Selection Schemes in Surrogate-Assisted Genetic Programming for Job Shop Scheduling

摘要

Designing effective dispatching rules is particularly important for dynamic job shop scheduling (JSS) problems. Recently, genetic programming (GP) and computer simulation have been combined to automatically design effective dispatching rules for different JSS problems. Although the literature has shown some success, expensive performance assessments or fitness evaluations still cause difficulty for design tasks, especially for very complicated and large-scale manufacturing systems. Therefore, it is important to effectively utilise the computational budget. The goal of this paper is to investigate the influence of surrogate models and the use of simulation replications on the performance of GP. The results show that the combination of the two techniques can enhance the quality of evolved dispatching rules. Analyses also show the advantages and disadvantages of different selection schemes in surrogate-assisted GP.