Performance comparison of search-based simulation optimisation algorithms for operations scheduling

This paper discusses the use of meta-heuristics coupled with discrete event simulations of various manufacturing systems to find the optimal operation schedules. Two search-based heuristic algorithms, OptQuest® (based on scatter search, tabu search and neural networks) and SimRunner® (based on genetic algorithm), are compared with respect to the quality of results and the computational time for a family of manufacturing system problems. The set of manufacturing systems configurations have been defined using the factors “type of shop” (flow shop and job shop), “number of part types” and “number of machines”. This family of problems is analysed based on the stochasticity of data, which is, using either deterministic or stochastic data for part inter-arrival times and processing times. A structured experiment has been conducted to test the responses of the two algorithms in optimising two different objective functions, maximising throughput rate and minimising percentage of tardy jobs. Arena® embedding OptQuest® and ProModel® embedding SimRunner® have been used in this research. Significant validation efforts have been made to ensure that simulation models built in Arena® and ProModel® are identical so that the performance difference only accrues from the heuristics. Evidence has been found to indicate that SimRunner® produced better results when the computation time is limited; however, OptQuest® produced comparable, sometimes superior results, when allowed infinite computation time.