In order to improve the prediction accuracy of gas emission quantity,aiming at the problems of linear overlapping and high-dimensional nonlinearity of the influencing factors of gas emission quantity,it was proposed to carry out the dimen-sionality reduction on the influencing factors by using the kernel principal component analysis(KPCA).Firstly,30 sets of sample data from a mine in Shenyang were selected,with the first 24 sets of data as the training set and the last 6 sets of data as the test set.Then the determined kernel principal components were used as the input variables of least squares support vec-tor machine(LSSVM)to establish the KPCA-LSSVM prediction model,and the prediction results were compared with the prediction results of PCA-LSSVM,LSSVM,multivariate nonlinear regression,KPCA-BP neural network,PCA-BP neural network,and BP neural network.Finally,the maximum absolute relative error was used as the evaluation index of model pre-diction accuracy.The results show that the requirements of model training are met when the first four kernel principal compo-nents are selected.The maximum absolute relative error of prediction by the KPCA-LSSVM model is 5.89％.The prediction accuracies are all better than the other six comparison models.The research results can provide a reference for realizing the high accuracy prediction of gas emission quantity.
关键词
瓦斯涌出量的预测/核主成分分析法(KPCA)/最小二乘支持向量机(LSSVM)/相对误差绝对值
Key words
prediction of gas emission quantity/kernel principal component analysis(KPCA)/least squares support vector machine(LSSVM)/absolute relative error
维普
万方数据