基于粗糙集和支持向量机的采空区煤自燃火灾预报

考虑到采用标志气体分析法对煤自燃火灾预报时特征维数较高、特征之间存在冗余且样本有限,文中提出基于粗糙集和支持向量机的采空区煤自燃火灾预报方法。该方法首先采用粗糙集对原始样本去除冗余和特征维数约简得到多组候选特征子集,然后对获得的多组候选特征子集利用支持向量机进行分类和性能评价,选取分类性能最好的一组特征子集用于设计支持向量机分类器,并对采空区遗煤自燃状态进行预测分析。实验选择大同矿区煤样自然发火实验数据,与4种典型分类预测算法的进行比较分析,实验结果表明文中算法预测准确率更高,训练速度更快。粗糙集为煤自燃火灾预报中标志气体选择提供了一个理论依据和新的思路,而支持向量机则提高了煤自燃火灾预测的精度。

Forecast of spontaneous combustion fire in goaf based on rough set and support vector machine

Considering the higher dimension of feature, redundancy existing in features and limited samples when using mark gas analysis method to forecast the spontaneous combustion fire, this paper proposed a approach to forecast spontaneous combustion fire in goal based on rough set （RS） and support vector machine （SVM）. This methods gets multiple groups of candidate feature subsets by using RS to eliminate redundancy and reduce feature dimension from original sample, then the candidate feature subsets above mentioned is classified by SVM and the classification performance is evaluated, the candidate feature subset with the best classification performance is used to design the SVM classifier to predict and analyse the state of residual coal spontaneous combustion in goaf. At last, the experimental data of spontaneous combustion of coal samples from Datong mine is selected in the experiment. Comparing with four typical classified and predict methods, the experimental result shows that the proposed method has higher predictive accuracy and faster training speed. RS provides a theory and new idea to select mark gas for the forecast of spontaneous combustion fire, while SVM improves the predict accuracy of spontaneous combustion fire.