矿井突水水源的多项逻辑回归判别

Multinomial Logistic Regression Discrimination of Mine Water Inrush Source

摘要: 准确判别突水水源是矿山安全生产需要解决的关键问题之一。以袁店二矿突水水源判别问题为例，选取归一化后的Na⁺ + K⁺、Ca²⁺、Mg²⁺、Cl⁻、$S{O}_4^{2-}$ 以及$HC{O}_3^{-}$ 离子浓度作为模型的输入，使用带Armijo线搜索的最速下降法优化交叉熵损失函数，建立突水水源判别的多项逻辑回归模型。计算结果表明，得到的多项逻辑回归模型对55组训练样本的判别正确率为100%，对8组测试样本的判别正确率为87.5%，验证了多项逻辑回归是突水水源判别的一种有效工具。

Abstract: Identifying the water inrush source correctly is one of the key problems that must be solved for the safety of coal mining operations. The water inrush source discrimination problem of Yuandian No. 2 mine is taken as an example. The normalized ion concentrations of Na⁺ + K⁺, Ca²⁺, Mg²⁺, Cl⁻, $S{O}_4^{2-}$ and $HC{O}_3^{-}$ are selected as the inputs of the model. The cross-entropy loss function is optimized using the steepest descent algorithm with the Armijo line search and the multinomial logistic regression model is established to discriminate the water inrush source. The computation results show that the discrimination accuracy of the obtained multinomial logistic regression model is 100% for the 55 training samples and 87.5% for the 8 testing samples, which indicate that multinomial logistic regression is an effective tool for mine water inrush source discrimination.