SURROGATE ASSISTED GENE EXPRESSION PROGRAMMING FOR AUTOMATED DESIGN OF JOB SHOP SCHEDULING RULES

Abstract

The field of automated design of dispatching rules for manufacturing systems has received considerable attention from both academia and industry. Machine Learning (ML) approaches, especially Gene Expression Programming (GEP), have been successfully applied to evolve superior dispatching rules for dynamic job shop scheduling problems. However, the extensive computational requirement for evaluating a large number of candidate rules using discrete event simulation models remains one of the major challenges. In this paper, we propose a new surrogate-assisted GEP approach to reduce the computational costs of fitness assessment. The proposed approach shortens the simulation length by reducing the number of jobs to be processed, i.e., replacing part of the simulation length with a surrogate model. A subset of rules is evaluated using the full simulation length to generate the training data for the surrogate model (ML model). Afterward, the remaining rules are (partially) evaluated using the shortened simulation model, and then their absolute fitness values are estimated using the surrogate model. The performance of the proposed approach is evaluated under different job shop settings with respect to computational time and accuracy in predicting jobs mean flow time. The experiments have verified the effectiveness of the proposed approach compared with those in the literature.