基于Boosting-决策树C5.0的岩体结构面粗糙度预测

针对目前岩体结构面粗糙度系数（JRC）定量评价模型构建困难且预测精度较低的问题，搜集包括10条Barton标准剖面线在内的112条岩体结构面JRC，统计各剖面线的8种形态参数. 采用主成分分析降维处理形态参数，共得到5个主成分. 以前102组剖面线参数作为训练样本，采用Boosting-决策树C5.0算法构建模型，以10条Barton标准剖面线验证模型精度. 对比决策树C5.0模型、CHAID决策树模型、支持向量机（SVM）模型、类神经网络模型，分析各模型预测效果. 结果表明，Boosting-决策树C5.0模型的预测结果平均误差、均方根误差均最小. 建立的显式JRC预测模型，包含8层共计68节点的判别阈值.

Prediction of joint roughness coefficient of rock mass based on Boosting-decision tree C5.0

The JRC values of 112 rock joints were collected, including 10 Barton standard profiles, and 8 morphological parameters of each profile were calculated in view of the difficulty and low accuracy for prediction of the current joint roughness coefficient (JRC) quantitative evaluation model. Principal component analysis was used to reduce the dimension of these morphological parameters, and 5 principal components were obtained. 102 groups of profile data were used as training samples, the Boosting-decision tree (DT) C5.0 algorithm was used to build the training model, and Barton 10 standard profiles were used for model verification. The DT C5.0 model, CHAID DT model, support vector machine (SVM) model, and artificial neural network models were selected to verify the prediction accuracy of each model. Results showed that the average error and root mean square error of the Boosting-DT C5.0 model were the least. Established explicit JRC prediction model included 8 layers and 68 nodes.