基于LightGBM算法的海洋土压缩参数预测模型  

Prediction of marine soil compressibility based on LightGBM algorithm

作　　者：汪明元 王振红 陈松庭 WANG Mingyuan;WANG Zhenhong;CHEN Songting(Zhejiang Huadong Construction Engineering Co.,Ltd.,Hangzhou 310014,China;Power China Huadong Engineering Co.,Ltd.,Hangzhou 311122,China;College of Civil Engineering,Zhejiang University of Technology,Hangzhou 310023,China)

机构地区：[1]浙江华东建设工程有限公司,浙江杭州310014 [2]中国电建集团华东勘测设计研究院有限公司,浙江杭州311122 [3]浙江工业大学土木工程学院,浙江杭州310023

出　　处：《浙江工业大学学报》2024年第1期17-24,共8页Journal of Zhejiang University of Technology

摘　　要：近年来海洋工程项目不断增多,海洋岩土参数的确定对于保证工程安全性、提高经济效益有重要意义。目前主要通过室内与原位试验对土体参数进行研究,存在着成本高、效率低的问题。以某海上风电场项目勘察中所获取的海洋土压缩系数av和压缩模量Es为研究对象,选取样底深度、含水率、湿密度、土粒相对密度和液塑限等为基本变量,利用机器学习算法研究各类参数的相互关系和规律。构建一种基于LightGBM(Light gradient boosting machine)算法的土体压缩参数预测模型,通过k折交叉验证方法及贝叶斯优化改善预测性能。研究结果表明:笔者模型能够有效预测土体压缩参数,k折交叉验证方法及贝叶斯超参数优化能够提高预测准确度。此外,分析了k折次数、人工特征、数据集大小、预测量与输入特征相关性等因素对模型准确性的影响。The number of marine engineering projects has been increasing in recent years,and the determination of marine geotechnical parameters is of great significance to ensure project safety and to save costs.The soil parameters have to be measured by field and indoor tests,which is time and cost consuming.In this paper,the compression coefficient a v and the compression modulus E s of marine soil are studied,which are obtained from the survey of an offshore wind farm project.Some parameters are chosen as the basic variables,including the sample depth,the moisture content,the wet density,the soil particle specific gravity,and the liquid plastic limit.A machine learning algorithm is used to study the relationships between soil parameters.The aim of this study is to develop a simple,efficient tool for predicting the compressibility of marine soils.The LightGBM(Light gradient boosting machine)technique is employed to achieve this goal,and better prediction performance is obtained by k-fold cross-validation and Bayesian hyperparameter optimization.The evaluation results show that the proposed model can effectively predict the compressibility parameters and the k-fold cross-validation method and Bayesian optimization can improve the prediction accuracy.The influential factors are analyzed,including the number of folds,the artificial features,the size of datasets,the correlation between the prediction indicators and the feature values.

关 键 词：海洋土 参数估计 LightGBM模型 贝叶斯调参 

分 类 号：TU43[建筑科学—岩土工程]