Transformer network for time series prediction via wavelet packet decomposition
Time series predictions are commonly used in the fields of energy, meteorol-ogy, and finance, among others. The accurate prediction of time series data iscritical for making decisions and planning. In the real world, non-stationarytime series data with statistical properties shift over time, making...
Saved in:
| Main Authors: | , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Electronics and Telecommunications Research Institute (ETRI)
2025-08-01
|
| Series: | ETRI Journal |
| Subjects: | |
| Online Access: | https://doi.org/10.4218/etrij.2024-0013 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1849224635430731776 |
|---|---|
| author | Zhichao Wu Aiye Shi Yan Ping Tao |
| author_facet | Zhichao Wu Aiye Shi Yan Ping Tao |
| author_sort | Zhichao Wu |
| collection | DOAJ |
| description | Time series predictions are commonly used in the fields of energy, meteorol-ogy, and finance, among others. The accurate prediction of time series data iscritical for making decisions and planning. In the real world, non-stationarytime series data with statistical properties shift over time, making predictionmore challenging. Although, conventional time series processing methods—such as multi-scale feature extraction or Transformer-based algorithms—produce superior prediction results, when dealing with data that contain morenoise and outliers, the prediction ability of such methods can suffer. Toaddress this problem, we proposed the WPFormer model, which incorporatedtime-frequency analysis into the Transformer architecture to increase thelong-term series prediction accuracy. The model employed wavelet packetdecomposition to identify and eliminate noise efficiently, increasing its immu-nity to interference. We evaluated WPFormer on four publicly available data-sets and compared its performance against the Informer, LogTrans, Reformer,LSTMa, LSTNet, and DeepAR models using MSE and MAE metrics. On aver-age, the WPFormer model surpassed the benchmark models by 16%. |
| format | Article |
| id | doaj-art-3d2f5b8896584ba581077381dde3dc0d |
| institution | Kabale University |
| issn | 1225-6463 2233-7326 |
| language | English |
| publishDate | 2025-08-01 |
| publisher | Electronics and Telecommunications Research Institute (ETRI) |
| record_format | Article |
| series | ETRI Journal |
| spelling | doaj-art-3d2f5b8896584ba581077381dde3dc0d2025-08-25T07:01:11ZengElectronics and Telecommunications Research Institute (ETRI)ETRI Journal1225-64632233-73262025-08-0147467268410.4218/etrij.2024-0013Transformer network for time series prediction via wavelet packet decompositionZhichao WuAiye ShiYan Ping TaoTime series predictions are commonly used in the fields of energy, meteorol-ogy, and finance, among others. The accurate prediction of time series data iscritical for making decisions and planning. In the real world, non-stationarytime series data with statistical properties shift over time, making predictionmore challenging. Although, conventional time series processing methods—such as multi-scale feature extraction or Transformer-based algorithms—produce superior prediction results, when dealing with data that contain morenoise and outliers, the prediction ability of such methods can suffer. Toaddress this problem, we proposed the WPFormer model, which incorporatedtime-frequency analysis into the Transformer architecture to increase thelong-term series prediction accuracy. The model employed wavelet packetdecomposition to identify and eliminate noise efficiently, increasing its immu-nity to interference. We evaluated WPFormer on four publicly available data-sets and compared its performance against the Informer, LogTrans, Reformer,LSTMa, LSTNet, and DeepAR models using MSE and MAE metrics. On aver-age, the WPFormer model surpassed the benchmark models by 16%.https://doi.org/10.4218/etrij.2024-0013long-time predictiontime series predictiontime-frequency analysistransformerwaveletpacket decomposition |
| spellingShingle | Zhichao Wu Aiye Shi Yan Ping Tao Transformer network for time series prediction via wavelet packet decomposition ETRI Journal long-time prediction time series prediction time-frequency analysis transformer waveletpacket decomposition |
| title | Transformer network for time series prediction via wavelet packet decomposition |
| title_full | Transformer network for time series prediction via wavelet packet decomposition |
| title_fullStr | Transformer network for time series prediction via wavelet packet decomposition |
| title_full_unstemmed | Transformer network for time series prediction via wavelet packet decomposition |
| title_short | Transformer network for time series prediction via wavelet packet decomposition |
| title_sort | transformer network for time series prediction via wavelet packet decomposition |
| topic | long-time prediction time series prediction time-frequency analysis transformer waveletpacket decomposition |
| url | https://doi.org/10.4218/etrij.2024-0013 |
| work_keys_str_mv | AT zhichaowu transformernetworkfortimeseriespredictionviawaveletpacketdecomposition AT aiyeshi transformernetworkfortimeseriespredictionviawaveletpacketdecomposition AT yanpingtao transformernetworkfortimeseriespredictionviawaveletpacketdecomposition |