A novel grid load management technique using electric water heaters and Q-learning

This paper describes a novel technique for controlling demand-side management (DSM) by optimizing the power consumed by Domestic Electric Water Heaters (DEWH) while maintaining customer satisfaction. The system has 18 states based on three factors: instantaneous grid load, water consumption, and the temperature of the water supplied. The current state of the system is defined based on its fuzzy membership for each factor. The resulting model represents a Semi-Markov decision process (SMDP) with two possible actions, “On” and “Off.” Rewards are assigned for each action-state pairs proportionally to the fuzzy membership of the system in the new state. A simulation study was conducted to compare the proposed method with three previous approaches. The proposed method demonstrated better performance in reducing the overall grid power demand and flattening its peaks. Furthermore, it provides better rate of customers' satisfaction than the uncontrolled operation.