This respository includes codes and datasets for paper "Spatiotemporal information conversion machine for time-series forecasting".
The paper is online on Fundamental Research journal.
- A spatiotemporal information conversion machine (STICM), was developed to efficiently and accurately render a multistep-ahead prediction of a time series by employing a spatial-temporal information (STI) transformation.
- We put data file including the locations of traffic loops in folder datasets/traffic.
- The locations of all 155 meteorological stations used in wind speed dataset are provided in folder datasets/ws.
- Other dataset are uploaded to Google Drive, and can be downloaded here.
- python = 3.6
- tenforflow = 2.1
- cuda-version = 10.1
- cudnn-version = 7.6.5
-
We release the sample training codes and predicting codes corresponding to the Lorenz dataset, which is located at folder
experiment/multi_sample_symmetric/
. The scripttrain.py
is used for training and the scripteval.py
is used for evaluation after training the model. -
We can make predictions on other datasets by modify the given sample codes for Lorenz dataset.
- The example movie in terms of traffic speed prediction mentioned in our paper is given in folder example_movies/traffic_movie.mp4.
If you find this repository useful in your research, please consider citing the following papers:
@article{peng2022spatiotemporal,
title={Spatiotemporal information conversion machine for time-series forecasting},
author={Peng, Hao and Chen, Pei and Liu, Rui and Chen, Luonan},
journal={Fundamental Research},
year={2022},
publisher={Elsevier}
}