Load Shifting of Electric Water Heaters through Priority-Based Control and Machine Learning Jackson Wilt (University of Tennessee, Knoxville, Tennessee 37916) Mentor : Jeffrey Munk (Oak Ridge National Laboratory, Oak Ridge, TN 37830) ________________________________________ The electricity supply from the national or regional grid must always match the electricity demand to maintain proper frequency and voltage levels. The fluctuating levels of daily electricity have posed significant challenges for energy production. In the U.S., residential energy consumption accounts for much of the total electricity used and contributes a significant amount to the irregularity of electricity load on the national grid. Water heating alone accounts for ~15-20% of the energy used in residential energy consumption. Here, we consider the feasibility of using residential electric water heaters to buffer peak load and provide peak load shifting. Electric water heater load shifting pertains to demand-side management which is attractive due to low first-costs and reduced transmission losses when compared to supply-side management or conventional electrochemical battery storage. We first evaluated the performance of a Priority Based Control system to limit power consumption of a group of water heaters. We simulated the performance of 25 water heaters using measured hot water use data from occupied homes, and we found the maximum peak power reduction. We calculated the reduction through logical operators and temperature states, which successfully optimized the heating cycles for the targeted power. Next, we used machine learning toolkits to determine the hot water draw patterns in a residential setting. Priority Based Control of the water heaters provided peak reduction up to 66%, and Recurrent Neural Networks showed the greatest time series predictivity and scalability for larger datasets. The results from this study show great potential for using predictive algorithms in electric water heaters to further strengthen national energy grid sustainability.
jwilt2 / hierarchy_optimization_wh Goto Github PK
View Code? Open in Web Editor NEWLoad Shifting of Electric Water Heaters through Priority-Based Control and Machine Learning Jackson Wilt (University of Tennessee, Knoxville, Tennessee 37916) Mentor : Jeffrey Munk (Oak Ridge National Laboratory, Oak Ridge, TN 37830) ________________________________________ The electricity supply from the national or regional grid must always match the electricity demand to maintain proper frequency and voltage levels. The fluctuating levels of daily electricity have posed significant challenges for energy production. In the U.S., residential energy consumption accounts for much of the total electricity used and contributes a significant amount to the irregularity of electricity load on the national grid. Water heating alone accounts for ~15-20% of the energy used in residential energy consumption. Here, we consider the feasibility of using residential electric water heaters to buffer peak load and provide peak load shifting. Electric water heater load shifting pertains to demand-side management which is attractive due to low first-costs and reduced transmission losses when compared to supply-side management or conventional electrochemical battery storage. We first evaluated the performance of a Priority Based Control system to limit power consumption of a group of water heaters. We simulated the performance of 25 water heaters using measured hot water use data from occupied homes, and we found the maximum peak power reduction. We calculated the reduction through logical operators and temperature states, which successfully optimized the heating cycles for the targeted power. Next, we used machine learning toolkits to determine the hot water draw patterns in a residential setting. Priority Based Control of the water heaters provided peak reduction up to 66%, and Recurrent Neural Networks showed the greatest time series predictivity and scalability for larger datasets. The results from this study show great potential for using predictive algorithms in electric water heaters to further strengthen national energy grid sustainability.
License: MIT License