Integrating meta-heuristics, simulation and exact techniques for production planning of a failure-prone manufacturing system

• Published in: European journal of operational research Vol. 266; no. 3; pp. 976 - 989
• Main Authors: Diaz, Juan Esteban, Handl, Julia, Xu, Dong-Ling
• Format: Journal Article
• Language: English
• Published: Elsevier B.V 01.05.2018
• Subjects: Combinatorial optimization; Genetic algorithms; Production planning; Simulation-based optimization; Uncertainty modelling; Combinatorial optimization; Production planning; Simulation-based optimization; Uncertainty modelling; Genetic algorithms
• ISSN: 0377-2217; 1872-6860
• DOI: 10.1016/j.ejor.2017.10.062

•We consider problems with uncertainty associated with the decision variables.•A simulation-based (meta-heuristic) optimization method is described.•We study seeding mechanisms exploiting a combination of mathematical programming and simulation.•Our method outperforms contestant (pure mathematical programming) techniques.•For growing levels of uncertainty, the performance advantage of our method remains. This paper considers a real-world production planning problem in which production line failures cause uncertainty regarding the practical implementation of a given production plan. We provide a general formulation of this problem as an extended stochastic knapsack problem, in which uncertainty arises from non-trivial perturbations to the decision variables that cannot be represented in closed form. We then proceed by describing a combination of exact optimization, simulation and a meta-heuristic that can be employed in such a setting. Specifically, a discrete-event simulation (DES) of the production system is developed to estimate solution quality and to model perturbations to the decision variables. A genetic algorithm (GA) can then be used to search for optimal production plans, using a simulation-based optimization approach. To provide effective seeding to the GA, we propose initialization operators that exploit mathematical programming in combination with the DES model. The approach is benchmarked against integer linear programming and chance-constrained programming. We find that our approach significantly outperforms contestant techniques under various levels of uncertainty.