Enhanced prediction of ammonia nitrogen levels in reverse osmosis brine from a full-scale water reclamation plant using machine learning
Water scarcity has led to the increased use of reclaimed water as a sustainable resource. However, reclaimed water generates byproducts like reverse osmosis brine, which can harm aquatic ecosystems if discharged directly. Effective monitoring and prediction of effluent quality are crucial. This stud...
Saved in:
| Main Authors: | , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Elsevier
2025-09-01
|
| Series: | Water Research X |
| Subjects: | |
| Online Access: | http://www.sciencedirect.com/science/article/pii/S2589914725000830 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Water scarcity has led to the increased use of reclaimed water as a sustainable resource. However, reclaimed water generates byproducts like reverse osmosis brine, which can harm aquatic ecosystems if discharged directly. Effective monitoring and prediction of effluent quality are crucial. This study focuses on a full-scale water reclamation plant (WRP) in Taiwan, which includes treatment units of both the wastewater treatment processes and the water reclamation processes, with comprehensive and detailed data across all treatment units—data rarely available in similar studies. Typical machine learning techniques, including shallow learning and deep learning, were systematically applied, along with automated machine learning frameworks, to predict ammonia nitrogen levels in reverse osmosis brine. The Long Short-Term Memory (LSTM) model outperformed other algorithms, achieving an R-squared value of 0.96 (mean absolute percent error (MAPE) of 2.6 %) with accurate peak level prediction using comprehensive treatment data and an R-squared value of 0.82 (MAPE of 20.5 %) using only influent wastewater quality and operational parameters. For one-day-ahead predictions, the LSTM model achieved an R-squared value of 0.64 (MAPE of 21.3 %). These findings demonstrate the potential of machine learning in full-scale WRP to provide accurate and early predictions, assisting plant operators in decision-making and reducing the risk of eutrophication in nearby rivers. |
|---|---|
| ISSN: | 2589-9147 |