关键词: 静力触探试验, 参数预测, 随机森林, 支持向量回归, 神经网络

Abstract:

The construction of offshore wind power platforms urgently requires seabed geotechnical data to enhance the efficiency and accuracy of geological parameter prediction and modeling. Cone Penetration Testing (CPT) data offers unique advantages and plays a crucial role in geological parameter modeling for offshore wind farms. Traditional CPT-based prediction methods mainly utilize techniques such as Bayesian and Kriging interpolation. This study introduces Support Vector Regression (SVR), Random Forest (RF), and Neural Network (NN) algorithms into geological parameter prediction, combining the spatial continuity advantage of 2D seismic data with the vertical resolution advantage of CPT data to predict and model the CPT geotechnical parameters in shallow strata at the Eastern Offshore Wind Farm site in the Yinggehai Basin, South China Sea. The accuracy of the three methods is evaluated using error histograms and validation scatter plots. Results indicate that the Neural Network delivers the best overall performance, while the Support Vector Regression model yields simpler predictions. Due to the truncation nature of the Random Forest method, it yields the least accurate results, exhibiting abrupt horizontal variations. This study presents a novel research approach for predicting seabed geotechnical parameters.

Key words: CPT, parameter prediction, Random Forest, Support Vector Regression, Neural Networks