Genetic Programming with Pareto Local Search for Many-Objective Job Shop Scheduling

Abstract

Genetic programming (GP) has been successfully used to automatically design effective dispatching rules for job shop scheduling (JSS) problems. It has been shown that hybridizing global search with local search can significantly improve the performance of many evolutionary algorithms such as GP because local search can directly improve the exploitation ability of these algorithms. Inspired by this, we aim to enhance the quality of evolved dispatching rules for many-objective JSS through hybridizing GP with Pareto Local Search (PLS) techniques. There are two challenges herein. First, the neighborhood structure in GP is not trivially defined. Second, the acceptance criteria during the local search for many-objective JSS has to be carefully designed to guide the search properly. In this paper, we propose a new algorithm that seamlessly integrates GP with Pareto Local Search (GP-PLS). To the best of our knowledge, it is the first time to combine GP with PLS for solving many-objective JSS. To evaluate the effectiveness of our new algorithm, GP-PLS is compared with the GP-NSGA-III algorithm, which is the current state-of-the-art algorithm for many-objective JSS. The experimental results confirm that the newly proposed method can outperform GP-NSGA-III thanks to the proper use of local search techniques. The sensitivity of the PLS-related parameters on the performance of GP-PLS is also experimentally investigated.