to a program function. 16   Figure 2.7: The deployment of sensor nodes.   Figure 2.8: Landscape of the Yosef Wosk reflecting pool. 17  Chapter 3: Adaptive Sampling In this chapter, to address the power management issue, the method of adaptive sampling based on reinforcement learning is presented. By changing the sampling frequency during the sampling process, an energy-neutral battery state is achieved, which manages to dynamically balance the relationship between consumed energy and harvested energy. Implementation of this method is described and some results are discussed.  3.1 Energy-neutral Battery State In a traditional battery-powered system, the main goal of power management is either to minimize the energy consumed in various operating process and/or to maximize the lifetime of the system itself while satisfying the applicable performance constraints. However, in practical environmental monitoring with energy harvesting, a battery state is preferred where the harvested energy is consumed at a proper rate compared to the remaining battery power level, so that the system can work perennially. Hence, this battery state is called energy-neutral: a node with energy harvesting is capable of keeping its battery state in a proper range while a desired performance level is supported indefinitely [9]. In the present case, the battery state should not be too low or too high since a sensor node working at a low battery state will undermine the data accuracy [10] and an over-charge problem might occur at an extremely high level of battery state. So generally, keeping the battery state at an energy-neutral state is the main goal of implementing adaptive sampling, in the present work.  18  3.2 Adaptive Sampling The use of adaptive sampling is not new. However, adaptive sampling is desirable for power management in environmental monitoring where the monitored parameters remain relatively stable. This is because the sampling rate can be adjusted according to the importance of parameters, accuracy requirements and battery storage state. Since a higher sampling frequency would increase the sampling accuracy but decrease the battery sate, and a lower sampling frequency would result in a higher battery state with energy harvesting but a undermined sampling accuracy, in this case, a trade-off (i.e., striking a balance between accuracy and battery state) is generally achieved between sampling accuracy and energy consumption while the system is reaching an energy-neural battery state.     Figure 3.2: How sampling affects battery state and accuracy.  This also means that if the battery state is low and some parameter; for example, the PH value of water hardly fluctuates and remains steady within a duty cycle, the sampling frequency may be lowered and even the sensor node may be turned off (i.e., sampling frequency = 0) until the next duty cycle in order to save energy. The same reasoning applies when the battery state is high with adequate energy constantly provided by the solar panel. Then the sampling frequency may be increased to achieve a higher sampling accuracy and avoid the over-charging issue.    Accuracy    Battery State Higher Sampling Frequency Lower Sampling Frequency   Battery State   Accuracy 19  3.2.1 Nyquist-Shannon Sampling Theorem According to the Nyquist-Shannon sampling theorem, the minimum sampling frequency sF , which could guarantee the signal reconstruction, must be at least twice the highest frequency of a given signal. If this highest frequency is denoted as maxF , generally, it is advised to have sF  three times greater than maxF , i.e., sF =c  maxF , where c  3 [11].  For example, consider the set of DO data obtained from National Water Information System (http://waterwatch.usgs.gov/wqwatch/?pcode=00300) in 10 hours continuously, as shown in Figure 3.3. Its frequency response function can be determined through Fourier transformation, as shown in Figure 3.4.   Figure 3.3: Original signal of DO data.  (http://waterwatch.usgs.gov/wqwatch/?pcode=00300) 20   Figure 3.4: Frequency response obtained through Fourier transform.  Using the frequency response, it is easy to determine the highest frequency of significance maxF  of the given signal, which then could be used as the basis for seeking a suitable sampling frequency. After conducting a number of Fourier transforms, it was concluded that any signal whose amplitude is below 50 was noise. Hence for this frequency response, a suitable estimate for maxF  is 0.75-310 Hz, as shown in Figure 3.5.  21    Figure 3.5: Determining the highest frequency of interest.  Particularly, as mentioned above, when the value of parameter c increases, a higher sampling frequency would assure a higher accuracy in reconstructed signal. But as a result, more energy would consequently be consumed during the data acquisition process. In environmental monitoring, since there exists uncertainty in the real signal of monitored data, it is proposed here to dynamically find a tradeoff between the accuracy of the reconstructed signal and the energy consumption, which depends on adaptive sampling. More details on this issue will be presented later in the thesis.  22  3.2.2 Mean Relative Error While dynamically changing the sampling frequency during water quality monitoring, to assure a satisfactory sampling frequency, Mean Relative Error (MRE) is used to constrain the sampling frequency. This is also called a process of tradeoff between power consumption and reconstructed signal’s accuracy. MRE is defined as:  1ˆ1 N i ii ix xMREN x                                            (3.2.2)  where ix  denotes the ith data value in a series of sensory data, say, the pH value. Then ix   represents the ith data value in the sequence of the reconstructed signal. Since the value of Temperature and ORP would possibly be 0 in the actual field testing, a constant parameter   is included in the denominator. However, for pH value, Conductivity and Dissolved Oxygen, there would rarely be 0 in the data values, 0   is used while calculating the MRE for pH value, Conductivity and DO value. A simple example is presented in Figure 3.6. The 5th data value in the original signal is actually the 3rd data value in the reconstructed signal, and the difference between it and the 3rd data value in the original sequence of data is then expressed as 3 3x x .     23   Figure 3.6: The process of calculating MRE.  3.2.3 Detection of Minimum Frequency Change Since a sequence of data will be sampled continuously during water quality monitoring, the highest frequency maxF  is subject to change over time. So, an algorithm is developed, which is able to detect the minimum frequency change and possibly adjust the value of maxF . Here two thresholds, upT  and downT , are defined, which can be used to detect a positive frequency change and a negative frequency change, separately. Also, a user-design parameter  , which stands for the minimum detectable frequency change, is introduced. This means, for example, if 2%  , then a frequency change more than 2% would be detected.[12]  If a boundary for the maximum positive frequency change is denoted as upF , and for the maximum negative frequency change as downF , the following equations may be written:  24                                                               maxupF F                                                                (3.2.3)                                                              maxdownF F                                                             (3.2.4)  Then the thresholds for the detectable frequency change can be written as:                                         maxmax(1 )2 2upupF FT F                                            (3.2.5)                                       max max(1 )2 2downdownF FT F                                          (3.2.6)  For an n number of consecutive samples, a frequency change can be easily illustrated in Figure 3.7.    Figure 3.7: Detection of a change in the maximum frequency  25  Basically, as several consecutive samples are acquired, if the current maximum frequency is denoted as cF  , then the following algorithm can be stated: Step 1: check if maxc up cF F F F    or maxc down cF F F F   , Step 2: if no, continue to sample based on the same maximum sampling frequency.             If yes, change the maximum sampling frequency. We have:              max cF F               maxsF c F                max(1 )upF F                 max(1 )downF F                 Then jump to step 3. Step 3: Start to sample based on a new maximum sampling frequency.  3.3 Reinforcement Learning Reinforcement Learning (RL) is a branch of machine learning, which is capable of modeling the real nature of learning. In RL, there is an agent that will find the optimal actions to execute by maximizing the rewards obtained through actions. More specifically, it is a process during which the agent interacts with its presented environment and as a feedback, a positive or negative reward is given upon to the agent’s action, which helps the agent to update its state and correspondingly decide what action to be taken subsequently. This process is represented in Figure 3.8.  26   Figure 3.8: The interaction between an agent and its environment.   Generally, in an unknown environment, the agent alone is unable to decide which action is the most appropriate one in the early phase of learning. In such situations, an exploration strategy is adopted, in which the action conducted by the agent with the highest reward is perceived for different states. This learning process, however, is often represented by an algorithm called Q-learning, which is used to accumulate rewards and therefore determine the best policy. This Q-learning algorithm is generally expressed as:                              1 1( , ) ( , ) ( max ( , ) ( , ))t t t t t t t tQ S A Q S A R Q S A Q S A                              (3.3) where Q represents the reward accumulated at state S along with an action A taken at time t, while   denotes the learning rate, which quantifies the agent’s learning pace, and   is termed the discount factor whose value falls between 0 and 1. Furthermore, R is called an immediate reward after taking certain action to transit from state tS  to the next state 1tS  . In general, the value of   plays a key role in controlling how fast the agent’s learning process would converge. Empirically, the value of   should decrease as the agent progressively learns, in order to achieve the best converging result. Moreover, the discount factor decides to what degree the future reward would Agent EnvironmentActionRewardState27  influence the current accumulative reward. More specifically, the closer the   is to 1, the more profound the influence of future rewards. To summarize, equation (3.3) explicitly conveys the idea of how to connect current and future cumulative rewards and as well, how the cumulative rewards update themselves after many rounds of trials.  3.3.1 Rewarding rules for adaptive sampling As it has been discussed before, the minimum sampling frequency sF  which can be used for signal reconstruction is decided by the value of parameter c, and based on this assumption, comprehending the scheme of rewarding rules in adjusting the value of c is quite necessary. The rewarding rules go as follows: Based on reinforcement learning, the S here denotes the battery state, and the Action that has to be taken during the sampling process is to increase or decrease the value of c. The corresponding rewards are -1, 0, +1 based on some action and the battery state. From Table 3.3.1, it is easy to conclude that, on the one hand, when the battery state is low, the system is encouraged to work at a relatively low sampling frequency in order to conserve energy as much as possible, waiting the battery to be gradually charged through solar energy. On the other hand, when the battery state is high, it is encouraged for the system to work at a high sampling frequency, where the value of c is between 7.0 and 8.0. This would assure a higher sampling frequency along with a higher energy consumption, and consequently, the over-charging issue might be avoided as well. And as a result of long-term monitoring, if there is no hardware failure, the system will eventually operate in a healthy condition where the battery state is between 40%-70% regardless the initial battery condition.  28  Table 3.1: The rewarding rules.                      Value of c Battery State (2.0 - 4.0] (4.0 - 7.0] (7.0-8.0] <40% 1 0 -1 40%-75% -1 1 -1 >75% -1 0 1  3.4 Simulation Example To summarize the steps implementing adaptive sampling based on reinforcement learning, the flowchart shown in Figure 3.9 may be used.          Figure 3.9: A flowchart for the algorithm. 29   Provided by [13], each sensor node in our laboratory has a battery capacity of 6600 mAh, while USB charging provides 5 V – 100 mA and the solar panel load is 9 V – 280  mA. In the simulation, the sensor node operates in the GPRS module, whose energy (current) consumption ranges from 10 mA – 400 mA depending on the sampling frequency. More specifically, for the parameter c whose value falls within (2,4], the energy (current) consumption is around 70 mA, while between (4,7] and (7,8] the energy (current) consumption is about 240 mA and 450 mA, respectively. Set MRE < 0.13, η = 0.6, γ = 0.8, c = 2.5. Then test the proposed algorithm with 1200 data samples of Dissolved Oxygen within 10 hours in 3 cases: battery storage initial state of 30%, 50%, 75%.     Figure 3.10: Simulation with 1200 data samples.  30   3.5 Conclusion Despite the difference in the initial battery state, with reinforcement learning, the battery state eventually converged to around 61% of the total battery state. This verifies that the proposed algorithm works satisfactorily for power management, making sure that the sensor node works at an energy-neural level. Further conclusions are provided in Chapter 5. 31  Chapter 4: Self-aware Scheduling In this chapter, the implementation of duty cycling is presented. In this method, by selectively turning on and turning off the sensor node depending on the battery state and sampling task, the goal of conserving energy is achieved.  4.1 Self-aware Scheduling of Sleep-Awake Strategy The scheme of self-aware scheduling aims to selectively sleep or wakeup the sensor nodes according to the needs of tasks, the status of other nodes, as well as the battery state of itself. In water quality monitoring, the self-aware scheduling can be implemented at two levels: 1. At the network level, in a distributed sensor network, optimization of power conservation can be achieved by globally evaluating the water quality in different areas and thus turning off the sensor nodes in the area where a healthy water condition is known to exist for an extended , and the sleep/awake cycle can thus be adjusted depending on the task needs and water qualities; 2. At a local level, it is feasible to apply self-aware scheduling into single nodes. Based on the battery state and the task needs, a single node itself can correspondingly choose a proper sleep/awake cycle with the objective of energy conservation.   4.2 Related Work By far, the scheme of sleep and awake scheduling has been studied and implemented by many researchers. The work in [14] proposed a queuing model, combined with the Nash solution, to efficiently calculate the optimal parameters for a sleep and wakeup strategy;. The work in [15] experimentally studied the behavior of sleep and wakeup scheduling on 3 different network topologies, and proposed to use reinforcement learning to schedule the sleep and wakeup cycle. 32  The work in [16] systematically analyzed the scheme of sleep and wakeup cycling, and proposed to wake up the nodes depending on the paging demands. The proposed solutions in these contributions are in essence capable of coordinating with the system demands despite their different research directions. On the basis of the previous work, the power management issue could also be studied by designing sleep and wakeup strategies in automated water quality monitoring, which has advantages.  4.3 Sleep-Awake Strategies at Single Nodes Level As it has been discussed before, the sleep and wakeup strategies can be implemented either at the single nodes level or at the distributed network level. Since it is much easier to explore the strategies that are implemented at the singe node level, that subject is investigated first. Implementation at the distributed network is discussed subsequently. It is known that a too low battery state may possibly result in data inaccuracy while a too high battery state with continuous energy harvesting may lead to problems of overcharging. Hence, it is necessary to specify thresholds for the battery to change its state from sleep to wakeup or from wakeup to sleep, according to the specific battery state. Additionally, in view of the characteristics of water sensory data, including pH value, DO, Temperature, Conductivity and ORP, fluctuations in these parameters always imply activities that affect water quality such as effluent contamination, acid rain, and lead leaching. Such changes are in fact what we are most concerned about in water quality monitoring. Then if all the values of water quality-related parameters are stable within the time period of interest, it is believed that no significant changes are happening, and some individual sensor nodes may turned off temporarily or periodically. 33   4.3.1 Battery State-Aware Sleep and Awake Strategy During the sampling process, consider a situation where the solar power is limited as an energy source and the remaining battery power is too low. Then if the battery of a node finally depletes it becomes quite necessary to switch the node into the sleep mode. Thus, if the battery state is below the threshold low , then some of the sensors by themselves may go into the sleep mode, waiting for the battery to be recharged until the remaining battery capacity is above the threshold low , when it will wake up and work. If the state of the sensor node is denoted as nS , which indicates whether it is sleeping or awake, and the current battery state is denoted as curr , then the overall process may be expressed as:                                                lo1,0,curr wncurr lowS                                               (4.3.1)  where the value of 1 means the sensor turns into sleep mode from active mode, while the value of 0 means the sensor wakes up from sleep mode. Also, the value of low  should be somewhat greater than low , which assures that the sensor will not again fall into sleep mode right after waking up. The same reasoning applies to the situation where the battery state is originally moderate. Then, if the battery turns sensors into sleep mode for a while in response to task needs (which will be discussed below), and the stored energy is about to exceed a certain threshold, then it is quite necessary for the sensor to wake up and start sampling, which makes sure the extra stored energy will be consumed in case of over-charging. In this case, if the total stored energy exceeds the 34  threshold high , the sensor will turn into active mode from sleep mode. Subsequently, if the task needs governed by the environmental changes are minor, the sensor can possibly fall into sleep mode again after the stored energy is below the threshold high . That is:                              1,0,curr highncurr highS                                         (4.3.2) Thus, the whole sensor node is aware of its own battery state and will be capable of wisely choosing when to sleep and wakeup certain sensor based on these thresholds (See Figure 4.3).    Figure 4.3: Thresholds for battery state-aware cycle scheduling.  35  4.3.2 Sleep-Awake Strategy Based on Data Standard Deviation If the water quality remains steady for a relatively long period of time, the parameters of water quality would remain steady as well. However, if during the period of interest, the value some parameters changes dramatically, it may imply that the water is experiencing some form of contamination. For example (see Figure 4.4) eutrophication in water will generally cause a sharp decrease of DO and a dramatic increase of nitrates. As eutrophication has an adverse effect on biodiversity, it would be desirable to carefully observe the value of DO and nitrates in water, carry out some analysis, and make appropriate decisions based on the findings.    Figure 4.4: Eutrophication and its effects on water. (http://toxics.usgs.gov/definitions/eutrophication.html)  36  In this backdrop, based on the trends of water quality parameters (e.g., whether they increase, decrease of remain steady), the standard deviation of the sensory data may be used to evaluate the stability of a parameter value that is capable of directly representing the water quality. If the standard deviation is fairly small in a situation where, originally, the water was considered to be of good quality, then it can be assumed that the water quality is currently acceptable and some of the sensors in the monitored area may be put to sleep for a predetermined period, to save energy. They may be awaken after this period or by another agent based on another finding related to water quality in the monitored area. Here, an appropriate scheme:                                     1 2[ , , ] 1,2,350n n n my y y y y nm n                             (4.3.2)                                      12( ) ( ) 1,2,3( ) ( )idD y E y y iS y D y                                                   (4.3.3)  where y is a set of water quality-related data acquired in a time sequence, y represents the mean value of 50 samples, which may change over time, and ( )dS y  represents the standard deviation, which triggers turning off of certain sensor as soon as its value becomes smaller than a defined threshold  . In this situation, the task needs are assumed to be satisfied. Figure 4.5 provides an example.  Here some of the sensors are switched to sleep mode when the value of parameter remains stable, since the value of the standard derivation is steady and sufficiently small.  37    Figure 4.5: Sleep strategy for a sensor node based on standard deviation.   4.3.3 Hybrid Sleep-Awake Strategy The hybrid sleep and wakeup strategy is actually a combination of the two strategies presented before (see Figure 4.6). The ideal battery state for a sensor node under normal operation is between the thresholds high  and low . 38    Figure 4.6: Hybrid sleep and wakeup strategy.  To ensure a healthy battery state, unless the battery state is so high that the sensors should be awaken, a fixed time must be set to awaken the sensors in a scenario where the task needs are currently satisfied (see Figure 4.7).   Figure 4.7: Fixed duration of a sleep mode. 39  4.4 Sleep-Awake Strategies at Distributed Sensor Network Level Implementing the strategies of sleep/awake cycle at the level of distributed sensor network has a rather significant meaning and implication in power management since it focuses on the global issue of energy efficiency and sustainability. The key to designing a proper scheme for this is to coordinate the work loads with the surrounding environment. The information of the surrounding environment, even in the present application, is not limited to water quality in certain testing area, but also the battery state of other sensor nodes in a group. An example is shown in Figure 4.8.    Figure 4.8: Scheduling among grouped sensor nodes.  Suppose that there are three sensor node groups in area A, B, and C in a practical field testing implementation, with each group having 4 sensor nodes. Also, suppose that the water flows 40  from left to right, and the water quality is evaluated to be good in area A after a relatively long period of monitoring. Then from the point of saving energy, it makes sense to turn off some of the sensors in area A, and lower the sampling frequency of the sensors in areas B and C, as discussed further in Chapter 3. Also, after establishing that the water quality is good over a period of time, it is also possible do arrange a duty scheduling among all the three groups of sensor nodes; for example, allow them to take turns to sleep and also make sure that there is always a group of sensor nodes that remain active for detecting potential change in water quality (see Figure 4.9).   Figure 4.9: Grouped sensor nodes take turns to switch from Active to Sleep mode.  Additionally, waking up a sensor node from sleep mode is of similar significance as turning it off. A simple strategy is to turn it on after a fixed time. But in real-time monitoring, a scheme of event detection is more suitable in determining when to wake up a sensor node. For instance, in the example of Figure 4.8, while the other two grouped sensor nodes are sleeping during duty scheduling, the awake group of sensor nodes may detect some significant changes in water quality-related parameters, signaling a possible contamination of the water. Then the awake 41  group can directly send messages to the other groups and wake them up. Later on, the three grouped sensor nodes can start to work and evaluate how severe the contamination is and what is the source of contamination.  4.5 Conclusion This chapter explored the possibility of turning off some sensor nodes to conserve energy. Several strategies were proposed in sufficient detail and rationale for sensor nodes to duty cycle between active mode and sleep mode. Those strategies, implemented at two different levels of a sensor network, offer alternative solutions for addressing potential energy-efficiency problems.       42  Chapter 5: Conclusion This chapter concludes the thesis. It summarizes the key contributions of the thesis and indicates some possible future work.  5.1 Developed Methods This thesis provided three solutions for power management in water quality monitoring, which are mainly focused on the battery-state awareness. Even though the specific methods and algorithms were specifically developed for applications of water quality monitoring, the developed methodologies of power management can be extended into other applications, for long-term monitoring. Also, these solutions may be applied for a variety of energy sources such as solar, wind, thermal, and piezoelectric. Furthermore, the characteristics of the data that would be acquired should also be considered before choosing a method for power management.  Some advantages and disadvantages of the methodologies developed in this thesis are summarized now. For the adaptive sampling based on reinforcement learning, the sensor node is capable of dynamically changing the sampling frequency by itself while ensuring a high sampling accuracy. This method is quite suitable for sampling those parameters that require a relatively high sampling frequency. Nevertheless, in continuous sampling the reliability of the solar energy method is higher compared to self-aware sleep/awake scheduling. In a nutshell, this method fits the sampling situations where continuous sampling is needed, especially when the water quality is vulnerable and prone to contamination.  43  For self-aware scheduling, the main advantage is that the sensor nodes typically do not rely on solar energy, compared to adaptive sampling. However, if some abrupt changes in water quality emerge right after the sensor nodes are turned off, the increase/decrease in the value of a certain parameter might not be detected sufficiently fast. To conclude, this methodology is appropriate for monitoring situations where the water quality remains stable and a relatively low sampling frequency is preferred.  5.2 Future Work Planned future work includes purchase some hardware for implementing the methods developed in this thesis. Also, specific algorithms for the methodology of self-aware scheduling will be developed. After setting up and improving the ICT platform, the equipment will be deployed for remote monitoring, in a practical setting possibly in a water-critical area. This will represent an important contribution particularly for people in rural area, who will benefit from an effective water quality monitoring system to ensure the availability of safe drinking water.  An effective and more realistic enhancement to the existing Water Quality Index (WQI) is being developed. This can be used to more accurately represent water quality, which is computed using water-related parameters. It is viable to consider local data fusion and aggregation for power management. What is more, the network routing and dynamic voltage scaling are two possible research directions, since these two topics are widely studied in the field of power management.  44  Bibliography [1] Zhu W, Xiao-qiang H A O, De-bao W E I. Remote Water Quality Monitoring System Based on WSN and GPRS [J][J]. Instrument Technique and Sensor, 2010, 1: 018. [2] De Silva, C.W., Sensors and Actuators—Engineering System Instrumentation, 2nd Edition. CRC Press/Taylor&Francis, Boca Raton, FL. 2016.  [3] Castagnetti A, Pegatoquet A, Belleudy C, et al. A framework for modeling and simulating energy harvesting WSN nodes with efficient power management policies[J]. EURASIP Journal on Embedded Systems, 2012, 2012(1): 1-20. [4] Chipara O, He Z, Xing G, et al. Real-time power-aware routing in sensor networks [C]//Quality of Service, 2006. IWQoS 2006. 14th IEEE International Workshop on. IEEE, 2006: 83-92.  [5] Hur J, Hwang S J, Shin J K. Using synchronous fluorescence technique as a water quality monitoring tool for an urban river[J]. Water, air, and soil pollution, 2Garcı́a H L, González I M. Self-organizing map and clustering for wastewater treatment monitoring[J]. Engineering Applications of Artificial Intelligence, 2004, 17(3): 215-225.008, 191(1-4): 231-243. [6] Storey M V, van der Gaag B, Burns B P. Advances in on-line drinking water quality monitoring and early warning systems[J]. Water Research, 2011, 45(2): 741-747. [7] Gomez Cid-Fuentes R, Cabellos-Aparicio A, Alarcon E. Energy buffer dimensioning through energy-erlangs in spatio-temporal-correlated energy-harvesting-enabled wireless sensor networks[J]. Emerging and Selected Topics in Circuits and Systems, IEEE Journal on, 2014, 4(3): 301-312. 45  [8] Van Hoesel L F W, Havinga P J M. A lightweight medium access protocol (LMAC) for wireless sensor networks: Reducing preamble transmissions and transceiver state switches[J]. 2004. [9] Alippi C, Anastasi G, Galperti C, et al. Adaptive sampling for energy conservation in wireless sensor networks for snow monitoring applications[C]//Mobile Adhoc and Sensor Systems, 2007. MASS 2007. IEEE International Conference on. IEEE, 2007: 1-6. [10] Waspmote plug and sense technical guide [11] Niyato D, Hossain E, Fallahi A. Sleep and wakeup strategies in solar-powered wireless sensor/mesh networks: Performance analysis and optimization[J]. Mobile Computing, IEEE Transactions on, 2007, 6(2): 221-236. [12] Mihaylov M, Le Borgne Y A, Tuyls K, et al. Reinforcement learning for self-organizing wake-up scheduling in wireless sensor networks[M]//Agents and artificial intelligence. Springer Berlin Heidelberg, 2011: 382-396. [13] Schurgers C. Wakeup strategies in wireless sensor networks[M]//Wireless Sensor Networks and Applications. Springer US, 2008: 195-217. [14] Alippi C, Anastasi G, Francesco M D, et al. An adaptive sampling algorithm for effective energy management in wireless sensor networks with energy-hungry sensors[J]. Instrumentation and Measurement, IEEE Transactions on, 2010, 59(2): 335-344. [15] Yang W, Nan J, Sun D. An online water quality monitoring and management system developed for the Liming River basin in Daqing, China[J]. Journal of environmental management, 2008, 88(2): 318-325. 46  [16] Shen H, Tan Y, Lu J, et al. Achieving autonomous power management using reinforcement learning[J]. ACM Transactions on Design Automation of Electronic Systems (TODAES), 2013, 18(2): 24. [17] Yau K L A, Goh H G, Chieng D, et al. Application of reinforcement learning to wireless sensor networks: models and algorithms[J]. Computing, 2015, 97(11): 1045-1075. [18] Tan Y, Liu W, Qiu Q. Adaptive power management using reinforcement learning[C]//Proceedings of the 2009 International Conference on Computer-Aided Design. ACM, 2009: 461-467. [19] Rucco L, Bonarini A, Brandolese C, et al. A bird's eye view on reinforcement learning approaches for power management in WSNs[C]//Wireless and Mobile Networking Conference (WMNC), 2013 6th Joint IFIP. IEEE, 2013: 1-8. [20] Shah K, Kumar M. Distributed independent reinforcement learning (DIRL) approach to resource management in wireless sensor networks[C]//Mobile Adhoc and Sensor Systems, 2007. MASS 2007. IEEE International Conference on. IEEE, 2007: 1-9. [21] Sutton R S, Barto A G. Reinforcement learning: An introduction[M]. MIT press, 1998. [22] Wang P, Wang T. Adaptive routing for sensor networks using reinforcement learning[C]//Computer and Information Technology, 2006. CIT'06. The Sixth IEEE International Conference on. IEEE, 2006: 219-219. [23] Yau K L A, Komisarczuk P, Teal P D. Reinforcement learning for context awareness and intelligence in wireless networks: review, new features and open issues[J]. Journal of Network and Computer Applications, 2012, 35(1): 253-267. 47  [24] Salvadori F, De Campos M, Sausen P S, et al. Monitoring in industrial systems using wireless sensor network with dynamic power management[J]. Instrumentation and Measurement, IEEE Transactions on, 2009, 58(9): 3104-3111. [25] Wang Q, Yang W. Energy consumption model for power management in wireless sensor networks[C]//Sensor, Mesh and Ad Hoc Communications and Networks, 2007. SECON'07. 4th Annual IEEE Communications Society Conference on. IEEE, 2007: 142-151. [26] Suryadevara N K, Mukhopadhyay S C, Kelly S D T, et al. WSN-based smart sensors and actuator for power management in intelligent buildings[J]. Mechatronics, IEEE/ASME Transactions on, 2015, 20(2): 564-571. [27] Seah W K G, Eu Z A, Tan H P. Wireless sensor networks powered by ambient energy harvesting (WSN-HEAP)-Survey and challenges[C]//Wireless Communication, Vehicular Technology, Information Theory and Aerospace & Electronic Systems Technology, 2009. Wireless VITAE 2009. 1st International Conference on. Ieee, 2009: 1-5. [28] Popovici E, Magno M, Marinkovic S. Power management techniques for wireless sensor networks: a review[C]//Advances in Sensors and Interfaces (IWASI), 2013 5th IEEE International Workshop on. IEEE, 2013: 194-198. [29] Castagnetti A, Pegatoquet A, Belleudy C, et al. A framework for modeling and simulating energy harvesting WSN nodes with efficient power management policies[J]. EURASIP Journal on Embedded Systems, 2012, 2012(1): 1-20. 48  [30] Sausen P S, Spohn M A, Perkusich A. Broadcast routing in wireless sensor networks with dynamic power management and multi-coverage backbones[J]. Information Sciences, 2010, 180(5): 653-663. [31] Stojčev M K, Kosanović M R, Golubović L A R. Power management and energy harvesting techniques for wireless sensor nodes[C]//Telecommunication in Modern Satellite, Cable, and Broadcasting Services, 2009. TELSIKS'09. 9th International Conference on. IEEE, 2009: 65-72. [32] Shao H, Tsui C Y, Ki W H. The design of a micro power management system for applications using photovoltaic cells with the maximum output power control[J]. Very Large Scale Integration (VLSI) Systems, IEEE Transactions on, 2009, 17(8): 1138-1142. [33] Alippi C, Camplani R, Galperti C, et al. A robust, adaptive, solar-powered WSN framework for aquatic environmental monitoring[J]. Sensors Journal, IEEE, 2011, 11(1): 45-55. [34] Passos R M, Coelho Jr C J N, Loureiro A A F, et al. Dynamic power management in wireless sensor networks: An application-driven approach[C]//Wireless On-demand Network Systems and Services, 2005. WONS 2005. Second Annual Conference on. IEEE, 2005: 109-118. [35] Gormus S, Kulkarni P, Fan Z. The power of networking: How networking can help power management[C]//Smart Grid Communications (SmartGridComm), 2010 First IEEE International Conference on. IEEE, 2010: 561-565. [36] Christmann J F, Beigné E, Condemine C, et al. Energy harvesting and power management for autonomous sensor nodes[C]//Design Automation Conference (DAC), 2012 49th ACM/EDAC/IEEE. IEEE, 2012: 1049-1054. 49  [37] Liu S, Lu J, Wu Q, et al. Harvesting-aware power management for real-time systems with renewable energy[J]. Very Large Scale Integration (VLSI) Systems, IEEE Transactions on, 2012, 20(8): 1473-1486. [38] Cheour R, Lahmar K, Abid M. Evolution of wireless sensor networks and necessity of power management technique[C]//Faible Tension Faible Consommation (FTFC), 2011. IEEE, 2011: 75-78. [39] Sausen P S, Sousa J R B, Spohn M A, et al. Dynamic power management with scheduled switching modes in wireless sensor networks[C]//Modeling, Analysis, and Simulation of Computer and Telecommunication Systems, 2007. MASCOTS'07. 15th International Symposium on. IEEE, 2007: 1-8. [40] Levron Y, Shmilovitz D, Martínez-Salamero L. A power management strategy for minimization of energy storage reservoirs in wireless systems with energy harvesting[J]. Circuits and Systems I: Regular Papers, IEEE Transactions on, 2011, 58(3): 633-643. [41] Cui X, Zhang X, Shang Y. Energy-saving strategies of wireless sensor networks[C]//2007 International Symposium on Microwave, Antenna, Propagation and EMC Technologies for Wireless Communications. 2007. [42] Smith R A, Schwarz G E, Alexander R B. Regional interpretation of water‐quality monitoring data[J]. Water resources research, 1997, 33(12): 2781-2798. [43] Loaiciga H A, Charbeneau R J, Everett L G, et al. Review of ground-water quality monitoring network design[J]. Journal of Hydraulic Engineering, 1992, 118(1): 11-37. [44] Baker A. Fluorescence properties of some farm wastes: implications for water quality monitoring[J]. Water Research, 2002, 36(1): 189-195. 50  [45] Strobl R O, Robillard P D. Network design for water quality monitoring of surface freshwaters: A review[J]. Journal of environmental management, 2008, 87(4): 639-648. [46] O'Flyrm B, Martinez R, Cleary J, et al. SmartCoast: a wireless sensor network for water quality monitoring[C]//Local Computer Networks, 2007. LCN 2007. 32nd IEEE Conference on. Ieee, 2007: 815-816. [47] Harmancioglu N B, Alpaslan N. WATER QUALITY MONITORING NETWORK DESIGN: A PROBLEM OF MULTI‐OBJECTIVE DECISION MAKING1[J]. JAWRA Journal of the American Water Resources Association, 1992, 28(1): 179-192. [48] Storey M V, van der Gaag B, Burns B P. Advances in on-line drinking water quality monitoring and early warning systems[J]. Water Research, 2011, 45(2): 741-747. [49] Le Dinh T, Hu W, Sikka P, et al. Design and deployment of a remote robust sensor network: Experiences from an outdoor water quality monitoring network[C]//Local Computer Networks, 2007. LCN 2007. 32nd IEEE Conference on. IEEE, 2007: 799-806. [50] Charef A, Ghauch A, Baussand P, et al. Water quality monitoring using a smart sensing system[J]. Measurement, 2000, 28(3): 219-224. [51] Rasin Z, Abdullah M R. Water quality monitoring system using zigbee based wireless sensor network[J]. International Journal of Engineering & Technology IJET Vol: 9 No: 10, 2009, 9(10): 24-28. [52] Sanders T G, Adrian D D. Sampling frequency for river quality monitoring[J]. Water Resources Research, 1978, 14(4): 569-576. [53] Dahlgren R, Singer M, Allen-Diaz B, et al. Timing, frequency of sampling affect accuracy of water-quality monitoring[J]. California Agriculture, 1999, 53(6): 44-48. 51  [54] Telci I T, Nam K, Guan J, et al. Optimal water quality monitoring network design for river systems[J]. Journal of environmental management, 2009, 90(10): 2987-2998. [55] Strobl R O, Robillard P D, Shannon R D, et al. A water quality monitoring network design methodology for the selection of critical sampling points: Part I[J]. Environmental Monitoring and Assessment, 2006, 112(1-3): 137-158. [56] Lo S L, Kuo J T, Wang S M. Water quality monitoring network design of Keelung River, Northern Taiwan[J]. Water Science and Technology, 1996, 34(12): 49-57. 

Cite

Citation Scheme:

        

Citations by CSL (citeproc-js)

Usage Statistics

Share

Embed

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"
                            src="{[{embed.src}]}"
                            data-item="{[{embed.item}]}"
                            data-collection="{[{embed.collection}]}"
                            data-metadata="{[{embed.showMetadata}]}"
                            data-width="{[{embed.width}]}"
                            async >
                            </script>
                            </div>
                        
                    
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:
http://iiif.library.ubc.ca/presentation/dsp.24.1-0300325/manifest

Comment

Related Items