Scholarlink

<p>Sewage treatment plants (STPs) are essential for protecting water resources, yet their performance prediction remains challenging due to highly variable influent characteristics and complex operational conditions. To address these challenges, this study develops a data-driven framework based on Light Gradient Boosting (LGB) and Histogram Gradient Boosting (HGB) regression models combined with Fast Fourier Amplitude Sensitivity Test–based sensitivity analysis (FAST-SA) to predict key effluent indicators, ammonium nitrogen (NH₄_N) and total nitrogen (TN), using data from a full-scale STP. The proposed models effectively capture nonlinear relationships among multiple process variables, while FAST-SA efficiently identifies the most influential inputs with reduced computational demand compared to conventional sensitivity approaches. The optimized models demonstrate high predictive accuracy, achieving test coefficients of determination (R²) of up to 0.978 for NH₄_N and 0.984 for TN, and consistently low error metrics across validation and test phases, indicating strong generalization. Sensitivity analysis reveals that suspended solids (SS) and total organic carbon (TOC) are the dominant factors governing NH₄_N prediction, whereas dissolved oxygen (DO), SS, and mixed liquor suspended solids (MLSS) exert the greatest influence on TN prediction, providing valuable process-level insights. Overall, the proposed integrated modeling and sensitivity analysis framework offers accurate and interpretable predictions, supports real-time monitoring and operational decision-making, and contributes to more efficient and sustainable wastewater treatment management.</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

Identification of High-Impact features in the prediction of sewage treatment plant performance

  • 期刊论文

  • 发表日期：

    2026年2月20日

查看全文
  • Miliang Liu,
  • Qianhui Liu

摘要

Sewage treatment plants (STPs) are essential for protecting water resources, yet their performance prediction remains challenging due to highly variable influent characteristics and complex operational conditions. To address these challenges, this study develops a data-driven framework based on Light Gradient Boosting (LGB) and Histogram Gradient Boosting (HGB) regression models combined with Fast Fourier Amplitude Sensitivity Test–based sensitivity analysis (FAST-SA) to predict key effluent indicators, ammonium nitrogen (NH₄_N) and total nitrogen (TN), using data from a full-scale STP. The proposed models effectively capture nonlinear relationships among multiple process variables, while FAST-SA efficiently identifies the most influential inputs with reduced computational demand compared to conventional sensitivity approaches. The optimized models demonstrate high predictive accuracy, achieving test coefficients of determination (R²) of up to 0.978 for NH₄_N and 0.984 for TN, and consistently low error metrics across validation and test phases, indicating strong generalization. Sensitivity analysis reveals that suspended solids (SS) and total organic carbon (TOC) are the dominant factors governing NH₄_N prediction, whereas dissolved oxygen (DO), SS, and mixed liquor suspended solids (MLSS) exert the greatest influence on TN prediction, providing valuable process-level insights. Overall, the proposed integrated modeling and sensitivity analysis framework offers accurate and interpretable predictions, supports real-time monitoring and operational decision-making, and contributes to more efficient and sustainable wastewater treatment management.