Distribution network fault reconfiguration method based on interpretable reinforcement learning

As the proportion of flexible resources such as distributed photovoltaics, energy storage, and electric vehicles continues to increase, the new distribution systems have undergone profound changes, with some regions experiencing large-scale power backfeeding, leading to frequent faults. Against this backdrop, this study proposes an interpretable reinforcement learning method that combines the Soft Actor-Critic (SAC) algorithm and decision trees, aimed at assisting with the rapid and reliable fault reconfiguration of distribution networks. The effectiveness of this method was verified through simulations of the IEEE 123-node distribution system. This approach can quickly provide optimal solutions for load recovery, network losses, and the number of switch operations under various conditions, thereby significantly enhancing the operational efficiency and reliability of the distribution network.