Scholarlink

<p>This study optimized the chemical oxygen demand (COD) removal efficiency using machine learning models—multivariate adaptive regression splines (MARS) and the group method of data handling (GMDH). The key operational parameters considered included time, flow rate, COD, pH, volatile fatty acid (VFA), total suspended solids (TSS), hydraulic retention time (HRT), alkalinity, and organic loading rate (OLR). Datasets for model development were collected from the treatment of low-strength synthetic and real domestic wastewater using an upflow anaerobic sludge blanket (UASB) reactor, followed by a coagulation‒flocculation process in a clariflocculator unit. The effluent from the UASB reactor, containing nutrients such as phosphorus that contribute to eutrophication, was further treated with a clariflocculator. Water treatment sludge (WTS) was used as a coagulant. A laboratory-scale UASB reactor and clariflocculator were interconnected to establish an integrated treatment system. Synthetic wastewater with an average COD of 329.72 ± 35.86&#xa0;mg/L was prepared, whereas real domestic wastewater—with an average COD of 181.65 ± 25.32&#xa0;mg/L—was collected from the inlet chamber of the wastewater treatment plant. The system was operated continuously in two phases: synthetic wastewater was treated from days 1-274, followed by actual wastewater from days 275–409 and monitoring throughout. The integrated system improved the COD removal efficiency from 74.88% to 82.44% and improved the phosphate removal efficiency from 8.63% to 74.41%. MARS outperformed GMDH, with an RMSE of 0.040, an MAE of 0.032, an R² of 0.955 during training, and comparable testing (RMSE of 0.062, an MAE of 0.049, and an R² of 0.954).</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

Optimizing COD removal using machine learning in a UASB-Clariflocculator wastewater treatment system

  • 期刊论文

  • 发表日期：

    2026年2月25日

查看全文
  • Saurabh Kumar,
  • Divesh Ranjan Kumar,
  • Abdur Rahman Quaff,
  • Warit Wipulanusat

摘要

This study optimized the chemical oxygen demand (COD) removal efficiency using machine learning models—multivariate adaptive regression splines (MARS) and the group method of data handling (GMDH). The key operational parameters considered included time, flow rate, COD, pH, volatile fatty acid (VFA), total suspended solids (TSS), hydraulic retention time (HRT), alkalinity, and organic loading rate (OLR). Datasets for model development were collected from the treatment of low-strength synthetic and real domestic wastewater using an upflow anaerobic sludge blanket (UASB) reactor, followed by a coagulation‒flocculation process in a clariflocculator unit. The effluent from the UASB reactor, containing nutrients such as phosphorus that contribute to eutrophication, was further treated with a clariflocculator. Water treatment sludge (WTS) was used as a coagulant. A laboratory-scale UASB reactor and clariflocculator were interconnected to establish an integrated treatment system. Synthetic wastewater with an average COD of 329.72 ± 35.86 mg/L was prepared, whereas real domestic wastewater—with an average COD of 181.65 ± 25.32 mg/L—was collected from the inlet chamber of the wastewater treatment plant. The system was operated continuously in two phases: synthetic wastewater was treated from days 1-274, followed by actual wastewater from days 275–409 and monitoring throughout. The integrated system improved the COD removal efficiency from 74.88% to 82.44% and improved the phosphate removal efficiency from 8.63% to 74.41%. MARS outperformed GMDH, with an RMSE of 0.040, an MAE of 0.032, an R² of 0.955 during training, and comparable testing (RMSE of 0.062, an MAE of 0.049, and an R² of 0.954).