Scholarlink

<p>Conventional desalination plants discharge high-salinity wastewater into the environment, disrupting natural salt balances and exacerbating soil salinity. Zero-discharge desalination (ZDD) plants have emerged as a promising alternative, aiming to recover salt and water rather than discharging saline wastewater. This study focuses on predicting the performance of wastewater pretreatment, a crucial step in ZDD plants, through both experimental and computational approaches. Experimentally, a pilot plant was utilized to investigate the influence of coagulant dosage, mixing rate, and additive concentrations on effluent quality. Computationally, three modeling techniques—artificial neural networks (ANNs), adaptive neuro-fuzzy inference systems (ANFIS), and a tree-based Random Forest model—were applied to predict effluent parameters including total hardness, CO₂ content, and electrical conductivity. While experimental trials identified effective pretreatment strategies, modeling extended these results by enabling process generalization, sensitivity analysis, and variable importance ranking. The integration of experimental and computational methods provides a comprehensive understanding of wastewater pretreatment in ZDD plants, paving the way for optimizing treatment processes and achieving sustainable desalination practices. Among the tested models, ANN achieved the highest predictive accuracy (R² &gt; 0.97), outperforming ANFIS (R² = 0.94–0.96) and Random Forest, although ANFIS provided interpretable fuzzy rules that highlight the relative influence of dosing and mixing parameters.</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

Experimental Optimization and Machine Learning Modeling of Desalination Wastewater Pretreatment Using ANN, ANFIS, and Random Forest

  • 期刊论文

  • 发表日期：

    2025年11月13日

查看全文
  • Farshad Farahbod,
  • Mohammad Afkhami Karaei

摘要

Conventional desalination plants discharge high-salinity wastewater into the environment, disrupting natural salt balances and exacerbating soil salinity. Zero-discharge desalination (ZDD) plants have emerged as a promising alternative, aiming to recover salt and water rather than discharging saline wastewater. This study focuses on predicting the performance of wastewater pretreatment, a crucial step in ZDD plants, through both experimental and computational approaches. Experimentally, a pilot plant was utilized to investigate the influence of coagulant dosage, mixing rate, and additive concentrations on effluent quality. Computationally, three modeling techniques—artificial neural networks (ANNs), adaptive neuro-fuzzy inference systems (ANFIS), and a tree-based Random Forest model—were applied to predict effluent parameters including total hardness, CO₂ content, and electrical conductivity. While experimental trials identified effective pretreatment strategies, modeling extended these results by enabling process generalization, sensitivity analysis, and variable importance ranking. The integration of experimental and computational methods provides a comprehensive understanding of wastewater pretreatment in ZDD plants, paving the way for optimizing treatment processes and achieving sustainable desalination practices. Among the tested models, ANN achieved the highest predictive accuracy (R² > 0.97), outperforming ANFIS (R² = 0.94–0.96) and Random Forest, although ANFIS provided interpretable fuzzy rules that highlight the relative influence of dosing and mixing parameters.