A two-stage stochastic programming model for perishable goods allocation under LIFO customer preferences and diverse demand patterns
针对易腐品零售网络，提出两阶段随机规划模型，考虑后进先出偏好和不确定需求，并开发三种启发式方法（需求比例分配、贪婪分配、加权K均值场景缩减）以提升计算效率，实验表明场景缩减可减少90%计算时间且保持高精度。
Efficient inventory management for perishable goods is a critical operational challenge for retailers due to the complexities introduced by limited shelf-life and uncertain demand. This paper presents a two-stage stochastic programming model tailored for optimising perishable goods distribution in retail networks, explicitly addressing last-in-first-out customer preferences, demand uncertainty and fixed shelf-life. The model integrates operational objectives including cost minimisation, demand satisfaction, and waste reduction through scenario-based stochastic optimisation. Due to computational complexity in solving large-scale stochastic problems, we introduce heuristic and approximation methodologies for improved scalability. Specifically, we develop three approaches: a Demand-Based Fractional Allocation heuristic fairly distributing goods among locations, a Ranked Greedy Allocation heuristic prioritising stores based on profitability, and a Scenario Reduction method using Weighted K-means clustering to manage scenario complexity efficiently. These heuristics are evaluated using synthetic datasets and real-world-inspired instances from the M5 forecasting competition dataset, covering various demand patterns including intermittent and smooth series. Results indicate that scenario reduction via Weighted K-means significantly improves computational efficiency (up to 90% reduction in computation time for short shelf-lives) while maintaining high solution accuracy for smooth demands (average optimality gap 1.37%). The greedy heuristic effectively minimises waste for intermittent demands, highlighting the need for testing general allocation approaches considering diverse demand patterns.