Effluent quality prediction of the sewage treatment based on a hybrid neural network model: Comparison and application

• Published in: Journal of environmental management 2024-02, Vol.351, p.119900-119900, Article 119900
• Main Authors: Wang, Zeyu, Dai, Hongliang, Chen, Beiyue, Cheng, Sichao, Sun, Yang, Zhao, Jinkun, Guo, Zechong, Cai, Xingwei, Wang, Xingang, Li, Bing, Geng, Hongya
• Format: Article
• Language: English
• Subjects: Artificial neural networks; Effluent quality prediction; Machine learning; Wastewater treatment plants
• ISSN: 0301-4797; 1095-8630
• DOI: 10.1016/j.jenvman.2023.119900

The accurate prediction and assessment of effluent quality in wastewater treatment plants (WWTPs) are paramount for the efficacy of sewage treatment processes. Neural network models have exhibited promise in enhancing prediction accuracy by simulating and analyzing diverse influent parameters. In this study, a back propagation neural network hybrid model based on a tent chaotic map and sparrow search algorithm (Tent_BP_SSA) was developed to predict the effluent quality of sewage treatment processes. The prediction performance of the propose hybrid model was compared with traditional neural network models using five performance indicators (MAE, RMSE, SSE, MAPE and R2). Specifically, in comparison with the prior Tent_BP_SSA, Tent_BP_SSA2 demonstrated notable enhancements, with the R2 increasing from 0.9512 to 0.9672, while MAE, RMSE, SSE, and MAPE decreased by 9.62%, 18.84%, 24.80%, and 47.10%, respectively. These indicators collectively affirm that the utilization of higher-order input parameters ensures improved accuracy of the Tent_BP_SSA2 hybrid model in predicting effluent quality. Moreover, the Tent_BP_SSA2 model exhibited robust prediction ability (R2 of 0.9246) when applied to assess the effluent quality of an actual sewage treatment plant. The incorporation of integrated models comprising the sparrow search optimizing algorithm, tent chaotic mapping, and higher-order magnitude decomposition of input parameters has demonstrated the capacity to enhance the accuracy of effluent quality prediction. This study illuminates novel perspectives on the prediction of effluent quality and the assessment of effluent warnings in WWTPs. [Display omitted] •The developed hybrid neural network model consisted of chaotic mapping and a sparrow search algorithm.•High order of magnitude input parameters and hybrid models demonstrated better predictive power.•The hybrid model was verified using Benchmark Simulation Model 1 and at an actual sewage treatment plant.•Accurate prediction of the effluent quality of the actual sewage treatment plant was obtained using the hybrid model.