Incorporating Constraints in Reinforcement Learning Assisted Energy System Decision Making: A Selected Review

With the widespread use of reinforcement learning (RL) in the energy system, the damage to the system caused by the agents' stochastic exploration is beginning to be appreciated. We need to ensure the safety of agents during the whole training process, even at each time step. Therefore, we need to consider how to incorporate safety constraints into RL. In this paper, we summarize two approaches to soft constraints and three approaches to hard constraints, respectively, and provide a comparative analysis of their advantages and disadvantages. Finally, we conclude approaches to how to incorporate constraints in RL, and give an outlook on future research directions.