MI和SVM算法在煤与瓦斯突出预测中的应用

Application of MI and SVM in coal and gas outburst prediction

为解决能用于煤与瓦斯突出预测模型的真实事故训练数据量小、数据集缺失严重的问题,提出采用数据挖掘多重填补(MI)算法填补事故数据中缺失参数,增大可用数据集,并将填补后的数据用于支持向量机(SVM)预测模型的训练与测试,选取K最近邻(KNN)算法与SVM进行对比。结果表明:SVM数据填补前后的平均识别率分别为88.37%和88.87%,事故数据的识别率分别79.71%和91.27%;KNN算法在数据填补前后,平均识别率分别为87.59%和88.37%,事故识别率分别为70.4%和84.23%;可见:MI对平均识别率的提升作用不大,对事故识别率的提升作用显著,可提高煤与瓦斯突出事故预测率,数据填补后SVM算法比KNN算法的事故识别率高。

In order to address problems of small quantity of accidents training data and lack of data set that can be used in coal and gas outburst prediction model,MI data mining algorithm was presented to fill up missing parameters in accident data and increase available data sets.Then,imputed data were employed in SVM prediction model's training and testing,and K-Nearest Neighbors(KNN)algorithm was selected and compared with SVM.The results show that the average recognition rate of SVM algorithm is 88.37%and 88.87%before and after data inputting respectively,and recognition rate of accident data is 79.71%and 91.27%respectively.That of KNN algorithm before and after data inputting is 87.59%and 88.37%respectively while accident recognition rate being 70.4%and 84.23%,respectively.Therefore,MI has little effect on improving average recognition rate,but has a greater one on improving accident recognition rate,which can improve prediction rate of coal and gas outburst accidents.The incident recognition rate of SVM algorithm is higher than that of KNN algorithm after data inputting.