Physics-Informed Graph Neural Networks for Supply Chain Disruption Prediction and Mitigation

Abstract

Global supply chains face unprecedented challenges from multi-modal disruptions including natural disasters, geopolitical tensions, and market volatility. Traditional data-driven approaches for disruption prediction often fail to capture the underlying physical constraints and causal relationships governing supply chain dynamics. This paper introduces a novel Physics-Informed Graph Neural Network (PI-GNN) framework that integrates domain knowledge from supply chain theory with graph-based deep learning architectures for enhanced disruption prediction and mitigation strategies. The proposed methodology embeds physical laws such as conservation of flow, capacity constraints, and lead time dependencies directly into the neural network training process through custom loss functions and architectural constraints. We demonstrate that by incorporating physics-based regularization terms derived from supply chain fundamentals, the PI-GNN achieves superior predictive performance compared to purely data-driven GNNs, particularly in scenarios with limited historical data. Experimental results on real-world supply chain networks show that the PI-GNN framework reduces prediction error by 23% for disruption events and provides interpretable insights for proactive mitigation strategies. The framework facilitates real-time risk assessment across multi-tier supply networks while maintaining computational efficiency suitable for large-scale deployments.