A Randomized Local Search Heuristic for Lot Sizing in Multistage Assembly systems

Multistage production systems consider the production of items at more than one level. There are three general classes of such systems: (1) assembly systems where each stage requires inputs from one or more immediate predecessor stages but supplies exactly one immediate successor, (2) arborescent systems where each stage has exactly one predecessor stage but could have several immediate successors, and (3) general systems which have a combination of assembly and arborescent structures. Usually an external demand exists only for the end item at the last stage, and this in turn generates dependent demands for items further upstream in the hierarchy. This demand interdependency complicates the production planning problem tremendously, since lot sizing decisions for items at one stage depend on decisions made by its successor and/or predecessor stages. In this study we propose a robust heuristic solution technique for the multistage assembly problem based upon stochastic neighborhood search. This heuristic procedure which we refer to as Randomized Local Search (RLS), intelligently searches the problem domain and while considering only a small portion of the search space yields low cost lot sizing decisions in very little time. Moreover it is very easy to understand and implement, while simultaneously being based on the theoretical properties of the optimum solution.