A Multi-Scale Temporal Representation-Enhanced Informer for Wastewater Effluent Quality Prediction
Applied Sciences · 22.04.2026 · 16 (9) · 4078
DOI: 10.3390/app16094078 ↗Kurz gefasst
Fachartikel in der Zeitschrift Applied Sciences, online erschienen am 22.04.2026. Verfasst von Juan Wu und 3 weiteren Autor:innen. Die Arbeit wurde bisher 1-mal zitiert (Crossref) und nennt 49 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: National Key Research and Development Program of China.
Abstract (Originalsprache)
Accurate prediction of effluent water quality is essential for the intelligent and sustainable operation of wastewater treatment plants (WWTPs). However, this task remains challenging due to the strong nonlinearity, long-term temporal dependencies, and severe fluctuations inherent in influent characteristics. In this study, a novel data-driven framework termed the Multi-Scale Temporal Representation-Enhanced Informer (MTRE-Informer), is proposed to predict key effluent quality indicators, including total nitrogen (TN), total phosphorus (TP), and chemical oxygen demand (COD). To ensure data quality and computational efficiency, a generative recurrent learning framework is first employed for anomaly detection and correction, followed by variance inflation factor (VIF)-based feature selection to mitigate multicollinearity. Furthermore, feature contribution analysis is conducted to improve model interpretability. Subsequently, the core MTRE-Informer architecture utilizes hierarchical multi-scale temporal representation learning to simultaneously capture local patterns and long-term dependencies within the complex dynamics of the wastewater treatment process. Experimental results demonstrate that the MTRE-Informer achieves robust and stable predictive performance across diverse operational datasets. For TN prediction, the proposed framework attains a coefficient of determination (R2) of 0.9637 and a mean absolute percentage error (MAPE) of 3.39%. Compared with baseline approaches, the R2 improvement ranges from 3.8% to 14.2%, validating its superior capability. To further enhance model robustness, an anomaly detection and correction strategy based on a generative recurrent learning framework is employed. In addition, feature contribution analysis and VIF-based feature selection are conducted to improve interpretability, mitigate multicollinearity, and enhance computational efficiency. Overall, this framework provides a reliable and practical solution for real-time effluent quality prediction, facilitating the intelligent management of WWTPs.
Förderung
- National Key Research and Development Program of China
Zitieren
APA
Wu, J., Wu, Y., Liu, Y., & Zhang, X. (2026). A Multi-Scale Temporal Representation-Enhanced Informer for Wastewater Effluent Quality Prediction. Applied Sciences, 16(9), 4078. https://doi.org/10.3390/app16094078
Vancouver
Wu J, Wu Y, Liu Y, Zhang X. A Multi-Scale Temporal Representation-Enhanced Informer for Wastewater Effluent Quality Prediction. Applied Sciences. 2026;16(9):4078. doi:10.3390/app16094078
BibTeX
@article{Wu2026,
author = {Wu, Juan and Wu, Yifan and Liu, Yongze and Zhang, Xiaoyu},
title = {A Multi-Scale Temporal Representation-Enhanced Informer for Wastewater Effluent Quality Prediction},
journal = {Applied Sciences},
year = {2026},
volume = {16},
number = {9},
pages = {4078},
doi = {10.3390/app16094078}
}
EUVIMED