Intelligent Scheduling Based on Reinforcement Learning Approaches: Applying Advanced Q-Learning and State–Action–Reward–State–Action Reinforcement Learning Models for the Optimisation of Job Shop Scheduling Problems

Abstract

Flexible job shop scheduling problems (FJSPs) have attracted significant research interest because they can considerably increase production efficiency in terms of energy, cost and time; they are considered the main part of the manufacturing systems which frequently need to be resolved to manage the variations in production requirements. In this study, novel reinforcement learning (RL) models, including advanced Q-learning (QRL) and RL-based state–action–reward–state–action (SARSA) models, are proposed to enhance the scheduling performance of FJSPs, in order to reduce the total makespan. To more accurately depict the problem realities, two categories of simulated single-machine job shops and multi-machine job shops, as well as the scheduling of a furnace model, are used to compare the learning impact and performance of the novel RL models to other algorithms. FJSPs are challenging to resolve and are considered non-deterministic polynomial-time hardness (NP-hard) problems. Numerous algorithms have been used previously to solve FJSPs. However, because their key parameters cannot be effectively changed dynamically throughout the computation process, the effectiveness and quality of the solutions fail to meet production standards. Consequently, in this research, developed RL models are presented. The efficacy and benefits of the suggested SARSA method for solving FJSPs are shown by extensive computer testing and comparisons. As a result, this can be a competitive algorithm for FJSPs.