Scholarlink

<p>The ocean mining riser represents the core equipment utilized for the commercial exploitation of deep-sea natural gas hydrates. It is subjected to a multitude of extreme environmental conditions, including ocean vortex excitation, the presence of large aspect ratios of structures, and the occurrence of internal multi-phase flow. These factors contribute to the generation of an exceptionally complex vibration mechanism. A machine learning model for predicting the nonlinear vibration of deep-sea hydrate mining riser is proposed by combining the Zebra Optimization Algorithm (ZOA) with the Random Forest (RF). The model utilizes chaos mapping, levy flight, and the random perturbation mutation strategies (CLR) to improve the performance of the ZOA and achieve real-time prediction of the vibration response of mining riser under multiple factors. The model effectively solves the problem of traditional mechanical modeling and numerical solution accuracy being greatly affected by complex external factors. Using the principle of similarity, a simulation experimental device for deep-sea hydrate mining riser vibration is developed, and multi factor vibration experiments are conducted to construct an experimental dataset. The proposed CLR-ZOA-RF machine learning model is used to predict the root mean square (RMS) displacement of in-line (IL) and cross flow (CF) direction vibration of the mining riser under multiple factors. The predicted results are compared with experimental test data, and the determination coefficient of the ZOA-RF model for IL direction vibration is 0.9359, and which for CF direction vibration is 0.9264. The CLR-ZOA-RF model has a determination coefficient of 0.9719 for IL direction vibration and 0.9667 for CF direction vibration. Compared with the ZOA-RF model, the CLR-ZOA-RF model has reduced the mean absolute error (MAE) and root mean square error (RMSE) of IL direction vibration prediction by 26.83% and 33.72%, respectively, and which of CF direction vibration prediction decreased by 45.15% and 29.49% respectively. Therefore, the research results indicated that the CLR-ZOA-RF model has high prediction accuracy and good generalization.</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

Machine-learning model of nonlinear vibration prediction for mining riser used into deep-sea hydrate exploitation based on CLR-ZOA-RF

  • 期刊论文

  • 发表日期：

    2025年4月26日

查看全文
  • Xiaoqiang Guo,
  • Junlin Lv,
  • Yingwei Li,
  • Yuanzhe Ren,
  • Ning Hu,
  • Yuxuan Song

摘要

The ocean mining riser represents the core equipment utilized for the commercial exploitation of deep-sea natural gas hydrates. It is subjected to a multitude of extreme environmental conditions, including ocean vortex excitation, the presence of large aspect ratios of structures, and the occurrence of internal multi-phase flow. These factors contribute to the generation of an exceptionally complex vibration mechanism. A machine learning model for predicting the nonlinear vibration of deep-sea hydrate mining riser is proposed by combining the Zebra Optimization Algorithm (ZOA) with the Random Forest (RF). The model utilizes chaos mapping, levy flight, and the random perturbation mutation strategies (CLR) to improve the performance of the ZOA and achieve real-time prediction of the vibration response of mining riser under multiple factors. The model effectively solves the problem of traditional mechanical modeling and numerical solution accuracy being greatly affected by complex external factors. Using the principle of similarity, a simulation experimental device for deep-sea hydrate mining riser vibration is developed, and multi factor vibration experiments are conducted to construct an experimental dataset. The proposed CLR-ZOA-RF machine learning model is used to predict the root mean square (RMS) displacement of in-line (IL) and cross flow (CF) direction vibration of the mining riser under multiple factors. The predicted results are compared with experimental test data, and the determination coefficient of the ZOA-RF model for IL direction vibration is 0.9359, and which for CF direction vibration is 0.9264. The CLR-ZOA-RF model has a determination coefficient of 0.9719 for IL direction vibration and 0.9667 for CF direction vibration. Compared with the ZOA-RF model, the CLR-ZOA-RF model has reduced the mean absolute error (MAE) and root mean square error (RMSE) of IL direction vibration prediction by 26.83% and 33.72%, respectively, and which of CF direction vibration prediction decreased by 45.15% and 29.49% respectively. Therefore, the research results indicated that the CLR-ZOA-RF model has high prediction accuracy and good generalization.