An Artificial-Immune-System-Based Algorithm Enhanced with Deep Reinforcement Learning for Solving Returnable Transport Item Problems

This paper proposes a new approach, i.e., virtual pooling, for optimising returnable transport item (RTI) flows in two-level closed-loop supply chain. The chain comprises set of suppliers delivering their products loaded on RTIs to customers. are various types. objective is model deterministic, multi-supplier, multi-customer inventory routing problem with pickup and delivery multi-RTI. includes inventory-level constraints, the availability empty suppliers, minimisation total cost, including holding, screening, maintenance, transportation, sharing, purchasing costs RTIs. Furthermore, common customers coordinate virtually pool held by so that, when delivered customers, each may benefit from this visit pick up RTI, regardless ownership. To handle combinatorial complexity model, artificial-immune-system-based algorithm coupled deep reinforcement learning proposed. combines artificial immune systems’ strong global search ability self-adaptability into goal-driven performance enhanced learning, all tailored suggested mathematical model. Computational experiments randomly generated instances highlight proposed approach. From managerial point view, results stress that approach allows economies scale cost reduction at level involved parties about 40%. In addition, sensitivity analysis unit transportation procurement conducted, highlighting benefits limits compared dedicated physical pooling modes.