Performance of MRP, RPS, and ConWIP in Stochastic Multi-Stage Systems: An Item-Level Simheuristic Study

This study conducted a comprehensive comparison of three production planning and control systems – MRP, RPS, and ConWIP – across 18 stochastic multi-item, multi-stage production environments. The evaluation incorporated varying production system structures (flow, job, hybrid shops), planned utilization levels, and customer-required lead-time tightness. A simheuristic combining Simulated Annealing, Simulation Budget Management, and Exponential Range Reduction was applied to tune planning parameters at the item and component level, significantly enlarging the search space compared to earlier, system-level studies. The simheuristic consistently outperformed the baseline, with gains most pronounced under high utilization and complex structures. ConWIP proved most robust in medium- to high-load environments by stabilizing work-in-process. In contrast, MRP performed best under low utilization and relaxed due-dates, as demand-driven logic is most effective when congestion is minimal. While RPS was the least flexible and benefited least from tuning, it offered the highest efficiency with minimal computational overhead, providing a clear trade-off between optimization depth and responsiveness.本研究针对物料需求计划（Material Requirements Planning, MRP）、RPS系统（RPS）以及常量在制品（Constant Work-In-Process, ConWIP）三类生产计划与控制系统，在18个随机多品种、多阶段生产环境中开展了全面对比分析。本次评估涵盖了不同的生产系统结构（流水车间、单件车间、混合型车间）、计划设备利用率水平以及客户要求的交货期紧密度。本研究采用一种融合模拟退火（Simulated Annealing）、仿真预算管理（Simulation Budget Management）与指数范围缩减（Exponential Range Reduction）的仿真启发式算法（simheuristic），在物料与零部件层级对规划参数进行优化，相较于此前的系统级研究，该算法大幅拓展了搜索空间。该仿真启发式算法始终优于基准方案，在高设备利用率与复杂生产结构场景下的优化增益最为显著。ConWIP通过稳定在制品库存，在中高负载生产环境中展现出最强鲁棒性。与之相对，MRP在低设备利用率与宽松交货期场景下表现最优，因其需求驱动逻辑在生产拥堵程度较低时效果最佳。尽管RPS灵活性最差，且从参数优化中获得的收益最小，但它的计算开销极低、运行效率最高，在优化深度与响应性之间实现了清晰的权衡。