RS-SVM组合模型下煤矿安全风险预测  

Coal mine safety risk prediction by RS-SVM combined model

作　　者：汪莹[1] 蒋高鹏[1] 

机构地区：[1]中国矿业大学(北京)管理学院,北京100083

出　　处：《中国矿业大学学报》2017年第2期423-429,共7页Journal of China University of Mining & Technology

基　　金：国家自然科学基金项目(61471362)

摘　　要：针对煤矿安全风险预测的可靠性与效率问题,构建了基于粗集-支持向量机(RS-SVM)的煤矿安全风险预测模型.在不改变样本分类质量的条件下,运用RS方法作为预处理器约简特征参数,然后基于SVM方法进行分类建模.以煤矿瓦斯爆炸风险预测为例,从人员、环境、设备及管理4方面建立煤矿安全生产预警指标体系.利用遗传算法进行RS粗糙集属性约简,将初始风险因子由31项剔除为5项.将得到的约简集作为新的论域,并基于约简集利用SVM进行样本训练,通过回判法对30个训练样本的计算结果进行验证,得出SVM模型精确可靠的结论.进而利用5个预测样本得到其2015年瓦斯爆炸的风险预测结果,与实际情况完全相符.表明本文建立的RS-SVM组合预测模型对煤矿安全风险预测具有良好的指导作用.Aiming at the reliability and efficiency problem of coal mine safety risk prediction,this paper constructed a coal mine safety risk prediction model based on rough set and support vector machine（RS-SVM）.Without changing the quality classification of the sample,the RS method was used as the preprocessor reduction characteristic parameter.Then,the classification modeling was built based on the SVM method.The coal mine production safety early warning index system for coal mine gas explosion risk prediction was established from four aspects：personnel,environment,equipment and management.The rough set attribute reduction conducted by genetic algorithm can remove the initial risk factors from 31 to 5key factors.The obtained reduction set forms a new domain,which serves as the basis for the samples training by SVM method.The results of 30 training samples were validated by the method of backtracking in order to testify the accuracy and reliability of the SVM model.Furthermore,the results of the gas explosion risk prediction in 2015 were derived by using the five forecast samples,and the predicting result was confirmed to be consistent with the actual situation.The proposed RS-SVM combined predicting model has great significance for coal mine safety risk prediction.

关 键 词：粗集 遗传算法 属性约简 RS-SVM 风险预测 

分 类 号：TD771[矿业工程—矿井通风与安全] X928.03[环境科学与工程—安全科学]