Scholarlink

<p>The dynamics of sewage treatment plants are characterized by complex, nonlinear processes that pose substantial challenges in modeling and controlling effluent quality to meet stringent water discharge standards and sustainability targets. This study develops and evaluates three supervised machine learning (ML) models—multilayer perceptrons (MLPs), support vector regression (SVR), and random forests (RF)—for predicting key nitrification-related nitrogen compounds: ammonium nitrogen (NH₄-N), nitrite nitrogen (NO<sub>2</sub>-N), total Kjeldahl nitrogen (TKN), and total nitrogen (TN). A total of 1729 daily operational records from Muharraq Sewage Treatment Plant (STP) in Bahrain (2018–2022) were utilized to train and validate the models using tenfold cross-validation. Results of the overall data show that the RF model consistently outperforms across all nitrogen species as it achieves lower mean absolute error, root-mean-square error values, and superior <i>R</i><sup>2</sup> compared to both MLP and SVR. Notably, RF models deliver best results for NH₄-N, NO<sub>2</sub>-N, TKN, and TN, achieving MAE values of 0.1631, 0.0398, 0.1936, and 0.2109, respectively, with corresponding <i>R</i><sup>2</sup> values of 0.86, 0.93, 0.89, and 0.92. Models were optimized using iterative hyperparameter tuning, and validation was performed across all folds to promote robustness and minimize bias. Gradient boosting machine (GBM) analysis ranked steady state design temperature (SSDT) as the most influential input feature for NH₄-N and TN, whereas influent flowrate (IF) has the highest importance for TKN and NO₂-N. The models offer plant operators data-driven insights for controlling effluent and limiting the need for laboratory testing, particularly in arid regions where nitrification kinetics is sensitive to temperature.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

AI-Driven Optimization Models for Prediction of Effluent Quality Parameters in a Wastewater Treatment Plant

  • 期刊论文

  • 发表日期：

    2025年9月13日

查看全文
  • Majeed S. Jassim,
  • Salman Hammad,
  • Gulnur Coskuner

摘要

The dynamics of sewage treatment plants are characterized by complex, nonlinear processes that pose substantial challenges in modeling and controlling effluent quality to meet stringent water discharge standards and sustainability targets. This study develops and evaluates three supervised machine learning (ML) models—multilayer perceptrons (MLPs), support vector regression (SVR), and random forests (RF)—for predicting key nitrification-related nitrogen compounds: ammonium nitrogen (NH₄-N), nitrite nitrogen (NO2-N), total Kjeldahl nitrogen (TKN), and total nitrogen (TN). A total of 1729 daily operational records from Muharraq Sewage Treatment Plant (STP) in Bahrain (2018–2022) were utilized to train and validate the models using tenfold cross-validation. Results of the overall data show that the RF model consistently outperforms across all nitrogen species as it achieves lower mean absolute error, root-mean-square error values, and superior R2 compared to both MLP and SVR. Notably, RF models deliver best results for NH₄-N, NO2-N, TKN, and TN, achieving MAE values of 0.1631, 0.0398, 0.1936, and 0.2109, respectively, with corresponding R2 values of 0.86, 0.93, 0.89, and 0.92. Models were optimized using iterative hyperparameter tuning, and validation was performed across all folds to promote robustness and minimize bias. Gradient boosting machine (GBM) analysis ranked steady state design temperature (SSDT) as the most influential input feature for NH₄-N and TN, whereas influent flowrate (IF) has the highest importance for TKN and NO₂-N. The models offer plant operators data-driven insights for controlling effluent and limiting the need for laboratory testing, particularly in arid regions where nitrification kinetics is sensitive to temperature.