Multi-class hazmat distribution network design with inventory and superimposed risks

•Develop a multi-class distribution network design problem with inventory and superimposed risks.•Introduce risk superposition coefficients for multi-class hazmat to capture the superimposed risks among different materials resulting from possible chemical reaction.•Formulate the long-term detour cost/risk and the cyclic time windows penalty costs under the time-dependent (periodic) road closure policy.•Propose a population-based risk definition that evaluates the risk for the population at any location and any time with respect to its multi-class hazmat logistics system Provide a metropolitan wide real-world application and new insights. Transportation and inventory are essential to hazardous materials logistics, while different classes of hazardous materials are often transported over a network simultaneously. Despite their in-transit and storage incompatibility, the superimposed risks among different materials, which results from possible chemical reaction once accidents (e.g., leakage, explosion) happen, further complicate the comprehensive plans. In this study, we introduce a new multi-class hazmat distribution network design problem with inventory and superimposed risks (MHND) in a multi-echelon supply chain, where the planning of locations, inventory, and routes are made together. The long-term detour cost/risk and the cyclic time windows penalty costs under the time-dependent (periodic) road closure policy are explicitly formulated. We further propose a new population-based risk definition that evaluates the risk for the population at any location and any time with respect to its multi-class hazmat logistics system. In particular, to capture the interactions between different types of materials, we introduce risk superposition coefficients to capture possible superimposed risks among different hazmat that accommodate a general system with more than two hazmat types. We develop a knowledge-based NSGA-II algorithm with cyclic dissimilarity-based elitist selection (NSGA-II-CD) to solve the problem. The devised cyclic dissimilarity-based elitist selection (CD) operator can tackle the issue of speeding proliferation, which greatly improves the solution quality. Our model is applied to a metropolitan-wide real-world case study in Guangzhou, China. The results suggest that, from the perspective of the traffic management sector, the periodic road closures policy in Guangzhou could be possibly upgraded to a full-time prohibition. Moreover, the results provid