Domain Randomization for Demand Response of an Electric Water Heater

Thermostatically Controlled Loads (TCLs) provide a source of demand flexibility, and are often considered good for Demand Response (DR) applications. Due to their heterogeneity, as such lack dynamics models, Reinforcement Learning (RL) is used exploit this flexibility. Unfortunately, RL requires exploratory interaction with the TCL, resulting in period potential discomfort users. We present an approach reduce time by pre-training RL-agent. Domain randomization facilitate knowledge transfer. evaluate DR energy arbitrage scenario Electric Water Heater (EWH). Our experiments show that priori about EWH can be initialize improve control policy. In our experiments, attributes 8.8% additional cost savings, compared starting from scratch.