Learning to Adapt Genetic Algorithms for Multi-Objective Flexible Job Shop Scheduling Problems

Abstract

The configuration of Evolutionary Algorithm (EA) parameters is a significant challenge. While previous studies have examined methods for configuring EA parameters, there remains a lack of a general solution for optimizing these parameters. To overcome this, we propose DEMOCA, an automated Deep Reinforcement Learning (DRL) method for online control of multi-objective EA parameters. When tested on a multi-objective Flexible Job Shop Scheduling Problem (FJSP) using a Genetic Algorithm (GA), DEMOCA was found to be as effective as grid search while requiring significantly less training cost.