Evolving Ensembles of Dispatching Rules Using Genetic Programming for Job Shop Scheduling

Job shop scheduling (JSS) problems are important optimisation problems that have been studied extensively in the literature due to their applicability and computational difficulty. This paper considers static JSS problems with makespan minimisation, which are NP-complete for more than two machines. Because finding optimal solutions can be difficult for large problem instances, many heuristic approaches have been proposed in the literature. However, designing effective heuristics for different JSS problem domains is difficult. As a result, hyper-heuristics (HHs) have been proposed as an approach to automating the design of heuristics. The evolved heuristics have mainly been priority based dispatching rules (DRs). To improve the robustness of evolved heuristics generated by HHs, this paper proposes a new approach where an ensemble of rules are evolved using Genetic Programming (GP) and cooperative coevolution, denoted as Ensemble Genetic Programming for Job Shop Scheduling (EGP-JSS). The results show that EGP-JSS generally produces more robust rules than the single rule GP.