A Multi-objective Reinforcement Learning Algorithm for JSSP

Beatriz M Méndez-Hernández, Erick D Rodriguez-Bazan, Yailen Martinez-Jimenez, Pieter Libin, Ann Nowé

Research output: Chapter in Book/Report/Conference proceedingConference paper

7 Citations (Scopus)
1 Downloads (Pure)


Scheduling is a decision making process that takes care of the allocation of resources to tasks over time. The Job Shop scheduling problem is one of the most complex scheduling scenarios and is commonly encountered in manufacturing industries. Most of the existing studies are based on optimizing one objective, but in real-world problems, multiple criteria often need to be optimized at once. We propose a Multi-Objective Multi-Agent Reinforcement Learning Algorithm that aims to obtain the non-dominated solutions set for Job Shop scheduling problems. The proposed algorithm is used to solve a set of benchmark problems optimizing makespan and tardiness. The performance of our algorithm is evaluated and compared to other algorithms from the literature using two measures for evaluating the Pareto front. We show that our algorithm is able to find a set of diverse and high quality non-dominated solutions, that significantly and consistently improves upon the results obtained by other state-of-the-art algorithms.

Original languageEnglish
Title of host publicationInternational Conference on Artificial Neural Networks
Number of pages18
ISBN (Electronic)0302-9743
Publication statusPublished - 2019
EventInternational Conference on Artificial Neural Networks - Munchen, Germany
Duration: 17 Sep 201919 Sep 2019


ConferenceInternational Conference on Artificial Neural Networks


Dive into the research topics of 'A Multi-objective Reinforcement Learning Algorithm for JSSP'. Together they form a unique fingerprint.

Cite this