基于主成分分析的多项Logistic回归模型的突水水源判别研究

Study on Water Source Discrimination Based on Multinomial Logistic Regression Model Using Principal Component Analysis

矿井水害是矿井生产过程中较常见的地质灾害之一。快速有效地判别突水水源是预防矿井水害的关键所在。选取袁店二矿59个水样资料(常量离子含量),利用主成分分析进行处理,得出主成分得分;以主成分得分为自变量,水源类别为因变量建立多项Logistic回归模型;运用该回归模型对59个水样资料进行类型判别,得出综合判别准确率达到86.4%;并通过实例对判别模型进行了验证。研究结果表明:主成分分析法与多项Logistic回归模型相结合的方法在水源判别上具有可行性,不仅消除了常量离子之间的内在影响,而且使判别结果具有一定的准确率,在突水水源判别问题上提供了一种新方法,为矿井防治水提供有效依据。

Mine water disaster is one of the most common geological disasters in mine production process. Quickly and effectively determine the source of water inrush is a key to the prevention of mine water damage. The data of 59 water samples were selected in Yuaner coal mine. The data of water sample mainly include constant ion content. The content of constant ions was treated by principal component analysis. The principal component scores were used as independent variables. The water source category is used as a dependent variable. Multinomial logistic regression models were established by using these. The regression model was used to classify the type of the data of 59 water samples. The discriminant rate （ 86. 4% ） of synthesis was obtained and the discriminant model was validated with examples. The results show that the combination of the principal component analysis and the multinomial Logistic regression model is feasible in water source identification. It not only eliminates the inherent influence of the constant ions, but also makes a certain degree of accuracy of the discriminant results. A new method is provided for discriminating water inrush source and it will provide an effective basis for preventing from and controlling of mine water.