基于Weka平台的矿井突水水源判别

Identification of mine water inrush source based on Weka platform

矿井突水水源判别一直是涉及矿井安全开采的一项重要内容。本文基于Weka平台,选取潘二煤矿34组二叠系砂岩水样、灰岩水样、老空水样作为训练水样以及9组突水水样,并对水样中的水温、矿化度、碱度、纯硬度、离子含量进行整理,而后将数据进行归一离散化。采用朴素贝叶斯、多分类Logistic回归模型、MLP以及C4.5算法模型对34组水样进行训练,选择效果最好的模型对9组突水水样进行预测。结果表明:在混淆矩阵、详细精度、节点错误率上,MLP的训练结果均优于其他算法,MLP在9组突水水样的预测结果中,有8组正确,1组错误,准确率88.9%。

The identification of mine water inrush source has always been an important content related to mine safety mining.Based on Weka platform,34 groups of Permian sandstone water samples,limestone water samples and old air water samples are selected as training water samples and 9 groups of water inrush water samples.The water temperature,salinity,alkalinity,pure hardness and ion content in the water samples are sorted out,and then the data are normalized and discretized.Naive Bayes,multi classification logistic regression model,MLP and C4.5 algorithm model trains 34 groups of water samples,and selects the best model to predict 9 groups of water inrush samples.The results show that the training results of MLP are better than other algorithms in terms of confusion matrix,detailed accuracy and node error rate.Among the prediction results of 9 groups of water inrush samples,8 groups are correct and 1 group is wrong,with an accuracy rate of 88.9%.