Data-Driven Optimal Resource Allocation with Shared Capacity in Cold Chains' Warehouses

As customers seek more healthy lifestyles, the demand for fresh products (e.g., foods, vegetables, fruits) is undergoing a significant increase in recent years, thereby rendering escalating pressure on modern cold chain management. Specifically, the mismatch between enormous demand and the limited capacity of cold chain infrastructure is viewed as the main barrier to the cold chains' development in emerging markets. In this article, we focus on a resource allocation problem with shared capacity in a cold warehouse of JD.com, and attempt to design an inbound and outbound control mechanism for the cold store to increase its resource utilization and system revenue. We classify the products into two categories according to their storage characteristics: type-C, continuity requests with high priority and type-T, terminability requests allowing early discharge for new arrivals to avoid a prematurely overloaded warehouse and missing high-value orders. We formulate this problem as an event-based Markov decision process and show that theoretical admission thresholds exist for individual products based on the concavity of its value function. Moreover, as the uncertain demands significantly damage the quality of decisions, we further propose a data-driven robust Markov decision process in which the transition probability matrix depends on a rectangular uncertainty set and an optimal iterative algorithm to solve the robust counterpart. Sensitivity analysis indicates that an optimal warehouse capacity is achievable, and the long inventory ages incur tremendous profits loss for the warehouse. Finally, the numerical studies based on real-world data show that the optimal decisions from the robust Markov decision process outperform the benchmark in usage by 17% on average, verifying the effectiveness of our policy in practical environments