A novel hybrid model for long-term water quality prediction with the ‘decomposition–inputs–prediction’ hierarchical optimization framework

• Published in: Journal of hydroinformatics Vol. 26; no. 11; pp. 3008 - 3026
• Main Authors: Han, Jianjun, Wang, Lingling, Yao, Qiwen, Zhu, Dawei, Zhu, Hai, Xu, Jin, Hu, Lihan, Yang, Xu
• Format: Journal Article
• Language: English
• Published: London IWA Publishing 01.11.2024
• Subjects: Accuracy; Ammonia; Artificial intelligence; Chemical oxygen demand; Correlation analysis; Datasets; Decomposition; Deep learning; Environmental management; Long short-term memory; Mathematical models; Neural networks; Optimization; Oxygen requirement; Pollutants; Prediction models; Regression analysis; Simulation; Statistical analysis; Stream flow; Time series; Trends; Water pollution; Water quality; Wavelet analysis; Wavelet transforms
• ISSN: 1464-7141; 1465-1734
• DOI: 10.2166/hydro.2024.244

Accurate, stable, and long-term water quality predictions are essential for water pollution warning and efficient water environment management. In this study, a hierarchical water quality prediction (HWQP) model was developed based on ‘data decomposition–predictor screening–efficient prediction’ via wavelet decomposition, Spearman correlation analysis, and long short-term memory network, respectively. The observed data from 14 stations in the Huaihe River–Hongze Lake system, including ammonia nitrogen (AN) and chemical oxygen demand (COD), were used to make long-term water quality predictions. The results suggested that, compared to existing water quality prediction models, the HWQP model has higher accuracy, with the root mean square errors of 6 and 17% for simulating AN and COD, respectively. The AN and COD concentrations will range from 0 to 1 mg/l and from 3 to 5 mg/l at 12 stations, respectively, and the COD concentrations will exceed the water quality target at Stations 4 and 5. The established model has great potential to address the challenges associated with the water environment.