对煤矿事故的预警是煤矿安全的一项重要的工作，目前对各类煤矿事故的预警基本上都是通过监测某单一因素是否达到临界值进行预警的，而当单一因素没有达到临界值，此时系统一般不预警，但有多个因素同时接近临界值时也是事故发生的危险状态。经过对煤矿爆炸事故的分析，发现影响煤矿爆炸的因素主要有瓦斯浓度、煤尘含量、CO含量、温度和风速等。本文通过对影响煤矿爆炸因素的综合分析研究，探索对煤矿爆炸事故进行综合预警的方法。 CBA (Classification Based on Association)是基于关联规则的分类算法。其主要特点是根据关联规则所产生的支持度大的规则去覆盖训练样本中支持度小的规则，从而得到分类规则。由于煤矿安全预警是一个比较复杂的工作，仅仅使用CBA算法不能达到一种好的效果。 本文使用一种改进的CBA算法为煤矿安全中的爆炸灾害的实时预警分析提供优化后的训练样本数据集及其对应的权值。改进的CBA算法对经过数据预处理后的训练样本集进行关联分析得到分类规则数据集，然后对分类规则数据集使用PFCM(Fuzzy C-Means Based on Solving Polynomial)算法，将其产生的隶属度值作为分类训练样本数据的权值进行加权贝叶斯分类。为了增加分类的准确率，可以将分类错误的样本进行反馈，重新使用PFCM算法计算权值和进行加权贝叶斯分类，直至分类的准确率不再改变。这样通过对大量的历史监测数据的算法分析，可以确定一个优化后的分类的训练样本及其对应的权值，以减少在实时预警时对实时监测数据使用加权贝叶斯分类的计算量，提高对煤矿爆炸事故进行实时预警的效率和准确率。

Coal mine accidents early warning is an important work for coal mine safety. At present, almost all coal mine accidents early warning are basically realized by monitoring a single factor whether meets the critical value or not. The system generally is not warning when each single factor not reaches the critical value, however, it is also a very dangerous state of the accident when multiple factors are close to the critical value. According to the analysis of coal mine explosion accidents, the main factors affecting the coal mine explosion are gas concentration, coal power content, carbon monoxide content, temperature and wind speed, etc. This paper explores the method of comprehensive warning for coal mine explosion disaster through synthetically analyze the factors of impact coal mine explosion. CBA (Classification Based on Association) is a classification algorithm based on class association rules. Its main feature is to get classification by the rule of covering the smaller support degree rules in the training samples based on the lager support degree rules which are generated by the association rules. As the coal mine safety warning is a complicated work, the single CBA algorithm can’t reach a good effect. The paper uses an improved CBA algorithm to provide an optimized training samples data set and the corresponding metric for real-time warning analysis of the coal mine explosion disaster. The improved CBA algorithm gets the classification rules data set through the correlation analysis of the training set in which the data is preprocessed. Then, the PFCM(Fuzzy C-Means Based on Solving Polynomial) algorithm is used. Finally, the membership value generated by the PFCM algorithm is used as the weight of the classified train sample data to do the weighted Bayesian classification. In order to increase the classification accuracy, feedback the wrong classification samples, and the weights are recalculated by the PFCM algorithm and the weighted Bayesian classification is done, until the classification accuracy rate does not change. So an optimized training samples data set and its corresponding weights can be determined through analyzing lots of historical monitoring data, which can improve the efficiency and accuracy in the real-time warning analysis for the coal mine explosion disaster.