Deep Reinforcement Learning for an Incentive-based Demand Response Model

Incentive-based demand response is often modeled as a stackelberg game, which usually need complete information of customers. Reinforcement learning is always utilized to solve this problem under incomplete information. Prior to conventional reinforcement learning, which is vulnerable to dimensional disasters because of state and action space discretization, this paper proposed a deep reinforcement learning algorithm that can solve continuous state and action space to obtain incentive-based demand response interaction strategies of both participants under incomplete information. Simulation experimental results show that the results of the proposed method can obtain the optimal solution in the case of incomplete information in different types customers.