摘要
简要介绍了基于统计学习理论的支持向量机回归(SVR)原理,针对边坡稳定性影响因素的复杂性,结合实例运用SVR技术构建了铀矿边坡稳定性的支持向量回归预报模型,并利用网格搜索与留一交叉验证方法(LOOCV)优化模型参数。研究表明,在小样本条件下,SVR预报模型对训练样本的计算值与实测值平均相对误差(MRE)为0.045967%,相对均方误差(MSRE)为0.046371%,拟合值(VOF)为1.999995765,相关系数(R)为0.9984,均比人工神经网络方法的相应指标值要小,说明支持向量回归方法是一种科学有效的矿山边坡稳定性的分析方法。
The regression principle of support vector machines（SVM）based on the statistic learning theoryis in troduced and mathematical model com-bined with grid search and Leave- one- out cross validation（LOOCV）which is used to predict the stability of Uranium Mine slope is buildby by support vector regression technology. The results show that the errors for training samples are 0.045967%（MRE）, 0.046371%（MSRE）, 1.999995765 （VOF） and 0.9984（R）usingasmal quantity of samples to build the mathematical model,and the predicting precision of SVR model is obviously betterthan that of BP neural network model.It is suggested that SVR is an effective and powerful tool for analysis of mine slope stability.
出处
《科技风》
2014年第6期131-134,共4页
