operations research

In supply chain decision-making, optimization recommendations suffer from poor interpretability, complex human–system interaction, and delayed model updates—resulting in decision cycles spanning days to weeks and heavy reliance on data science teams. This paper introduces the first LLM-powered intelligent interaction layer tailored for supply chain optimization, integrating natural language understanding, knowledge-based reasoning, and optimization tool orchestration. Our approach delivers three key contributions: (1) automated generation of human-interpretable explanations for optimization outputs; (2) natural-language-driven dynamic scenario simulation and “what-if” analysis; and (3) business-feedback-guided adaptive retraining and updating of mathematical optimization models. By eliminating manual intermediation, the framework reduces decision latency to minutes, significantly enhancing autonomous decision-making capabilities for planners and executives. It advances the democratization and real-time operation of supply chain decision technologies.

This work addresses the dynamic replenishment scheduling problem under shared capacity constraints for multiple commodities—a setting long hindered by a 2-approximation barrier. We present the first polynomial-time algorithm that provably breaks this barrier by introducing a novel analytical framework that directly compares dynamic and classical static policies. Central to our approach is a refined cost-capacity balancing mechanism that yields a randomized, capacity-feasible dynamic policy. Our method constructs, in polynomial time and for any instance, a solution whose expected long-run average cost is at most $(2 - 17/5000 + \varepsilon)$ times the optimal, thereby achieving the first provably sub-2 approximation for dynamic scheduling in this setting.

This paper studies the multi-stage dynamic assortment optimization problem under knapsack-style inventory constraints: a retailer must dynamically adjust its product assortment each period—based on the multinomial logit (MNL) choice model—to maximize cumulative profit as inventory depletes over time. Since the problem is NP-hard and conventional approaches (e.g., static planning or greedy heuristics) lack theoretical performance guarantees, we propose the first epoch-based re-optimization algorithm with provable bounds. Our key innovation lies in reformulating the denominator structure of the MNL objective as linear constraints, enabling tractable fluid approximations and rigorous stochastic analysis. The algorithm achieves an $O(log(TC))$ regret bound—logarithmic in the time horizon $T$ and total capacity $C$—while maintaining computational efficiency and asymptotic optimality. It significantly outperforms existing methods both theoretically and empirically.

This paper addresses the last-truck scheduling problem in e-commerce middle-mile transportation, aiming to maximize next-day delivery order fulfillment under fixed inventory locations and sufficient last-mile delivery capacity. We formulate this problem as an NP-hard submodular optimization problem with coverage constraints—the first such formulation in the literature. To solve it, we propose three scenario-adaptive algorithms integrating greedy submodular maximization, pipage rounding, and Lagrangian relaxation heuristics, achieving both theoretical guarantees (worst-case approximation bounds) and scalability. Extensive experiments on real-world logistics networks and datasets demonstrate that our solutions closely approximate the global optimum, significantly improving next-day delivery coverage while maintaining computational efficiency. The proposed framework is production-ready and demonstrates strong industrial deployability.

This work proposes the first polynomial-time approximation scheme with provable performance guarantees for the economic lot-sizing problem, a classic inventory management challenge that has lacked such algorithms since its formulation in the 1950s. By integrating dynamic programming encodings, approximation techniques, and combinatorial optimization, the authors establish an efficient representation and optimization framework for dynamic replenishment policies, resolving the long-standing open issue that such policies inherently require exponential space. When the number of item types is constant, the method constructs an ε-optimal dynamic policy in polynomial time, overcoming prior limitations that relied on restrictive structural assumptions or offered no performance guarantees. This breakthrough substantially advances the algorithmic tractability of this fundamental problem.

This study addresses the service network design challenge in intermodal logistics networks arising from uncertain travel times and limited truck fleet availability. The authors propose a two-stage optimization framework that jointly handles tactical decisions—such as transport request acceptance and capacity reservation—and operational decisions, including dynamic vehicle dispatching and route replanning. Innovatively integrating an adaptive machine learning surrogate model with a simulated annealing algorithm, the approach effectively mitigates cascading disruptions caused by uncertainty-induced perturbations. Computational experiments demonstrate that the method achieves up to a 20-fold reduction in solution time while maintaining solution quality within a 5% deviation from the optimal objective value, significantly outperforming existing benchmark approaches.

This work addresses the design of potential flow networks—a class of optimization problems that are significantly harder than classical network flow due to their inherent nonlinearity. The paper presents the first effective approximation algorithm framework for this problem by introducing a refined reduction to well-studied combinatorial optimization problems such as constrained shortest paths, thereby enabling efficient solutions through existing algorithmic techniques. The study establishes matching complexity lower bounds that precisely delineate the approximability frontier of the problem and further demonstrates the NP-hardness and inapproximability of several key variants. Collectively, these results provide a comprehensive characterization of the computational complexity and algorithmic tractability of potential flow network design.

This study addresses the insufficient supply chain resilience in existing research, which often overlooks inter-tier disruption dependencies. To bridge this gap, we propose a two-stage mixed-integer programming model that explicitly captures disruption dependencies among facilities and jointly optimizes resilience (via backup reallocation), agility (through mobile facilities), and carbon emission constraints. To handle service-level probability requirements, we introduce linear cuts derived from machine learning classifiers—such as L1-regularized logistic regression—as surrogates for chance constraints. The model is efficiently solved by integrating sample average approximation with a Fix-and-Relax heuristic. Computational experiments demonstrate that our approach significantly improves solution efficiency while maintaining a 95% service level, enabling rapid generation of high-quality solutions for medium- to large-scale instances.

This study addresses the single-product lot-sizing problem under uncertain lead times and a budget constraint, incorporating practical considerations such as backlogging, inventory holding, and the infeasibility of splitting or crossing orders. The authors propose a robust optimization approach based on the R* criterion, which transcends the conventional min-max framework by constructing a continuum of production plans ranging from the most optimistic to the most pessimistic scenarios, thereby expanding the feasible solution space. Notably, this work is the first to integrate a budgeted uncertainty set with non-crossing order constraints and develops both polynomial- and pseudo-polynomial-time algorithms alongside a mixed-integer programming formulation for solution. Computational experiments demonstrate that the R* criterion significantly enhances decision flexibility and system performance while preserving feasibility.

This work addresses the challenge of automatically translating complex business requirements into optimization models for multi-warehouse inventory allocation in e-commerce. To this end, the authors propose ORLA, a novel framework that, for the first time, integrates solver feedback into the generative loop of large language models to automatically construct, validate, and select mixed-integer programming formulations from natural language or semi-structured inputs. ORLA supports dynamic constraints, infeasibility recovery, and modular extensibility, while incorporating modeling paradigms such as deviation minimization, soft bandwidth limits, and knapsack-style formulations. Evaluated on 29 real-world production batches from JD.com, ORLA improves allocation accuracy by 4.5 percentage points overall, significantly outperforming existing approaches.