Dual‐Sourcing Logistics Network Design Considering Time‐Based Service Levels

ABSTRACT Effective logistics networks drive cost efficiency, reliable delivery, and customer satisfaction for e‐commerce companies. Motivated by JD Logistics' regional warehouse network operations problem under uncertain customer demand and delivery times, this paper proposes a dual‐sourcing location‐inventory model with time‐based service levels (DSLI‐T). We formulate the problem within a scenario‐wise distributionally robust optimization framework that incorporates continuous delay times via a conditional value‐at‐risk measure to address tail risks. By deriving tractable closed‐form expressions for the worst‐case risk‐adjusted costs and for distributionally robust cycle service‐level constraints, we reformulate the DSLI‐T model as an equivalent mixed‐integer nonlinear programming (MINLP) model with submodular objective costs. To solve this model efficiently, we propose improved extended polymatroid inequalities (I‐EPIs) that jointly exploit submodularity and generalized upper bound constraints. We prove that the I‐EPIs are facet‐defining for the corresponding polyhedron and, under mild conditions, strictly dominate classical extended polymatroid inequalities (EPIs). Based on these I‐EPIs, we develop an enhanced branch‐and‐cut (B&C) algorithm accelerated by a two‐phase implementation framework. Case studies on JD Logistics demonstrate that the dual‐sourcing strategy reduces the total cost by 12.01%. Numerical experiments show that the proposed enhanced B&C algorithm with I‐EPIs achieves reductions of 77.7% and 98.4% relative to EPIs and MINLP, respectively.