TED4STL: Trend-Error Decomposition For Self-Supervised Time Series Learning in Multivariate Forecasting Task

Datasets

The datasets are available at the following links:

Dataset	Link
ETT*	https://github.com/zhouhaoyi/ETDataset
Exchange	https://github.com/laiguokun/multivariate-time-series-data
WTH	https://drive.google.com/drive/folders/1ohGYWWohJlOlb2gsGTeEq3Wii2egnEPR
Electricity	https://archive.ics.uci.edu/ml/datasets/ElectricityLoadDiagrams20112014
Traffic	http://pems.dot.ca.gov/
Weather	https://www.bgc-jena.mpg.de/wetter/
Ili	https://gis.cdc.gov/grasp/fluview/fluportaldashboard.html

All datasets are provided in .csv format, except for the electricity dataset, which requires a preprocessing step. To preprocess it, please navigate to the datasets directory of each model and run the following command:

python3 preprocess_electricity.py

Run models

All the models require Python 3.11 and each has its requirements reported in the requirements.txt file in the model root directory. Please create a .venv environment for each model and install the requirements before running. To run the models move to the root directory of each of them and follow the instructions reported in the README.md file.

Statistics

To extract the statistics of the models, move to their root directory and run the following command:

python3 exctract_csv.py --directory forecasting/B<batch_size>_E<repr_dim>/ [--type raw]

Citation

@article{DBLP:journals/access/PederzoliBVG26,
  author       = {Sara Pederzoli and
                  Francesco Del Buono and
                  Maurizio Vincini and
                  Francesco Guerra},
  title        = {Trend-Error Decomposition for Self-Supervised Time Series Learning
                  in Multivariate Forecasting Task},
  journal      = {{IEEE} Access},
  volume       = {14},
  pages        = {8618--8631},
  year         = {2026}
}