Scholarlink

<p>Due to water resource limitations and the environmental challenges associated with wastewater generated during oil and gas well drilling processes, the treatment and reuse of drilling wastewater have become essential. In Iran, most drilling wastewater treatment is conducted chemically using coagulant and flocculant agents, typically managed by on-site jar testing, which requires high technical expertise and can be time-consuming and prone to human error. Replacing this conventional approach with artificial intelligence techniques can significantly accelerate the process and reduce operational inaccuracies. In this study, data from 200 drilling waste management reports across various wells in the West Karun oilfields were collected, including input wastewater characteristics, dosages of polyaluminum chloride (coagulant) and polyacrylamide (flocculant), and the quality of the treated effluent. After conducting sensitivity analysis to select relevant input-output parameters, predictive models were developed using Recurrent Neural Networks (RNN), a hybrid PSO-RNN model, Extreme Learning Machines (ELMs), and Random Forest (RF). Each model was trained, tested, and validated, and their performance was evaluated using correlation coefficient (R) and root mean square error (RMSE). The validation results showed that for coagulant prediction, the RF model achieved the highest R value (0.89), while for flocculant prediction, the ELMs model outperformed others with an R value of 0.95. In terms of error, the ELMs model demonstrated the lowest RMSE values for both coagulant (0.13) and flocculant (0.10) predictions. ELM and Random Forest showed strong predictive performance (<i>R</i> ≈ 0.95, RMSE ≈ 0.10&#xa0;g/m³), with high NSE (&gt; 0.85) and low AIC (&lt; 110), confirming model robustness and stability through cross-validation. Overall, Among the four models tested, the ELMs model demonstrated relatively strong predictive performance in both coagulant and flocculant estimation tasks, though limitations in capturing extreme values remain.</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

Predictive modeling of coagulant dosing in drilling wastewater treatment using artificial neural networks

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年8月16日

查看全文
  • Mahyar Kalhormohammadi,
  • Sanaz Khoramipour

摘要

Due to water resource limitations and the environmental challenges associated with wastewater generated during oil and gas well drilling processes, the treatment and reuse of drilling wastewater have become essential. In Iran, most drilling wastewater treatment is conducted chemically using coagulant and flocculant agents, typically managed by on-site jar testing, which requires high technical expertise and can be time-consuming and prone to human error. Replacing this conventional approach with artificial intelligence techniques can significantly accelerate the process and reduce operational inaccuracies. In this study, data from 200 drilling waste management reports across various wells in the West Karun oilfields were collected, including input wastewater characteristics, dosages of polyaluminum chloride (coagulant) and polyacrylamide (flocculant), and the quality of the treated effluent. After conducting sensitivity analysis to select relevant input-output parameters, predictive models were developed using Recurrent Neural Networks (RNN), a hybrid PSO-RNN model, Extreme Learning Machines (ELMs), and Random Forest (RF). Each model was trained, tested, and validated, and their performance was evaluated using correlation coefficient (R) and root mean square error (RMSE). The validation results showed that for coagulant prediction, the RF model achieved the highest R value (0.89), while for flocculant prediction, the ELMs model outperformed others with an R value of 0.95. In terms of error, the ELMs model demonstrated the lowest RMSE values for both coagulant (0.13) and flocculant (0.10) predictions. ELM and Random Forest showed strong predictive performance (R ≈ 0.95, RMSE ≈ 0.10 g/m³), with high NSE (> 0.85) and low AIC (< 110), confirming model robustness and stability through cross-validation. Overall, Among the four models tested, the ELMs model demonstrated relatively strong predictive performance in both coagulant and flocculant estimation tasks, though limitations in capturing extreme values remain.