摘要
为了能够准确识别煤矿开采中瓦斯爆炸事故的风险类别,以提升对瓦斯爆炸事故的应急防范能力,提出了一种高效准确的风险识别模型。考虑到瓦斯爆炸涉及特征指标之间并非单纯的线性关系,采取核主成分提取方法(KPCA)对涉及的煤层瓦斯含量、瓦斯涌出量、风量供需比等19种特征指标进行属性约简,从而消除信息重叠,并使用遗传算法(GA)对BP神经网络的权值和阈值进行全局寻优,避免了过拟合状况的出现,在降低模型预测损失值的同时提高模型运算准确率。采用KPCA-GA-BP模型对83组实例数据进行分析,选取其中62组数据进行训练,其余21组数据进行测试。预测结果中20组风险等级分类正确且用时较短,识别准确率为95.24%,表明模型对瓦斯爆炸灾害风险等级具有高识别精度和高效率。
In order to recognize the risk category of the gas explosion accident in the coal mining process accurately to promote the emergency precaution and preventing ability of the gas explosion accidents,the paper intends to propose a kind of efficient and accurate risk recognition model. The given paper intends to extend the classification to 19 indexical categories in evaluating the gas explosion risk grade. However,since the characteristic indexes related to the gas explosion are not simply linear with a large number of information items that involve the prediction or forecast functions of the model,the paper has to introduce the extraction method( KPCA) with its kernel components by eliminating the information overlapping and extracting the 7 chief elements to reflect the hidden information of the original data properly to reach 90. 71% of the original functions by reducing the above mentioned coal seam gas-bearing capacity,the gas emission quantity,the blast capacity supply and the demand ratio in addition to the other 16 specific indexes of the accumulative interpretation ratio of the main elements. For,it is only through the global optimization of the genetic algorithm( GA) to the weight and threshold of the BP neural network that it would be possible to get rid of any over-fitting condition by reducing the model prediction loss and improving the model operation accuracy. And,in so doing,the optimal fitness of the genetic algorithm can be made stable and fit for implementation,for it involves the optimal operational fitness of 3 439 times to reduce the mean square error of the lowest loss up to mere 0. 029. The reduction also involves the analysis of 83 groups of instance data by KPCA-GA-BP model,by choosing 62 groups of data training,in addition to the rest testing of 21 data groups. What is more,20 groups of risk level classification are to be corrected in a short instant from the predicted results so as to attain the recognition accuracy rate of 95. 24%. For,all the above said accuracy can indicate and repres
作者
温廷新
孔祥博
WEN Ting-xin;KONG Xiang-bo(System Engineering Institute,Liaoning Technical University,Huludao 125105,Liaoning,China)
出处
《安全与环境学报》
CAS
CSCD
北大核心
2021年第1期19-26,共8页
Journal of Safety and Environment
基金
国家自然科学基金项目(71371091)
辽宁省社会科学规划基金项目(L14BTJ004)。
关键词
安全工程
瓦斯爆炸
风险
主成分
遗传算法
神经网络
safety engineering
gas explosion
risk
principal components
genetic algorithms
neural network
