Building load management clusters using reinforcement learning

In this paper we introduce a customer selection model for decision making in a Demand response program. In particular, we focus on modelling demand response as a reinforcement learning problem that decomposes the customers into clusters based on their ability to provide curtailments at time of Demand response signal. The reinforcement learning approach allows the retailer to make fast informed decision on the customers reliable to provide demand management capabilities without the need of exploring the entire set of customers when needed in real-time and allow for classification of future customers in appropriate clusters. We demonstrate using this approach on a representative example to create clusters based on provided customer profiles for varying DR signals.