Extending no-MASS : multi-agent stochastic simulation for demand response of residential appliances

Demand Response (DR) has been proposed as an efficient and inexpensive solution to face the challenges of the evolving power system: to reduce peak demands and thus demands for additional generating capacity, and to improve localized energetic autonomy. The success of DR programs is highly dependent on the acceptance and reactions of end-users: their willingness to devolve control and/or to proactively adjust their energy using behaviours. However, experimental trials to identify the best DR configuration for each type of consumer is very costly. Here simulation has a valuable role to play, in identifying promising DR strategies, taking into account occupants characteristics. This paper describes and evaluates a proof-of-concept extension of a multi-agent stochastic simulation platform that simulates occupants behaviours (No-MASS) to also simulate DR strategies. This is applied in the first instance to electrical devices in households: categorizing these as demand devices (e.g. large and small appliances, heating and hot water systems), supply devices (e.g. PV, cogeneration, wind turbines) and electrical storage devices (e.g. batteries, HEVs). The device-specific DR strategies are learned by our device-agents using machine (specifically reinforcement) learning, to maximize a function rewarding the utilization of locally generated energy. We close this paper by discussing our next steps to add further functionality to No-MASS for DR simulation.