Deep reinforcement learning based proximal policy optimization algorithm for dynamic job shop scheduling

doi:10.1016/j.cor.2025.107149

Deep reinforcement learning based proximal policy optimization algorithm for dynamic job shop scheduling

2025-05-28

0

PRE

AI

DOI：10.1016/j.cor.2025.107149

Original

Original request for help

Save

Abstract

En 中文

A scheduling method combining an improved scheduling strategy with deep reinforcement learning (DRL) is proposed for the dynamic job-shop scheduling problem (DJSP) considering uncertainties such as machine failures and order changes. Firstly, the job shop data is preprocessed to obtain the corresponding system state, and secondly, the DJSP is defined as a Markov decision-making process, the job-shop model is established, and the reward function based on the scheduling goal is set to obtain the optimal scheduling strategy for each decision point by interacting with the environment. Then, a training method based on the proximal policy optimization algorithm is proposed, so as to realize the mapping of agents from production state to scheduling rules. Finally, the efficiency and effectiveness of the algorithm are proved by comparing with the intelligent optimization algorithm and scheduling rule on the test problem set.

Deep reinforcement learning based proximal policy optimization algorithm for dynamic job shop scheduling

Abstract

Journal

Organization