Prediction Method of Rock Cohesion and Internal Friction Angle Based on Ensemble Tree Algorithm
The cohesion(c)and internal friction angle(φ)of rock are critical parameters in the design and stability assessment of rock engineering projects.Direct measurement of these parameters necessitates condu-cting numerous rock triaxial or shear tests,which are both time-intensive and expensive.This study proposes the development of intelligent models to predict the values of c and φ based on four readily obtainable parameters:P-wave velocity(VP),density(ρ),uniaxial compressive strength(UCS),and Brazilian tensile strength(BTS).A total of 199 datasets containing various rock types were collected and randomly partitioned into a training set(80%)and a test set(2%).The distribution characteristics and correlations among the data were analyzed using scatter plots for data distribution and correlation plots for variables.To address discrepancies in characteristic attributes,such as magnitude and order of magnitude across different input variables,a normalization function was applied.Subsequently,five ensemble trees were utilized to develop predictive models for rock shear strength parameters.Bayesian optimization was employed to optimize the hyperparameters of the models.Concurrently,five-fold cross-validation was implemented during model training.To evaluate the performance of the models,four widely recognized regression metrics were utilized:The coefficient of determination(R²),root mean square error(RMSE),mean absolute error(MAE),and variance accounted for(VAF).Additionally,a ranking system was introduced to provide a comprehensive assessment of the five models.The model evaluation demonstrated that the constructed models exhibited robust predictive performance,with the extremely randomized tree model outperforming others.Specifically,for predicting the value of c,the R2 was 0.993,the RMSE was 0.45,the MAE was 0.309,and the VAF was 99.306%.For predicting the value of φ,the R2 was 0.97,the RMSE was 0.823,the MAE was 0.612,and the VAF was 97.058%.Furthermore,the application of the SHAP interpretation method for sensitivity analysis indicated that VP,UCS,and BTS significantly influenced on the prediction of c,whereas ρ had a substantial impact on the prediction of φ.Finally,rock blocks were collected and processed into samples for physical-mechanical testing to determine the VP,ρ,UCS,BTS,c,and φ values of rocks at various locations within the Jinchuan Ⅱ and Ⅳ mining areas in China.The model was effectively utilized to predict the c and φ values for rocks in the Jinchuan mining area,thereby validating its practicability.Furthermore,a graphical user interface was developed to facilitate ease of use for engineers and technicians in the field.
cohesioninternal friction anglemachine learningensemble tree algorithmBayesian optimi-zationintelligent prediction
万方数据
维普
