充填体强度的支持向量机设计匹配模型  

Matching Model of Backfill Strength Design on Support Vector Machines

作　　者：王志军[1,2] 吕文生[1,2] 杨鹏[1,3] 王志凯[1,2] 王金海 

机构地区：[1]北京科技大学土木与环境工程学院,北京100083 [2]金属矿山高效开采与安全教育部重点实验室,北京100083 [3]北京联合大学机器人学院,北京100101 [4]北方爆破科技有限公司,北京100089

出　　处：《金属矿山》2016年第11期34-38,共5页Metal Mine

基　　金："十二五"国家科技支撑计划项目(编号:2012BAB08B01)

摘　　要：充填体强度设计的预测受到多种高维度、非线性、随机性因素的影响。为改善当前充填体强度设计预测效果不佳的现状,使用支持向量机(SVM)方法在matlab软件中借助Lib SVM工具箱建立充填体强度设计匹配模型。分析并筛选出8个主要因素作为条件属性,充填体强度作为决策属性,并挑选出72组训练样本和6组校验样本。模型选择径向基函数(RBF)为核函数,采用网格搜索法对参数寻优,再通过交叉验证检验最优参数组合。结果表明:SVM匹配模型做出的回归预测平均误差为1.94%,校验预测平均误差为2.23%,相对于BP神经网络模型,预测准确度更高。在保证采场稳定性的前提下,SVM匹配模型更为有效地减少水泥消耗、降低充填成本,提高企业经济效益。The forecast of backfill strength design can be influenced by many factors,such as high-dimensional parameters,nonlinear and random elements. In order to improve the current status of the poor prediction effect of backfill strength design,the support vector machines was adopted to establish a backfill strength design matching model with Lib SVM toolbox in Matlab software. In this paper,eight major factors have been selected as condition attribute,backfill strength as decision attribute,as well as 72 sets of training samples and 6 sets of check samples are determined. With radial basis function（ RBF） as its kernel function,parameters are optimized by grid search method,and the optimal parameter combination is tested through crossvalidation method. The results show that： the average deviation of regression forecast and calibration made by SVM model are1. 94% and 2. 23% respectively,which are of higher accuracy than the BP neural network. On the premise of ensuring the stope stability,the model based on SVM can effectively reduce cement consumption,lower backfill costs,and therefore improve economic benefits of enterprises.

关 键 词：胶结充填 充填体强度 支持向量机 BP神经网络 预测模型 

分 类 号：TD853[矿业工程—金属矿开采]