Deep-Learning-Driven Fault Location and Service Restoration for Distribution-Network Apparatus

The growing topological complexity and load heterogeneity of modern distribution networks have exposed the limitations of conventional fault-location and service-restoration schemes that rely on rigid rules and operator experience. To overcome these limitations, this paper proposes a deep-learning-based framework for pinpointing equipment faults and enabling rapid service restoration. First, a node energy-deviation and restoration-demand model is established to quantify fault-state characteristics. Second, a CNN-BiLSTM spatio-temporal feature-fusion network is designed to achieve high-accuracy fault identification from multi-source signals. Third, a state-reconstruction layer and a reinforcement-learning agent are integrated to dynamically optimize the restoration path. Pilot tests conducted at a 110 kV substation demonstrate that the proposed method significantly improves fault-location accuracy, shortens response time, and raises the load-recovery rate, thereby enhancing the intelligence and operational stability of distribution-network equipment.