Introduciton

Disclaimer on 2025 Mar 18:

Simulatior in this repository is vulnerable to job-overstay which leads to training instability and failure, and no longer maintained. Kindly consider using our latest sim model for your training and experiment: https://github.com/RK0731/DRLforDynamicScheduling/tree/main/src

Repository for the code of COR paper entitled A Deep Multi-Agent Reinforcement Learning Approach to Solve Dynamic Job Shop Scheduling Problem

Please consider citing our paper if you found it helpful ^-^

@article{LIU2023106294,
author = {Liu, Renke and Piplani, Rajesh and Toro, Carlos },
title = {A deep multi-agent reinforcement learning approach to solve dynamic job shop scheduling problem},
journal = {Computers & Operations Research},
volume = {159},
pages = {106294},
year = {2023},
issn = {0305-0548},
doi = {https://doi.org/10.1016/j.cor.2023.106294},
url = {https://www.sciencedirect.com/science/article/pii/S0305054823001582},
}

Repository Overview

This repo includes the code of simulation model, DRL learning algorithm, and sequencign rules for benchmark. Scripts' function can be identified by their prefix:

1. "asset_" under "src" directory: simulation model representing assets on shop floor, such as machines;
2. "brain_" under "src" directory: learning algorithm for job sequencing;
3. "event_" under "src/event_generator" directory: dynamic events on shop floor, such as job arrival;
4. "main_" in root directory: endpoints to start simulation processes, to train, or validate deep MARL-based algorithms;
5. "static_" under "src/static_experiment" directory: experiments on static problem instances.

Data and trained parameters can be found in folders:

1. "experiment_result": results of experiments in dynamic environments;
2. "GA_static_validation": static problem instances and experiment result on static problems;
3. "trained_models": the trained parameters.

User Guide

To use our code as the benchmark, kindly refer to "trained_models" folder for the trained parameters, and use the class "network_value_based" within "brain_sequencing.py" file to build the neural network. The state and action functions can also be found in "Brain_sequencing.py" file.

An alternative way is to test your approach in our simulation model and context, you may create your algorithm and run the simulation in "Main_experiment.py" for comparison. Please refer to the comments in each module to see how to interact with the simulation model.