Simulating and predicting the performance of a horizontal subsurface flow constructed wetland using a fully connected neural network

Date

2022

Authors

Li, P.
Zheng, T.
Li, L.
Lv, X.
Wu, W.J.
Shi, Z.
Zhou, X.
Zhang, G.
Ma, Y.
Liu, J.

Editors

Advisors

Journal Title

Journal ISSN

Volume Title

Type:

Journal article

Citation

Journal of Cleaner Production, 2022; 380(pt. 1, article no. 134959):1-11

Statement of Responsibility

Conference Name

Abstract

Constructed wetland systems, as an engineered ecological system, are being increasingly employed for wastewater treatment. However, owing to the complex incentives for pollutant removal in ecological treatment systems, it is challenging to simulate and optimize the operation of constructed wetlands to advance ecological wastewater treatment systems. In this study, a horizontal subsurface flow constructed wetland (HSCW) system was constructed and applied to a rural wastewater treatment system. Reeds (Phragmites australis) were planted in the HSCW to remove pollutants from the wastewater. Further, a fully connected neural network (FCNN) was designed based on the Adam optimization algorithm with weather conditions, quality, and quantity of influent and effluent as input to simulate and predict the performance of the HSCW. The results of the FCNN simulation analysis showed that the relative errors of the simulated concentrations of CODcr, NH4+-N, total nitrogen (TN), and total phosphorus (TP) for the FCNN model were 8.07 ± 10.73%, 18.34 ± 17.75%, 9.90 ± 11.91%, and 9.47 ± 10.98%, respectively. The mean absolute errors (MAEs) of CODcr, NH4+-N, TN, and TP for the FCNN model were 2.17, 1.06, 1.21, and 0.54, respectively. The root-mean-squared errors (RMSEs) of CODcr, NH4+-N, TN, and TP for the FCNN model were 3.91, 2.05, 2.22, and 0.80, respectively. The correlation coefficients (R2) of CODcr, NH4+-N, TN, and TP for the model were 0.99, 0.91, 0.92, and 0.82, respectively. These results indicate that the model performed well. Sensitivity analysis results also showed that temperature, solar radiation intensity, and rainfall had a strong impact on the model accuracy. This study verifies that an artificial neural network can effectively reflect the nonlinear function of each factor and is suitable for simulating HSCW treatment for wastewater under various conditions, providing a new optimization method for wastewater ecological treatment systems.

School/Discipline

Dissertation Note

Provenance

Description

Data source: Supplementary data, https://doi.org/10.1016/j.jclepro.2022.134959

Access Status

Rights

Copyright 2022 Elsevier

License

Grant ID

Call number

Persistent link to this record