Load Scheduling for an Electric Water Heater With Forecasted Price Using Deep Reinforcement Learning

Abstract

Electric water heaters have time-accumulating effect due to the ability to store energy, which makes them attract much attention in the field of smart homes. The load scheduling problem of an electric water heater is challenging due to the uncertainty of future electricity prices and the applicability of different types of electric water heaters. This paper formulates the problem as a Markov decision process (MDP) and uses deep reinforcement learning (DRL) to obtain the optimal scheduling policy. Specifically, Long Short-Term Memory (LSTM) is applied to forecast the future prices. Then adopt Deep Q Network (DQN) to determine the optimal strategy for this problem. LSTM can deal with the uncertainty of future prices. This model-free approach can handle the applicability of different types of electric water heaters. Simulation results indicate that the proposed algorithm can minimize the cost of electricity and dissatisfaction. In addition, the electricity cost and user dissatisfaction can be balanced by introducing a balance parameter according to user preferences.