POMDP-Based Dispatch Scheme for Residential Distributed Energy Resources Under Customer Fatigue Consideration

With rapidly growing penetration of renewable energy, there is an increasing need for real-time balance in the power system. In recent years, as a vital component of the end user side, residential distributed energy resources (RDER) have shown promising prospects in demand response (DR). This paper addresses the RDER optimal dispatch problem from an online learning perspective. The Partially Observable Markov Decision Process (POMDP) and the logistic regression model are employed to describe how the DR service contractor learns the uncertain response model of the customer affected by the fatigue and recovery effect. A non-stationary combinatorial multi-armed bandit paradigm is applied to dispatch RDER. In addition, we use the maximum likelihood estimation (MLE) technique to estimate the customer’s unobservable states. Through simulations based on public datasets, our proposed Greedy-MLE and CUCB-MLE policies for dispatch decision-making outperform existing methods in relative power adjustment mismatch, collective fatigue, and cumulative regret. The simulation results highlight the benefits of the proposed RDER dispatch approach in ensuring customer satisfaction, service quality, and economic efficiency.