徐志军, 赵世鹏, 王政权,等.基于GBDT算法的基桩竖向承载力预测方法[J].河南理工大学学报（自然科学版）,doi:10.16186/j.cnki.1673-9787.2022110051

XU Z J, ZHAO S P, WANG Z Q, et al.Prediction method of vertical bearing capacity of piles assisted with GBDT algorithm[J].Journal of Henan Polytechnic University(Natural Science) ,doi:10.16186/j.cnki.1673-9787.2022110051

基于GBDT算法的基桩竖向承载力预测方法(网络首发)

徐志军, 赵世鹏, 王政权, 田江涛, 宗飞龙

河南工业大学 土木工程学院，河南 郑州 450000

摘要:针对基桩竖向承载力的影响因素众多，且相互之间存在着复杂的非线性关系，导致基桩 竖向承载力难以准确预测的问题，采用机器学习中的梯度提升决策时（GBDT）算法，进行了基 桩竖向承载力的预测研究。首先，通过文献收集到174组基桩竖向承载力实测数据，建立了机 器学习预测模型需要的数据库。然后，采用GBDT算法建立基桩竖向承载力的预测模型。结 合均方误差（MSE）、平均绝对误差（MAE）、判定系数（R2）评价指标，将建立的预测模型与决策 树（DT）、随机森林（RF）、BP神经网络（BP）、线性回归（LR）模型，以及规范中的基桩竖向承载 力计算公式进行综合对比。研究结果表明：相对于DT模型、RF模型、BP模型、LR模型和规范 中的计算公式，建立的GBDT模型在测试集上的MAE、MSE最小，R2最大。结论是GBDT模型 具有准确度高、稳定性好、鲁棒性强，模型误差小的优势，能够更好地预测基桩竖向承载力。

关键词:基桩竖向承载力;梯度提升决策树;预测模型;评价指标;鲁棒性

doi:10.16186/j.cnki.1673-9787.2022110051

中图分类号：TU473.11

基金项目:国家自然科学基金资助项目（51978247）

收稿日期:2022-11-21

修回日期:2023-03-13

网络首发日期:2023-09-26

Prediction method of vertical bearing capacity of piles assisted with GBDT algorithm（Online）

XU Zhijun, ZHAO Shipeng, WANG Zhengquan, TIAN Jiangtao, ZONG Feilong

School of Civil Engineering，Henan University of Technology，Zhengzhou 450000，Henan，China

Abstract:Aiming at the problem that the vertical bearing capacity of piles is difficult to predict accurately due to the many factors affecting the vertical bearing capacity of piles and the complex nonlinear relationship between them，the gradient lifting decision tree （ GBDT） algorithm in machine learning is used to predict the vertical bearing capacity of piles.First of all，174 groups of measured data of vertical bearing capacity of foundation piles were collected from the literature，and the database required by the machine learning prediction model was established.Then，the GBDT algorithm is used to establish the prediction model of the vertical bearing capacity of the foundation pile. Combined with the three evaluation indicators of mean square error （MSE），mean absolute error （MAE） and judgment coefficient （R2），the prediction model established is comprehensively compared with the three models of decision tree （DT），BP neural network （BP） and linear regression （LR），as well as the calculation formula of vertical bearing capacity of piles in the specification. The research results show that compared with DT model，BP model，LR model and calculation formula in the specification，the established GBDT model has the smallest MAE and MSE in the testset and the largest R2.The conclusion is that GBDT model has the advantages of high accuracy，good stabil⁃ity，strong robustness and small model error，and can better predict the vertical bearing capacity of foundation piles.

Key words:vertical bearing capacity of piles;gradient boosting decision tree;prediction model;evaluation index;robustness

CLC：TU473.11