Optimal supply chain design using machine learning, risk assessment and optimisation applied to coal distribution

Abstract

This paper presents a novel integration of machine learning and optimization techniques for robust supply chain network design under uncertainty, demonstrated through a regional coal distribution system (150 suppliers, 500 consumers, 7.5 million tonnes annually). Unlike traditional approaches that focus solely on cost minimization, our progressive methodology uniquely combines k-means clustering with mixed-integer programming to identify configurations that are both cost-effective and resilient to demand variability. Starting from a baseline single-warehouse configuration costing $404.6 million annually, our approach systematically evaluates multi-facility alternatives through Monte Carlo simulation (50,000 iterations), revealing remarkable system stability—only 0.96 % cost variation despite 25 % individual volume uncertainty. The k-means analysis identifies optimal clustering patterns, while subsequent mixed-integer programming confirms that a five-warehouse configuration reduces annual transportation costs by 45.8 % ($185.3 million savings) while maintaining 90–100 % capacity utilization. This configuration demonstrates a Net Present Value of $3.02 billion over 50 years, significantly outperforming traditional single-facility designs. Critically, correlation analysis reveals that shipment volumes (ρ=0.739) drive costs more than distances (ρ=0.556), challenging conventional distance-minimization paradigms. The integrated framework is computationally efficient (4.2 s to optimality), scalable to larger networks, and applicable to various bulk commodity distribution challenges, offering supply chain managers a robust tool for strategic network design under uncertainty.