| With the rapid development of underground engineering construction in water project,transportation,mining and other industries in China,the scope and depth of excavating works are increasing,and rockburst disasters caused by high ground stress and excavation disturbance occur frequently.As a dynamic instability geological disaster with strong burst,great harm and complex action mechanism,rockburst directly threatens the safety of personnel,equipment and buildings,and even induces earthquakes in serious cases.Therefore,in order to reduce the risk and loss of rockburst,accurately predicting rockburst intensity grade is the key to the safe and efficient construction of underground engineering.In this paper,the rockburst prediction model is established from two aspects: dataset optimization and algorithm improvement,aiming to improve the prediction accuracy of rockburst intensity grades.First,based on lithology conditions,surrounding rock conditions and stress conditions,the maximum tangential stress,uniaxial compressive strength,uniaxial tensile strength,stress coefficient,brittleness coefficient and elastic energy index of surrounding rock are selected as rockburst prediction indexes.According to the different classification standards of rockburst intensity proposed by relevant scholars,four rockburst intensity grades of no rockburst,slight rockburst,intermediate rockburst and strong rockburst are determined,and 188 groups of rockburst samples at home and abroad in the existing literature are taken as the experimental dataset.Then,the rockburst dataset is preprocessed,one is to use the multiple imputation by chained equations random forest(MICE＿RF)method to interpolate the missing values in rockburst samples,through the verification of the effectiveness of the method,the advantages and effectiveness of MICE＿RF method for rockburst missing data interpolation are confirmed.Second,in view of the unbalanced number of intensity samples in the rockburst dataset and the phenomenon of overlapping and abnormal values of sample points among various categories,the adaptive synthetic sampling approach(ADASYN)is improved from the perspectives of evaluating the quality of synthetic sample points(introducing the sample selection module)and selecting the sampling direction(introducing the hellinger distance),and a new oversampling algorithm(HDADASYN)is proposed,which is verified by the effectiveness of the algorithm,the advantages and effectiveness of HDADASYN algorithm in the balanced processing of rockburst dataset are confirmed.Next,the rockburst prediction algorithm is improved,based on particle swarm optimization(PSO),the nonlinear dynamic decreasing inertia weight and random behavior parameters are introduced to design the improved particle swarm optimization(IDPSO).Optimize the random initialization parameters of extreme learning machine(ELM)algorithm with better fitting to the nonlinear rockburst prediction problem,and establish the IDPSO-ELM rockburst intensity grade prediction model.Through the validation of the model,the advantages and accuracy of IDPSO optimized ELM are confirmed,and the feasibility and effectiveness of IDPSO-ELM model in predicting rockburst intensity grade are further verified.Finally,the IDPSO-ELM rockburst intensity grade prediction model based on data preprocessing is used to predict 15 groups of rockburst samples in Sichuan Jiangbian hydropower station diversion tunnel to verify the model’s practical application value,at the same time,it is compared with the prediction results of five models in the existing literature.The results show that the rockburst intensity grade prediction model has the highest prediction accuracy and the prediction results are relatively safe,it provides a new idea and method for rockburst prediction,and provides a strong reference for the next work of rockburst prevention and control.This thesis has 30 figures,20 tables and 100 references... |
PDF Full Text Request