Deep learning-based dynamic tracking and optimization of power supply paths in main distribution networks

Modern electrical distribution networks need intelligent and adaptive systems to manage power flow under changing load and network conditions. Conventional optimization methods struggle with real-time adaptation and cannot fully capture temporal variations and spatial relationships. Most existing deep learning models focus on either time dynamics or network topology, limiting their effectiveness. To address this, an Intelligent Big Bang–Big Crunch driven Spatiotemporal Graph Neural Network (IBBB-STGNN) is proposed. The model integrates the IBBB optimization algorithm with a Spatiotemporal GNN to jointly learn network structure and time-varying load behavior for efficient power routing. A realistic synthetic dataset is developed, including voltage, active and reactive power, losses, resistance, load types, and switching events. Data preprocessing, feature extraction, and dimensionality reduction are applied. The GNN captures spatial dependencies, the temporal encoder models time patterns, and IBBB optimizes switching decisions. Results show up to 75% power loss reduction, demonstrating effective, real-time, topology-aware power distribution management.