A Two‐Stage Strategy for Black‐Start Restoration by Coordinating Deep Reinforcement Learning and Mixed‐Integer Linear Programming

ABSTRACT Extreme events can lead to extensive blackouts, emphasising the crucial significance of power system black‐start restoration. In this paper, a two‐stage strategy coordinating deep reinforcement learning (DRL) and mixed‐integer linear programming (MILP) for power system black‐start restoration is proposed. In the first stage, a convolutional neural network‐based (CNN‐based) invalid action masking assisted proximal policy optimisation (CI‐PPO) algorithm is developed to determine component connections. Furthermore, a MILP model for optimal power dispatch is formulated with the objective of minimising generation costs in the second stage. The effectiveness of the proposed method is validated using the New England 10‐unit 39‐bus power system. The results show strong learning ability and computational efficiency for varying system conditions.