Skip to content

Official implementation of the Heterogeneous Graph Attention Networks for Scheduling in Cloud Manufacturing and Logistics

License

Notifications You must be signed in to change notification settings

AIRI-Institute/heterogat-scheduling-logistics

Repository files navigation

Heterogeneous Graph Attention Networks for Scheduling in Cloud Manufacturing and Logistics

This repository is the official implementation of methods from the paper Graph Attention Networks for Scheduling in Cloud Manufacturing and Logistics.

Overview

We investigate task scheduling in cloud manufacturing systems emphasizing logistics integration. We propose a graph neural network (GNN) architecture for optimizing task scheduling by representing the problem on a heterogeneous graph, where nodes denote tasks and locations.

An example of the problem graph (left) and the solution graph (right). The lower subgraph represents a task consisting of 4 subtasks to be scheduled. Dotted edges between subgraphs in the problem graph show available enterprises to perform operations. The encoders generate node embeddings, the dot product decoder predicts a probability of possible assignments. The solution is sampled from the predicted probabilities. Solid edges in the solution graph show an assigned enterprise for each subtask.

Requirements

To install dependencies, run

pip install -r requirements.txt

Running experiments

To run experiments on the baselines and the GNN models, run the following command:

python main.py

The results will be stored in the results.log file.

To train the GNN models from scratch, run the following command:

python train.py

Then replace the checkpoints in the folder checkpoints with the new ones stored in the folder training_gnn and run the experiments as described before.

Experimental results

The total cost (the lower the better) of solutions to the problems in the LMSC dataset.

Tasks-Operations-
Enterprises
No. Optimal
solution
Random
solution
Greedy
algorithm
Genetic
algorithm
GNN (Ours)
5-5-5 1 2648.63 3484.19 3495.44 2648.63 2780.38
2 5944.42 8704.79 8266.23 5944.42 5944.42
3 6653.88 8896.25 8868.77 6653.88 6653.88
5-10-10 1 5086.08 12599.87 6325.05 5145.03 5141.23
2 7352.66 18310.91 10974.05 7409.04 7352.66
3 7652.32 19556.74 12521.69 7714.85 7652.32
10-10-10 1 12221.76 33980.4 14849.09 12656.17 15306.85
2 14275.53 39574.01 21835.97 14322.24 14275.53
3 14200.75 40062.74 23082.48 14303.49 14200.75
5-10-20 1 4252.04 11542.44 6091.69 4289.83 4701.41
2 5497.29 18336.67 9377.16 5543.21 5497.29
3 5866.18 20916.08 9414.35 5925.18 5866.18
5-20-10 1 13512.96 36326.45 15446.19 13805.60 13921.15
2 15803.26 42108.59 19195.71 16137.08 16116.05
3 16703.56 45502.72 20396.13 17243.44 17491.83
5-20-20 1 11392.82 34663.67 14002.05 11875.55 13500.28
2 13045.16 42312.09 17038.35 13580.50 14550.33
3 14366.02 45445.61 18597.85 14899.99 16083.00

Please note that the results may differ slightly from the paper due to randomness.

Citation

Please cite the paper as follows:

@article{fomin2024heterogeneous,
  author={Fomin, Dmitrii and Makarov, Ilya and Voronina, Mariia and Strimovskaya, Anna and Pozdnyakov, Vitaliy},
  journal={IEEE Access}, 
  title={Heterogeneous Graph Attention Networks for Scheduling in Cloud Manufacturing and Logistics}, 
  year={2024},
  volume={12},
  pages={196195-196206},
  doi={10.1109/ACCESS.2024.3522020}
}

About

Official implementation of the Heterogeneous Graph Attention Networks for Scheduling in Cloud Manufacturing and Logistics

Resources

License

Stars

Watchers

Forks

Languages