Graph Neural Networks for Real-Time Supply Chain Risk

Abstract

Constant supply chain disruption due to a changing demand, supplier disruptions, supply chain delays, geopolitical disruptions, cyber events, environmental shocks, and more, makes understanding the status of the supply chain more important than ever. Many of the traditional risk prediction models are risk-centric, considering supply chain information just as an it Fact record, and overlooking the sophisticated interdependencies that supply chain inputs, including suppliers, production facilities and warehouses, transportation networks, retailers and customers have with each other. This research work attempts to develop a framework, feasible through GNN for the real-time supply chain risks identification, prediction and response. Logistics platforms, market indicators and outside risk feeds are dynamically inserted into a graph of the supply chain, while various nodes represent the entities of the supply chain and edges represent the material flow, financial dependency, transportation, information exchange, contractual associations etc. between the entities. Real-time data feeds from enterprise resource planning data systems and IoT sensors are introduced into a dynamic supply chain graph. The Graph Network Module is used to discover hidden patterns, propagation of risk, detection of vulnerable nodes and prediction of the impact of disruptions across the network. The framework consists of four main steps: data acquiring and integrating, dynamic graph constructing, constructing graphs by GNN and risk analysis and visualization for decision-support. The goal is to model the risks on each node of the distribution network, along with the knock-on effect throughout the network, and bolster the early-warning system to better facilitate quicker mitigation measures, including the substitution of supplies, inventory relocation, rerouting routes, and relocating demand. The study shows the feasibility of applying graph-based learning methods, in smart supply chain management aimed at enhancing SC visibility, resilience and real-time decision making in complex SC in the presence of uncertainties.