基于支持向量机回归的岩体变形模量预测  

Prediction of Rock Deformation Modulus Based on Support Vector Machine Regression

作　　者：刘超 唐锡彬 邓冬梅 孙自豪[2] 姜耀飞 邓姗 Liu Chao;Tang Xibin;Deng Dongmei;Sun Zihao;Jiang Yaofei;Deng Shan(Guizhou Electric Power Design and Research Institute,Guiyang 550002,China;Faculty of Engineering,China University of Geosciences（Wuhan）,Wuhan 430074,China)

机构地区：[1]贵州电力设计研究院,贵阳550002 [2]中国地质大学(武汉)工程学院,武汉430074

出　　处：《地质科技情报》2018年第5期275-280,共6页Geological Science and Technology Information

基　　金：建设部科技项目(2015-K2-008)

摘　　要：岩体变形模量是表征岩体变形特性的最重要参数之一,其获取手段有室内试验与现场试验法、经验关系法、数值模拟法、人工智能预测方法等。人工智能预测方法中常用的是神经网络方法,但神经网络易陷入局部极小值和过学习而导致低精度,支持向量机回归(SVR)方法能有效地避免神经网络的以上缺陷,并在小样本、非线性预测方面具有较大优势,但目前SVR应用于岩体变形模量预测的研究较少。以某水电站坝址区英安岩的试验数据为依托,采用灰色关联分析筛选出与变形模量最相关的纵波波速作为输入变量。在此基础上,以3个国内的水电站为例,分别建立相应的以实测纵波波速作为输入变量的粒子群算法优化-支持向量机回归(PSO-SVR)变形模量预测模型,同时,通过与BP神经网络(BP-NN)、RBF神经网络(RBF-NN)2种预测方法进行对比,对比分析表明SVR模型具有更高的预测精度,预测效果较好,说明PSO-SVR方法更适用于岩体变形模量预测。The deformation modulus is one of the most important parameters for characterization of rock mass deformation.It can be obtained by laboratory and field experiments,empirical relation method,numerical simulation method and intelligent nonlinear methods.In these methods,the artificial neural network method is the most widely used intelligent nonlinear method,such as back propagation neural network（RBF-NN）and radial basis functionneural network（RBF-NN）.But the structure of neural network is difficult to determine,and the disadvantage of local minimization and over learning leads to low prediction accuracy.While support vector machine regression（SVR）method has more advantages in small samples and nonlinear prediction,but there are few researches in prediction of rock mass modulus with SVR method currently.With experimental data from dacite of a hydropower station dam area,this paper selects the optimal input variable using grey relational analysis（GRA）as the input variables of particle swarm optimization（PSO）algorithm optimized SVR model to predict the deformation modulus of rock mass.In order to reveal its supremacy of the PSO-SVR model over its comparison objects,i.e.,BP-NN,RBF-NN models,in-situ test data of longitudinal wave velocity and deformation modulus from 3 hydropowers are obtained to develop the corresponding PSO-SVR model,BP-NN model and RBF-NN model.It turns out that the PSO-SVR prediction model has higher prediction accuracy,and is more suitable for the prediction of rock mass deformation modulus.The simulation results proved the presented method to be applicable,and the obtained results confirm the superiority of the PSO-SVR in modeling of deformation modulus of rock mass.

关 键 词：岩体变形模量 支持向量机回归 粒子群算法 灰色关联分析 纵波波速 

分 类 号：TU454[建筑科学—岩土工程]