基于PCA-RVM的矿山岩石爆破粒径预测模型

PCA-RVM based model for particle size prediction of blasted mine rock

为解决露天矿山爆破开采过程中岩石爆破粒径大小难以获取的问题,提出一种基于主成分分析法(PCA)及相关向量机(RVM)相结合的矿山岩石爆破粒径预测模型.该模型利用PCA对样本数据进行降维处理,选取出4个相互独立的主成分变量,并借助RVM构建主成分与爆破粒径之间的非线性映射关系,从而建立预测模型.将该模型应用于工程实例,并与BP神经网络和LM双隐含层模型进行对比.结果表明,在相同学习样本下,PCA-RVM模型预测结果与实际值更加接近,在平均相对误差和均方差上远小于另两种模型.

In order to resolve the difficulties in obtaining the particle size of blasted rock in the process of blasting in open pit mines, a predication model based on principal component analysis(PCA) and relevance vector machine(RVM) for the particle size of blasted rock in mine was proposed. The model reduced the dimensionality of sample data by using PCA, selected four independent principal component variables and constructed a non-linear mapping relationship between the principal components and the blasting size with the help of RVM, and the prediction model was formulated consequently. This model was applied to an engineering case, in comparison with BP neural network and LM double hidden layer model. The results indicate that the prediction results of PCA-RVM model are closer to the actual values with much smaller average relative error and mean squared error than those of other two models, under the condition of identical learning samples.