Multi-period stochastic optimisation of the integrated production-inventory-distribution problem
研究了随机需求下多周期集成生产-库存-分销优化问题，提出随机对偶动态规划算法求解，并用饮料制造商实例验证算法能在1秒内响应新信息，为管理者提供决策支持。
This paper investigates the multi-period integrated production-inventory-distribution optimisation problem of a complex supply chain under stochastic demands. The network includes heterogeneous plants with various capacities and unit production costs, distribution centers (DCs), and retailers across multiple locations. Finished products can be stored in plant warehouses or transported to DC warehouses with varying holding costs. At the end of each period, the manufacturer faces random demands from retailers, which can be fulfilled from inventories of the plants and DCs. Unfulfilled demands are backlogged incurring backlogging cost. Transportation between different locations incurs various costs. The objective is to make integrated production-inventory-distribution decisions to minimise total cost over multiple periods, consisting of production, transportation, backlogging, and holding. To address the challenges arising from intricate objective and high-dimensional stochastic demands, states, and actions, we formulate the problem as a stochastic dynamic programming model and propose stochastic dual dynamic programming (SDDP) algorithm to solve it. Using real-case-based instances derived from a beverage manufacturer, we conduct numerical experiments to evaluate the performance of the SDDP algorithm by comparing it with two benchmark approaches and the perfect information bound. Results demonstrate the SDDP algorithm reacts quickly to new information within one second and provide valuable managerial insights.