IBBB-STGNN: A Hybrid Spatiotemporal Graph Neural Network with Big Bang–Big Crunch Optimization for Real-Time Power Distribution Path Optimization

Abstract

Modern electrical distribution networks require intelligent and adaptive systems that can dynamically track and optimize power flow paths under variable load and topology conditions. Traditional optimization methods often struggle to adapt in real-time and cannot capture both temporal fluctuations and spatial dependencies inherent in distribution systems. Many existing deep learning approaches focus on either temporal or static topological modeling, limiting their performance in real-time path optimization tasks. To address these challenges, an Intelligent Big Bang–Big Crunch-driven Spatiotemporal Graph Neural Network (IBBB-STGNN) is introduced. This hybrid model integrates an IBBB optimization algorithm with an STGNN to simultaneously learn the network structure and time-based load dynamics for efficient power routing. A synthetic yet realistic dataset is constructed, incorporating parameters such as voltage, active/reactive power, power losses, resistance, load types, and switching events. Preprocessing includes missing value imputation and Min-Max normalization, ensuring data quality and consistent scale. Feature extraction involves lag-based attributes, rolling statistics, power factor, load gradients, and principal component compression via Principal Component Analysis (PCA). The proposed model components function in synergy: the GNN captures spatial relationships, the temporal encoder models timeseries behavior, and the IBBB module optimizes switching decisions for optimal path selection. Experimental evaluation demonstrates superior performance in terms of power loss minimization of up to 75%. This approach effectively bridges the gap between graph-based learning and metaheuristic optimization, enabling intelligent, real-time, and topology-aware power distribution management.