Research And Application Of Rockburst Intensity Classification Prediction Model Based On Machine Learning Algorithms

Rockburst is one of the difficult problems faced by large-scale underground geotechnical and deep resource mining projects.Accurate prediction of rockburst intensity classification has important engineering significance and academic value.Rockburst intensity classification prediction is an important scientific basis for rockburst prevention and control,and accurate and practical prediction model can effectively guide rockburst prevention and control.However,prediction model is affected by a variety of complex factors,and its effectiveness needs to be improved in terms of index weight determination and practical engineering applications.Based on the established rockburst intensity classification prediction database,aiming at the randomness,ambiguity,finiteness,non-linearity,and discreteness of the rockburst prediction data,this paper adopts machine learning technology to deeply explore the value of data.Three types of rockburst intensity classification prediction models are proposed,and the effectiveness of the prediction models are verified.At the same time,the prediction models are applied to the deep mining rockburst engineering practice of a gold mine in Chifeng,Inner Mongolia.The main content of the paper:(1)Rockburst intensity classification prediction database has been established.By analyzing four rockburst engineering examples,and after comprehensive consideration of the internal and external causes,affecting factors and characteristics of rockburst,this paper selects four rockburst evaluation indices,i.e.,tunnel-wall surrounding rock's maximum tangential stress(),rock uniaxial compressive strength(),rock uniaxial tensile strength(),and rock elastic energy index().Through comparing and analyzing of the existing rockburst intensity classification schemes at home and abroad,and considering the strength of the rockburst and the main influencing factors,the rockburst intensity is divided into four levels,i.e.,level I(no rockburst),level II(light rockburst),level III(moderate rockburst),and level IV(strong rockburst).According to the determined rockburst evaluation index and rockburst intensity classification,a database containing 301 rockburst engineering examples has been established as sample data for rockburst intensity classification prediction.(2)Rockburst intensity classification prediction model based on RF-AHP-CMhas been proposed.Considering the timeliness of rockburst prediction,the analytic hierarchy process(AHP)is used to calculate the weight of the rockburst evaluation index,and the random forest(RF)algorithm which can effectively deal with the fuzzy data characteristics is used to establish a random forest-based importance analysis model of rockburst evaluation index.According to the quantitative analysis results of index importance,the analysis matrix in the analytic hierarchy process is constructed,and the analytic hierarchy process is optimized,and the RF-AHP index weight calculation method is constructed.Combined with the cloud model(CM),the RF-AHPCM rockburst prediction model is constructed,and its prediction accuracy rate can reach 85%.The prediction model can judge the main rockburst intensity classification,and can also judge the possible rockburst intensity classification at the same time,effectively solving the rockburst prediction problems with uncertainty,randomness and ambiguity.(3)Rockburst intensity classification prediction model based on IGSO-SVM has been proposed.Aiming at the finiteness and non-linearity of the rockburst prediction data,the improved glowworm swarm optimization(IGSO)is used to optimize the penalty parameter(C)and the radial basis function parameter(g)of the support vector machine(SVM),and construct the IGSO-SVM rockburst prediction model,and its prediction accuracy rate can reach 90%.The prediction model avoids the problem of determining index weight,and effectively solves the problem of non-linear rockburst prediction under limited sample conditions by directly learning the data of rockburst engineering examples.(4)Rockburst intensity classification prediction model based on DA-DNN has been proposed.In order to meet the needs of larger-scale rockburst data processing,deep neural network(DNN)is used.Aiming at the discreteness and finiteness of rockburst prediction data,Dropout is used to regularize the model to prevent overfitting.At the same time,in order to improve the timeliness and stability of the prediction model,the improved Adam algorithm is used to optimize the parameters.The DA-DNN rockburst prediction model is constructed,and its prediction accuracy rate can reach98.3%.The prediction model effectively solves the problem of rockburst prediction with larger data scale.(5)Comparative analysis and engineering example application of different rockburst intensity classification prediction models.The RF-AHP-CM,IGSO-SVM and DA-DNN rockburst prediction model are compared and analyzed from the prediction accuracy,timeliness and scope of application.Three rockburst prediction models have their own advantages,and effectively solve the rockburst prediction problem from different angles.Three established rockburst prediction models are used to predict the deep mining of a gold mine in Chifeng,Inner Mongolia.The prediction results agree well with the actual situation on site,which verifies the accuracy and practicality of the three models.According to the prediction results of rockburst and the actual production of the mine,eight corresponding rockburst prevention measures are proposed.