Deep Reinforcement Learning-Based Explainable Pricing Policy for Virtual Storage Rental Service

Abstract

The shared community energy storage system (CESS) can reduce energy storage costs by exploiting the complementarity of end-users and economies of scale. To further improve the economic feasibility of the CESS, we propose a novel business model and pricing method for the virtual storage rental service (VSRS). In this model, the rental users aim to minimize the electricity bill by renting the virtual capacity and optimizing its operation, while the CESS operator seeks to maximize the revenue from the combination of energy arbitrage and the VSRS. The pricing problem and the optimal operation problems of the CESS and users’ virtual batteries are modeled as a bi-level optimization problem. Next, the proposed problem is solved through transformer-based deep deterministic policy gradient (TDDPG) method and mixed-integer linear programming (MILP) due to the non-convexity and non-continuity of the original problem. The post-hoc interpretability of the policy network is provided based on the Shapley value to reveal the importance of different input features for decision-making. Numerical simulations suggest that the proposed VSRS could benefit the CESS operator and users. Moreover, the explanation based on the Shapley value could effectively generate an implicit solution for understanding the policy network.