A novel rolling horizon product-centric MILP based approach for a real-world integrated production and distribution scheduling problem

This paper addresses an integrated, real-world detailed scheduling problem of OCP Group operations, in the context of the phosphate mining industry. The studied problem incorporates a production scheduling problem in multiple washing stations, a multiproduct pipeline network scheduling problem, and a multiproduct inventory management problem, where tanks are shared for storing different products at different periods. We propose a discrete-time, product-centric MILP framework based on a rolling-horizon decomposition with two components. First, a MILP model generates scheduling solutions for each one of the sub-horizons while incorporating maintenance windows and minimizing product shortages and the total volume of injected water into pipelines. Each sub-problem is solved with a total discharge of batches still in the pipelines at the end of the scheduling horizon, ensuring that downstream inventory levels and demands are accurately updated. Additionally, a fixed portion of each sub-solution is discarded to mitigate edge effects and preserve solution quality across sub-horizons. Second, once all the sub-solutions are assembled, a post-processing MILP model refines the final solutions by minimizing unnecessary state changes across multiple units, resulting in good-quality, directly implementable solutions. The developed approach provides a holistic framework that addresses the limitations of previous attempts to solve the studied problem. To evaluate its performance, it was tested on ten challenging high-demand industrial instances over a seven-day rolling horizon. The experimental results demonstrate its effectiveness in providing high-quality solutions with high demand satisfaction rates and low water consumption in a reasonable CPU time.