Maximum Order Tree Algorithm for Optimal Scheduling of Product Distribution Lines
Mokashi, S.D., Kokossis, A.C.
AIChE Journal, vol.48, no.2, p.287-301
Research on scheduling and planning in the chemical engineering community subscribes to one of two schools of thought. A general-purpose optimization approach resorts to using conventional mathematical programming techniques on generic models of a scheduling problem, which has a limitation in its application to large-scale industrial problems in terms of the computational time involved. The other extreme of heuristic methods lacks guarantees on the quality of the solution. A philosophy is proposed of contextual optimization that exploits problem-specific knowledge to develop efficient algorithms. This concept is applied to a delivery scheduling problem to generate a tailored graph-based method called the maximum order tree algorithm, which reduces the CPU time dramatically compared to conventional methods without compromising on the quality of the solution. When applied to a single-site distribution case study, it resulted in savings of over a quarter of a million dollars per year over the existing heuristic-rule-based system.