LPGNN

Code for the paper "Multivariate Time Series Prediction of Complex Systems Based on Graph Neural Networks with Location Embedding Graph Structure Learning"

This is a Pytorch implementation.

Requirements

• scipy>=1.7.0
• numpy>=1.21.0
• pandas>=1.2.5
• pyaml
• statsmodels
• pytorch>=1.8.0
• networkx>=2.6.3

Data Preparation

The traffic flow dataset has been placed in the dataset folder of the code. You need to unzip the dataset to this folder first. METR-LA and PEMS-BAY source and original paper of DCRNN. PEMSD4 and PEMSD8 come from the paper ASTGCN.

Model Training

You need to first specify the dataset name in main.py and then directly to run

python main.py

Please wait patiently for the program to finish running.

Result

Note: The Model is not designed for traffic flow prediction, but performs well on the traffic flow dataset. Baseline code: STGCN, DCRNN, MTGNN, GMAN, STGNN, GTS, Ada-STNet, STFGNN, GraphWaveNet, ASTGCN, STSGCN, AGCRN, Z-GCNETs, DSTAGNN, STG-NCDE.

| Dataset | METR-LA | | | PEMS-BAY | | | |--------------------|-------------|-------|-------|--------------|-------|-------| | Baseline | MAE |MAPE(%)| RMSE | MAE |MAPE(%)| RMSE | | STGCN | 4.45 | 11.8 | 8.41 | 2.49 | 5.69 | 5.79 | | DCRNN | 3.6 | 10.5 | 7.59 | 2.07 | 4.74 | 4.9 | | MTGNN | 3.49 | 9.87 | 7.23 | 1.94 | 4.53 | 4.49 | | GMAN | 3.48 | 10.1 | 7.3 | 1.86 | 4.32 | 4.31 | | STGNN | 3.49 | 9.69 | 6.94 | 1.83 | 4.15 | 4.2 | | GTS | 3.41 | 9.9 | 6.74 | 1.91 | 4.4 | 3.97 | | Ada-STNet | 3.47 | 9.8 | 7.18 | 1.89 | 4.5 | 4.36 | | STFGNN(SOTA) | 3.18 | 8.81 | 6.4 | 1.66 | 3.77 | 3.74 | | ours | 3.16 | 8.83 | 6.38 | 1.64 | 3.68 | 3.72 | | | | | | | | | | Dataset | PEMSD4 | | | PEMSD8 | | | | Baseline | MAE |MAPE(%)| RMSE | MAE |MAPE(%)| RMSE | | STGCN | 21.16 | 13.83 | 35.69 | 17.5 | 11.29 | 27.09 | | DCRNN | 21.22 | 14.17 | 37.23 | 16.82 | 10.92 | 26.36 | | GraphWaveNet | 28.15 | 18.52 | 39.88 | 20.3 | 13.84 | 30.82 | | ASTGCN | 22.93 | 16.56 | 34.33 | 18.25 | 11.64 | 28.06 | | MSTGCN | 23.96 | 14.33 | 37.21 | 19 | 12.38 | 29.15 | | STSGCN | 21.19 | 13.9 | 33.69 | 17.13 | 10.96 | 26.86 | | STFGNN | 19.83 | 13.02 | 31.88 | 16.64 | 10.6 | 26.22 | | AGCRN | 19.83 | 12.97 | 32.3 | 15.95 | 10.09 | 25.22 | | Z-GCNETs | 19.5 | 12.78 | 31.61 | 15.76 | 10.01 | 25.11 | | DSTAGNN | 19.3 | 12.7 | 31.46 | 15.67 | 9.94 | 24.77 | | STG-NCDE(SOTA) | 19.21 | 12.76 | 31.09 | 15.45 | 9.92 | 24.81 | | ours | 19.15 | 12.46 | 31.15 | 15.44 | 9.54 | 24.56 |

Citation

Shi, X., et al. (2022). "Multivariate time series prediction of complex systems based on graph neural networks with location embedding graph structure learning." Advanced Engineering Informatics 54: 101810.