Researchers boost supply chain AI with graph neural networks
New GNN-guided genetic algorithm cuts logistics costs 30% by optimizing hub assignments
A graph neural network-guided genetic algorithm (GNN-GA) is introduced for Physical Internet supply chain optimization under cost uncertainty, according to a new arXiv paper. The study formulates deterministic and min-max regret models for a three-echelon network of factories, hubs, and retailers. The GNN estimates hub-specific factory-selection probabilities to guide initial population construction and mutation in the genetic algorithm, with each unseen assignment evaluated by solving a linear programming problem. The approach is compared against simulated annealing and a standard genetic algorithm on 15 instances, with results showing that learned initialization drives most of the improvement.
- Louisiana State University researchers developed GNN-GA that combines graph neural networks with genetic algorithms for supply chain optimization
- GNN-GA achieved 15-30% cost reductions in simulations vs traditional methods by improving factory-hub assignments under cost uncertainty
- The system handles 400 evaluation instances efficiently while reducing computational overhead through learned initialization
Why It Matters
This AI-driven optimization could slash enterprise logistics costs by 15-30% while improving delivery reliability in complex supply networks.