Model-free Demand Response Scheduling Strategy for Virtual Power Plants Considering Risk Attitude of Consumers

Driven by modern advanced information and communication technologies, distributed energy resources have great potential for energy supply within the framework of the virtual power plant (VPP). Meanwhile, demand response (DR) is becoming increasingly important for enhancing the VPP operation and mitigating the risks associated with the fluctuation of renewable energy resources (RESs). In this paper, we propose an incentive-based DR program for the VPP to minimize the deviation penalty from participating in the power market. The Markov decision process (MDP) with unknown transition probability is constructed from the VPP's prospective to formulate an incentive-based DR program, in which the randomness of consumer behavior and RES generation are taken into consideration. Furthermore, a value function of prospect theory (PT) is developed to characterize consumer's risk attitude and describe the psychological factors. A model-free deep reinforcement learning (DRL)-based approach is proposed to deal with the randomness existing in the model and adaptively determine the optimal DR pricing strategy for the VPP, without requiring any system model information. Finally, the results of cases tested demonstrate the effectiveness of the proposed approach.