基于特征选择与机器学习的煤与瓦斯突出危险等级协同预测方法

Cooperative prediction method of coal and gas outburst risk grade based on feature selection and machine learning algorithm

煤与瓦斯突出危险性预测可有效防止煤矿井下突出灾害事故。为进一步提高煤与瓦斯突出危险等级预测的科学性及准确性,构建了基于多算法和多元分析的煤与瓦斯突出动态预测模型。选择51组煤与瓦斯突出工程案例数据作为样本集,对样本数据进行空值填补、数据标准化等预处理,通过引入6种特征选择方法及6种有监督机器学习算法构建了42种煤与瓦斯突出危险等级预测模型。采用准确率、混淆矩阵、Kappa系数及F1值等指标对预测模型的性能进行验证与评估,筛选出精度及稳定度高的4种机器学习算法和3种特征参数组合,确定了8种最优分类模型,并对8组典型的煤与瓦斯突出事故案例进行等级预测。结果表明:8种最优分类预测模型准确率为0.667~0.961,Kappa系数为0.625~0.920,F1值为0.615~1;实际案例煤与瓦斯突出预测准确率为100%,突出等级预测准确率为87.5%。所构建的多参数、多算法、多组合、多判定指标的煤与瓦斯突出等级协同预测模型精度较高,且具有一定的普适性,可为煤与瓦斯突出危险等级预测提供一种新思路。

Coal and gas outburst risk prediction is essential to effectively prevent underground coal and gas outburst disasters.In order to improve the scientificity and accuracy of coal and gas outburst risk level prediction,proposed a dynamic prediction model of coal and gas outburst based on multi-algorithm and multivariate analysis.The system selected 51 sets of coal and gas outburst engineering case data as the sample set.Null filling and data standardization were used for preprocessing of the sample data.The 42 prediction models of the coal and gas outburst risk level were built by introducing 6 feature selection methods and 6 supervised machine learning algorithms.The accuracy,confusion matrix,Kappa coefficient and F1 value were used to verify and evaluate the performance of the prediction mode. 8 optimalclassification models were determined. Finally,the classification model is used for the prediction of 8typical coal and gas outburst cases. The results show that the accuracy rate of the 8 optimal classificationprediction models is 0. 667-0. 961,the Kappa coefficient is 0. 625-0. 920,and the F1 value is 0. 615-1.The actual case of coal and gas outburst prediction accuracy rate is 100%,and the outburst grade predictionaccuracy rate is 87. 5%. The constructed multi-parameter,multi-algorithm,multi-combination,andmulti-identification index collaborative prediction system of coal and gas outburst level has high accuracyand a certain degree of universality,which can provide a new way for the prediction of the coal and gasoutburst risk level.