Research On Identification Method Of Gas Disaster Risk

Gas disaster seriously threatens the safety of coal mine production.In order to avoid blindness and achieve predictability in gas disaster prevention and control.Coal mine gas disaster prediction technology has become a research focus in the field of coal mine safety.However,the existing gas disaster prediction technology can not meet the practical application requirements.In order to improve the practicability of gas disaster prediction technology,this paper has carried out in-depth research on coal mine gas disaster risk identification methods.The main research contents are as follows.(1)Aiming at the problem of high dimensionality in coal mine monitoring data,an ensemble classification model based on data dimensionality reduction(ECM-DDR)is studied.Firstly,the model adopts principal component analysis,local linear embedding,kernelized principal component analysis,isometric mappilng and multidimensional scaling dimensionality reduction methods to reduce data dimensionality.Then,five ensemble classification models is constructed on the data sets generated by five dimension,ality reduction methods.Finally,the optimal model is selected for gas disaster risk identification.The experimental results show that ECM-DDR has less running time.And the accuracy of the ensemble classification model based on local linear embedding is 1.78%higher than that of the ensemble classification model without data dimensionality reduction.(2)For the shortcomings of traditional dimensionality reduction methods in the improvement of gas disaster risk identification accuracy,a selective ensemble regression learning model based on correlation analysis(CA-SERL)is proposed.Firstly,CA-SERL model realizes attribute reduction by analyzing the correlation between gas concentration and sample atrtributes.Then,the base learners are modeled,and the selective ensemble regression learning model is established by the optimization ensemble forward sequential selection method.Finally,the model is used for gas disaster risk identification.The experimental results show that the recognition rate of CA-SERL model for gas disaster risk compared with four single learners without correlation analysis improves 15.12%on average,and compared with the selective ensemble regression learning model without correlation analysis improves 4.11%.(3)In order to further improve the performance of gas disaster risk identification,a selective ensemble classification model based on clustering selection and a new degree of combination fitness(CS-NDCF)is proposed.Firstly,CS-NDCF method uses clustering algorithm to screen base classifiers for the first time.Then,the result set of base classifiers selected by clustering algorithm is used as the input of degree of combination fitness method.The second screening of base classifiers is carried out by degree of combination fitness,and it gets the base classifier result set.Finally,a selective ensemble classification model is constructed by using the base classifier result set.The experimental results show that CS-NDCF selection strategy has high search efficiency for classifiers.And it improves the classification performance of selective ensemble classification model.