Scholarlink

To minimize the risk of marine pollution due to the discharge of the produced water in offshore oil and gas production, many operators have installed online Oil-in-Water (OiW) monitors along with their produced water treatment (PWT) processes. However, these measurements have not been integrated with the PWT control systems due to a lack of proper quantitative models to be able to describe the key separation dynamics related to these OiW measurements, particularly toward the de-oiling hydrocyclone-based PWT processes. In contrast, these models are unavoidable for design of an OiW-based PWT feedback control. This paper investigated the modeling of the de-oiling hydrocyclone system’s separation dynamics by proposing a combined configuration of Long Short-Term Memory (LSTM) and Convolutional Neural Networks (CNN), based on the data retrieved from a pilot plant. To improve network performance, the network hyper-parameters optimization problem is given and solved using a generic Algorithm (GA). The prediction performances and training efficiencies of the proposed models are also discussed. The obtained results show that the GA-CNN-LSTM model exhibits the best prediction performance in terms of RMSE of 0.009, MAE of 0.007, and \(R^2\) of 0.975. This provides a promising opportunity to support the development of an GA-CNN-LSTM based OiW control solution in the next step.

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

Monitoring and Prediction of the De-oiling Dynamic Efficiency of Offshore Produced Water Treatment Using Deep-Learning Models

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Mahsa Kashani,
  • Stefan Jespersen,
  • Zhenyu Yang

摘要

To minimize the risk of marine pollution due to the discharge of the produced water in offshore oil and gas production, many operators have installed online Oil-in-Water (OiW) monitors along with their produced water treatment (PWT) processes. However, these measurements have not been integrated with the PWT control systems due to a lack of proper quantitative models to be able to describe the key separation dynamics related to these OiW measurements, particularly toward the de-oiling hydrocyclone-based PWT processes. In contrast, these models are unavoidable for design of an OiW-based PWT feedback control. This paper investigated the modeling of the de-oiling hydrocyclone system’s separation dynamics by proposing a combined configuration of Long Short-Term Memory (LSTM) and Convolutional Neural Networks (CNN), based on the data retrieved from a pilot plant. To improve network performance, the network hyper-parameters optimization problem is given and solved using a generic Algorithm (GA). The prediction performances and training efficiencies of the proposed models are also discussed. The obtained results show that the GA-CNN-LSTM model exhibits the best prediction performance in terms of RMSE of 0.009, MAE of 0.007, and \(R^2\) of 0.975. This provides a promising opportunity to support the development of an GA-CNN-LSTM based OiW control solution in the next step.