基于SPSS的PCA-MRA回采工作面瓦斯涌出量预测

Prediction of gas emission quantity in the working face based on SPSS PCA-MRA

为克服自变量之间的多重共线问题,增强多元回归模型预测的精确性,将主成分分析(PCA)与多元回归分析(MRA)相结合提出了PCA-MRA模型,并将该模型用于实际瓦斯涌出量预测。结果表明,采用SPSS软件直接对影响回采工作面瓦斯涌出量的因素进行主成分回归分析,避免了复杂的推导计算以及繁琐编程,预测精度较高。

In order to forecast the gas emission quantity in the working face in a coal mine, this paper intends to make evaluation of gas emission quantity in the working face based on the data of the gas emission so as to establish a principal component-multivariate retrogressive analysis （PCA - MRA） model by means of the statistical product and service solutions （SPSS） software. Before establishing the model, first of all, in this study we would like to work on the collinearity diagnostics. According to the results, the minimum of variance inflation factor （VIF） is 5.426 and the maximum tolerance is 0.184. However, when VIF is greater than 2, there may exist the problem of collinearity. And the smaller the tolerance is, the more serious the problem of collinearity will be. If the tolerance is smaller than 0.1, the problem of collinearity is likely to be sewere, and we have to make sure accordingly that there exist some serious collinearity problems among the independent variables. To get over the problem of collinearity, in this study we may choose the first five principal components to establish the PCA - MRA model. The cumulative contribution rate for the first five principal components should be up to 96. 138% , which can replace the 14 raw data variables. Thus, using the established model, we could predict the quantity of gas emission from the working faces numbered as 15, 16, 17, 18, with the maxi- mum relative error being - 3. 684 % and the smallest relative error （only -0.091% and the average error 1.364%）. Compared with grey relational grade analysis, the fuzzy comprehensiw： evaluation model and artificial neural network model, it can be concluded that the accuracy of PCA - MRA is strongly beneficial, and, this method can help to forecast the quantity of gas effectively by avoiding complicated derivation and calculation. Hence, our method can lead to precise results, which can significantly reduce the calculation time and is convenient in use.