基于机器学习的海上风电场浅部地层CPT参数预测方法

  • 李书兆 ,
  • 魏澈 ,
  • 申辰 ,
  • 孙国栋 ,
  • 杨叶涛 ,
  • 罗进华 ,
  • 王教龙
展开
  • 1. 中海石油(中国)有限公司北京研究中心, 北京 100028;

    2. 中国地质大学(武汉)地球物理与空间信息学院, 武汉 430074;

    3. 中海油田服务股份有限公司, 天津 300459

收稿日期: 2024-12-04

  修回日期: 2025-01-08

  录用日期: 2025-01-15

  网络出版日期: 2025-01-15

基金资助

国家“十四五”重大科技项目“新型能源开发及 CCUS 关键技术”(KJZX-2022-12-XNY-0100)

A machine Learning-Based method for predicting shallow subsurface CPT parameters in offshore wind farms

Expand
  • 1. CNOOC Research Institute Ltd., Beijing 100028, China;

    2. Institute of Geophysics & Geomatics, China University of Geosciences(Wuhan), Beijing 100049, China;

    3. China Oilfield Services Limited, Tianjin 300459, China

Received date: 2024-12-04

  Revised date: 2025-01-08

  Accepted date: 2025-01-15

  Online published: 2025-01-15

Supported by

China "14th Five Year Plan" Major Science and Technology Project "New Energy Development and CCUS Key Technologies"(KJZX-2022-12-XNY-0100) 

摘要

海上风电平台的建设,迫切需要利用海底的地质工程资料,提高地质参数预测与建模的计算效率和准确率。传统的地质参数预测主要采用贝叶斯、克里金插值等方法。本研究将支持向量回归、随机森林和神经网络算法引入地质参数预测领域,融合二维地震数据的空间连续性优势和静力触探数据的垂向分辨率优势,对南海莺歌海盆地的东方海上风电场浅部地层进行岩土参数预测与建模。利用误差直方图和验证散点图对三种方法的准确率进行比较,结果表明神经网络预测值综合性能较为优秀,支持向量回归模型预测结果较为简单,而随机森林由于其截断性,导致预测结果在水平方向出现了突变。本研究为海底地质岩土参数预测提供了崭新的思路。

本文引用格式

李书兆 , 魏澈 , 申辰 , 孙国栋 , 杨叶涛 , 罗进华 , 王教龙 . 基于机器学习的海上风电场浅部地层CPT参数预测方法[J]. 热带海洋学报, 0 : 1 . DOI: 10.11978/2024228

Abstract

Construction of offshore wind power platforms requires the use of geological engineering data from the seabed for geological parameter prediction and modeling. Traditional methods for predicting geological parameters include Bayesian and Kriging interpolation. With continuous advancements in computational power, machine learning has demonstrated outstanding performance in prediction tasks. This study introduces support vector regression, random forests, and neural network algorithms into the field of geological parameter prediction. Using 2D seismic data and CPT data, we predict geotechnical parameters for the Yinggehai Basin in the South China Sea. Finally, error histograms and validation scatter plots are used to compare the results of the three methods. It is found that neural network predictions exhibit the best overall performance, support vector regression results are relatively simple, and random forest predictions show abrupt changes in the horizontal direction due to truncation. These three methods provide new approaches for predicting submarine geological and geotechnical parameters.
文章导航

/