Predicting the ammonia nitrogen of wastewater treatment plant influent via integrated model based on rolling decomposition method and deep learning algorithm. (July 2023)

• Record Type: Journal Article
• Authors: Yan, Kefen; Li, Chaolin; Zhao, Ruobin; Zhang, Yituo; Duan, Hengpan; Wang, Wenhui
• Is Part Of: Sustainable cities and society. Volume 94(2023)
• Publication Date: 2023-07
• Subjects: NH3-N prediction -- Rolling decomposition -- Decomposition-integrated model -- Wastewater treatment plant; Sustainable urban development -- Periodicals; Sustainable buildings -- Periodicals; Urban ecology (Sociology) -- Periodicals; 307.76
• DOI: 10.1016/j.scs.2023.104541
• Languages: English
• ISSNs: 2210-6707

Abstract

Highlights: A decomposition-integrated model was proposed to predict influent NH3 -N of WWTP. Rolling decomposition method ensures an information leakage-free modeling process. The proposed integrated model shows better performance compared with single model. The model outperforms the integrated models with information leakage in practice. Abstract: Timely and accurate assessment of key sewage quality indicators based on deep learning models has attracted much attention for intelligent wastewater treatment. Decomposition algorithms are widely adopted to further enhance the prediction performance of deep learning models, but traditional one-time decomposition method suffers from information leakage. Herein, an information leakage-free integrated model enabled by rolling decomposition method was built to predict influent ammonia nitrogen (NH3 -N), a key indicator for wastewater treatment. Firstly, the original NH3 -N sequence is decomposed into different subsequences using variational mode decomposition (VMD) under rolling method, which adds data successively for decomposition and excludes the future data, thus avoiding information leakage. Secondly, model and predict the subsequences by gated recurrent unit (GRU). Finally, summing the prediction results of the subsequences to obtain the prediction result of NH3 -N. The results show that the proposed model outperforms single GRU, with 16.69% lower RMSE, 13.02% lower MAE, and 11.90% lower MAPE. Moreover, the model also has advantages over the integrated model that trained under information leakage, with 42.34% lower RMSE, 41.06% lower MAE, and 39.34% lower MAPE. This work improves the applicability of integrated models and assists in the intelligent wastewater treatment and the construction of sustainable cities.