基于随机森林方法的岩石节理粗糙度系数研究  

Method to Calculate the Joint Roughness Coefficient Based on Random Forest

作　　者：宋康明 姜阳厚 谭志祥[2] 周才文 陈锐[3] 朱冬丽 Song Kangming;Jiang Yanghou;Tan Zhixiang;Zhou Caiwen;Chen Rui;Zhu Dongli(Guangzhou Urban Planning Survey Design and Research Insitute, Guangzhou 510000, China;School of Environment and Survey and Mapping, China University of Mining and Technology, Xuzhou Jiangsu 221116, China;School Architectural and Surveying of Mapping Engineering, Jiangxi University of Science and Technology, Ganzhou Jiangxi 341000, China)

机构地区：[1]广州市城市规划勘测设计研究院,广州510000 [2]中国矿业大学环境与测绘学院,江苏徐州221116 [3]江西理工大学建筑与测绘工程学院,江西赣州341000

出　　处：《地质科技情报》2018年第3期263-267,共5页Geological Science and Technology Information

基　　金：国家自然科学基金项目(41272389);江苏高校优势学科建设工程项目(SZBF2011-6-B35)

摘　　要：岩石节理粗糙度系数(JRC)是研究岩石力学的重要参数之一。为精确有效地描述JRC,提出了一种基于随机森林(Random forest,RF)算法研究JRC的新方法。首先,详细叙述了RF算法的原理和实现流程;然后,简要分析了影响JRC的一些统计参数,确定了7个重要的基本变量,即节理表面最大峰高Sp、表面最大高度Sz、表面最大谷深Sv、峰度系数Sku、偏斜度系数Ssk、均方根高度Sq、算数平均高度Sa;最后,结合R语言构建了一种RF回归预测分形维数D和JRC值的模型,其中用于训练和测试RF回归模型的样本资料源于某高校的实测数据。用6组实测数据对训练后的RF回归模型进行了测试,试验结果表明:(1)利用RF回归模型预计的D值、JRC值与实测值的最大相对误差仅为3.844%、4.553%。(2)RF回归模型具有较强的泛化能力,需要考虑的模型参数少,预测精度高,为今后继续研究D值和JRC值提供了一种新思路。Rock joint roughness coefficient is one of the important parameters for studyig the rock shear strength.The coefficient is influenced by many complex factors and thus is difficult to describe by using some expressions.In order to calculate the JRCeffectively and accurately,a new method of JRCbased on random forest（RF）algorithm is proposed.Firstly,the basic principle of RF algorithm and its implementation process are narrated.Then,statistical parameters that may affect the value change of the JRCare discussed.The seven variables are maximum peak high Sp,surface and the maximum height Sz,maximum surface valley deep Sr,kurtosis coefficient Sku,partial slope coefficient Ssk,square highly Sqand arithmetic mean height Sa.Finally,the RF regression model is established based on R language to predict the fractal dimension（D）and the JRCvalue.Sample data for training and testing the RF regression model are obtained from a college.The test results of the trained RF regression model by using 6 sets of measured data show that the maximum relative error of the measured data and the prediction results of the D and JRC based on the RF regression respectively are 3.844%,4.553%.The RF regression model can be generalized with fewer model parameters to be considered so as to achieve a high prediction accuracy in calculating the D and JRC.Above all,the RF regression model provides a new way to continue studying the D and JRC.

关 键 词：决策树 随机森林 R语言 岩石节理粗糙度系数 分形维数 

分 类 号：P554[天文地球—构造地质学]