基于静力触探与机器学习的打入桩竖向承载力预测

Prediction of Vertical Bearing Capacity of Driven Pile Based on Cone Penetration Test and Machine Learning

摘要:

分析了单桩承载力的主要影响因素，利用降维算法对静力触探数据进行预处理，使用树模型的重要性分析函数筛选主要特征，由贝叶斯优化算法确定超参数，提出了一种基于机器学习算法的打入桩竖向承载力预测模型。经过准度测试、蒙特卡洛模拟检验及与经验关系模型的对比，对预测模型的有效性和鲁棒性进行了评估。结果表明，支持向量机（SVM）模型、随机森林（RF）模型、极限梯度提升算法（XGBoost）模型均能较准确地预测打入桩竖向承载力，且准确性显著高于经验关系模型。根据综合分析结果，建议选择XGBoost模型用于打入桩竖向承载力的预测。

Abstract:

The main factors influencing the bearing capacity of a single pile were analyzed， the cone penetration test data were preprocessed using the dimensionality reduction algorithm， the main features were screened using the importance analysis function of tree model， and the hyperparameters were determined using the Bayesian optimization algorithm. A prediction model of vertical bearing capacity of driven pile was proposed based on the machine learning algorithm. The validity and robustness of the prediction model were evaluated using accuracy test and Monte Carlo simulation test， and compared with the empirical models. The results show that the support vector machine （SVM） model， the random forest （RF） model， and the extreme gradient boosting （XGBoost） model can accurately predict the vertical bearing capacity of driven piles， whose accuracy is significantly higher than that of the empirical relationship model. The comprehensive analysis indicates that the XGBoost model should be selected to predict the vertical bearing capacity of driven piles.