岩土体单桥静力触探曲线MCP-DD预测方法研究  

MCP-DD prediction method of single bridge static cone penetration curve of rock and soil

作　　者：高睿[1,2] 胡修任 王泽良 李畅 吴月秀[1] GAO Rui;HU Xiuren;WANG Zeliang;LI Chang;WU Yuexiu(School of Civil Engineering,Wuhan University,Wuhan Hubei 430072,China;Hubei Data-Driven Intelligent Construction Enterprise-School Joint Innovation Center,Wuhan University,Wuhan Hubei 430072,China)

机构地区：[1]武汉大学土木建筑工程学院,湖北武汉430072 [2]武汉大学湖北省数据驱动智能建造企校联合创新中心,湖北武汉430072

出　　处：《矿业科学学报》2025年第2期226-235,共10页Journal of Mining Science and Technology

摘　　要：工程中常采用静力触探获得地层信息,但因触探孔位相对稀疏,场地中存在大量未知区域,影响了设计和施工对地层信息的准确判断。提出一种岩土体单桥静力触探曲线的MCP-DD预测方法:①通过邻域半径搜索算法筛选相关数据点;②基于B样条基函数的改良MCP算法计算相关数据的趋势估计函数;③通过空间相关函数对趋势估计函数进行DD加权,得到综合预测模型。运用该方法预测某工程场地的单桥静力触探曲线,结果表明:相比于线性插值法,MCP-DD预测方法有更高的决定系数R^(2),平均绝对误差MAE减小26.8%~55.8%,均方根误差RMSE减小25.2%~54.9%。此外,预测半径R0的最佳取值范围为25.0~37.8 m,平均相关数据点个数越多,平均相对距离越小,模型预测效果越好。Static cone penetration test is often used in engineering practice to obtain relevant geological information.Yet the relatively sparse arrangement of test holes results in a large number of unknown areas in the site,which could affect the accurate judgment of the actual geological information in engineering design and construction.Therefore,this study proposes MCP-DD(minimax concave penaltydistance determined)prediction method for single bridge static cone penetration curve of rock and soil.It contains three main parts:1)screen neighboring training data through the neighborhood radius search algorithm;2)compute trend estimation functions of selected data through the modified MCP algorithm based on the B-spline basis function;3)weight the trend estimation function by the spatial correlation function to obtain an integrated prediction model.This method was used to predict the single bridge static cone penetration curve of a certain engineering site.Results show that compared with the traditional linear interpolation method,the proposed method exhibits a higher coefficient of determination R2.The mean absolute error MAE reduction reduced by 26.8%~55.8%and the root mean squared error RMSE reduces by 25.2%~54.9%.The optimal range for predicting the radius R^(0) is 25.0~37.8 m.Increasing average number of relevant data points leads to smaller average relative dis-tance between the predicted points and the relevant data points,thus bettering the prediction perform-ance of the model.

关 键 词：单桥静力触探 趋势估计函数 空间相关性 MCP-DD算法 

分 类 号：TU412[建筑科学—岩土工程] TD166[建筑科学—土工工程]