Rolling-horizon lot-sizing when set-up times are sequence-dependent

Abstract

The challenging problem of efficient lot sizing on parallel machines with sequence-dependent set-up times is modelled using a new mixed integer programming (MIP) formulation that permits multiple set-ups per planning period. The resulting model is generally too large to solve optimally and, given that it will be used on a rolling horizon basis with imperfect demand forecasts, approximate models that only generate exact schedules for the immediate periods are developed. Both static and rolling horizon snapshot tests are carried out. The approximate models are tested and found to be practical rolling horizon proxies for the exact model, reducing the dimensionality of the problem and allowing for faster solution by MIP and metaheuristic methods. However, for large problems the approximate models can also consume an impractical amount of computing time and so a rapid solution approach is presented to generate schedules by solving a succession of fast MIP models. Tests show that this approach is able to produce good solutions quickly.