On the optimization of warehousing handling unit partitioning
研究了随机需求下零件到拣选系统的搬运单元划分问题，提出精确两阶段随机混合整数规划和遗传算法两种求解器，并用美国配送中心数据验证了多模式划分可节省约10.6-12.6%的货架面。
This study examines handling-unit partitioning (HUP) for parts-to-picker systems under stochastic demand—an NP-hard design problem—and propose a practical framework for industrial deployment. We develop two complementary solvers: an exact two-stage stochastic mixed-integer program (SMIP) for moderate sizes and a two-stage genetic algorithm (GA) for larger instances, both using sample-average approximation to represent demand uncertainty. We introduce a simple HUP pattern-distance metric and show that maximizing the minimum distance among selected patterns reliably shortens SMIP convergence time without degrading objective value. Using data from a U.S. distribution center, we find that allowing two patterns (vs. one) reduces required pod faces by about 10.6–12.6% (statistically significant), with diminishing returns beyond two—yielding meaningful footprint and capital savings. The stochastic design consistently outperforms a deterministic average-demand baseline, while the incremental benefit available even with perfect foresight is modest, indicating that most of the achievable improvement is already captured. Across instances, GA attains about 95% of SMIP’s objective on average while generally running faster and scaling to e-commerce–sized assortments.