WaterTS: Integrating Enhanced Transformer, Sliding Block, and Channel Independence for Long-term Water Quality Prediction
Document Type
Conference Proceeding
Publication Date
1-1-2024
Abstract
Nowadays, the deterioration of water resources leads to negative ecological impacts. To effectively inhibit the deterioration of water resources, a water quality prediction model based on enhanced transformer, sliding block, and channel independence (WaterTS) is proposed by comprehensively analyzing the indicators of water resources and making long-term predictions of the dissolved oxygen index. WaterTS adopts a sliding block method to extract the short-term temporal features of the water quality series and combine them with channel independence to make independent predictions of multi-featured data. Moreover, it upgrades the internal encoder structure of the transformer and improves the attention mechanism to Probsparse-attention and Auto-Correlation to speed up the prediction speed. Furthermore, Post LayerNormal is adjusted to Pre LayerNormal, which makes the training gradient more stable and enhances the accuracy of predictions. Experiments are conducted using real-world water environment data, and comparison results with state-of-the-art prediction models show that the WaterTS achieves accurate predictions on both short-term and long-term water quality data.
Identifier
85208228971 (Scopus)
ISBN
[9798350358513]
Publication Title
IEEE International Conference on Automation Science and Engineering
External Full Text Location
https://doi.org/10.1109/CASE59546.2024.10711632
e-ISSN
21618089
ISSN
21618070
First Page
270
Last Page
275
Grant
L233005
Fund Ref
Natural Science Foundation of Beijing Municipality
Recommended Citation
Bi, Jing; Xu, Lifeng; Wang, Ziqi; Yuan, Haitao; Chen, Shichao; Gu, Mu; and Zhou, Meng Chu, "WaterTS: Integrating Enhanced Transformer, Sliding Block, and Channel Independence for Long-term Water Quality Prediction" (2024). Faculty Publications. 823.
https://digitalcommons.njit.edu/fac_pubs/823
