Rockbursts pose severe risks to underground engineering projects,including mining and tunnelling,where sudden rock failures can lead to substantial infrastructure damage and loss of human lives.An accurate assessment ...Rockbursts pose severe risks to underground engineering projects,including mining and tunnelling,where sudden rock failures can lead to substantial infrastructure damage and loss of human lives.An accurate assessment of rockburst damage is essential for safety and effective risk mitigation.This study investigates the effectiveness of ensemble machine learning models optimized through Bayesian optimization(BO)in predicting rockburst damage scales.Nine classifier algorithms,including random forest(RF),were evaluated using a dataset of 254 samples.The research considered factors such as stress conditions,support system capacity,excavation span,geological characteristics,seismic magnitude,peak particle velocity,and rock density as input variables.The rockburst damage scale,categorized into four severity levels based on displaced rock mass,served as the target variable.Among the models evaluated,BO-RF model demonstrated the highest predictive accuracy and generalization capability,achieving 92%testing accuracy.BO-RF model also ranked top in a multi-criteria evaluation framework.This devised ranking system underscores the importance of evaluating model performance on both training and unseen testing data to ensure robust generalization.The findings underscore the effectiveness of BO-RF in enhancing rockburst risk assessment and providing reliable predictive insights for underground engineering applications.展开更多
The uniaxial compressive strength(UCS)of rocks is a crucial indicator for evaluating the bearing capacity of geological structures in rock engineering,and it holds significant implications for disaster management.Howe...The uniaxial compressive strength(UCS)of rocks is a crucial indicator for evaluating the bearing capacity of geological structures in rock engineering,and it holds significant implications for disaster management.However,direct measurement poses a significant challenge.Therefore,simpler alternatives such as Schmidt hammer rebound number(SRn),P-wave velocity(Vp),and point load index(Is)are frequently used to estimate UCS indirectly.In this study,we compiled a comprehensive dataset of 1168 samples that included SRn,Vp,Is,and UCS values.The dataset was refined using an isolation forest algorithm,which identified and removed 280 outliers,leaving a dataset of 888 samples for analysis.We developed and assessed an automated machine learning(AutoML)model for predicting UCS,introducing a novel approach to tackle this prediction challenge.Additionally,we compared models enhanced by Bayesian optimization,including multi-layer perceptron(MLP),support vector machine(SVM),Gaussian process regression(GPR),and K-nearest neighbor(KNN).Among these,the AutoML model demonstrated superior performance in UCS prediction,offering a rapid and efficient method for estimating UCS in engineering applications and enabling intelligent classification of rock masses.The study also evaluated the sensitivity and contribution of SRn,Vp,and Is in UCS estimation by various techniques,including permutation feature importance(PFI),SHapley Additive exPlanations(SHAP),and local interpretable model-agnostic explanations(LIME).The results underscore that the AutoML approach not only streamlines UCS modeling but also provides a robust and comprehensive solution,significantly enhancing the accuracy and efficiency of the prediction process.展开更多
Engineering disasters, such as rockburst and collapse, are closely related to structural instability caused byinsufficient bearing capacity of geological materials. Uniaxial compressive strength (UCS) holds considerabl...Engineering disasters, such as rockburst and collapse, are closely related to structural instability caused byinsufficient bearing capacity of geological materials. Uniaxial compressive strength (UCS) holds considerablesignificance in rock engineering projects. Consequently, this study endeavors to devise efficient models for theexpeditious and economical estimation of UCS. Using a dataset of 729 samples, including the Schmidt hammerrebound number, P-wave velocity, and point load index data, we evaluated six algorithms, namely AdaptiveBoosting (AdaBoost), Gradient Boosting Decision Tree (GBDT), Extreme Gradient Boosting (XGBoost), LightGradient Boosting Machine (LightGBM), Random Forest (RF), and Extra Trees (ET) and utilized Bayesian Optimization (BO) to optimize the aforementioned algorithms. Moreover, we applied model evaluation metrics suchas Root Mean Square Error (RMSE), Mean Absolute Error (MAE), Variance Accounted For (VAF), Nash-SutcliffeEfficiency (NSE), Weighted Mean Absolute Percentage Error (WMAPE), Coefficient of Correlation (R), and Coefficient of Determination (R2). Among the six models, BO-ET emerged as the most optimal performer duringtraining (RMSE ¼ 4.5042, MAE ¼ 3.2328, VAF ¼ 0.9898, NSE ¼ 0.9898, WMAPE ¼ 0.0538, R ¼ 0.9955, R2 ¼0.9898) and testing (RMSE ¼ 4.8234, MAE ¼ 3.9737, VAF ¼ 0.9881, NSE ¼ 0.9875, WMAPE ¼ 0.2515, R ¼0.9940, R2 ¼ 0.9875) phases. Additionally, we conducted a systematic comparison between ensemble andtraditional single machine learning models such as decision tree, support vector machine, and K-NearestNeighbors, thus highlighting the advantages of ensemble learning. Furthermore, the enhancement effect of BO ongeneralization performance was assessed. Finally, a BO-ET-based Graphical User Interface (GUI) system wasdeveloped and validated in a Tunnel Boring Machine-excavated tunnel.展开更多
基金supported by the Young Elite Scientist Sponsorship Program by China Association for Science and Technology under Grant No.YESS20230742.
文摘Rockbursts pose severe risks to underground engineering projects,including mining and tunnelling,where sudden rock failures can lead to substantial infrastructure damage and loss of human lives.An accurate assessment of rockburst damage is essential for safety and effective risk mitigation.This study investigates the effectiveness of ensemble machine learning models optimized through Bayesian optimization(BO)in predicting rockburst damage scales.Nine classifier algorithms,including random forest(RF),were evaluated using a dataset of 254 samples.The research considered factors such as stress conditions,support system capacity,excavation span,geological characteristics,seismic magnitude,peak particle velocity,and rock density as input variables.The rockburst damage scale,categorized into four severity levels based on displaced rock mass,served as the target variable.Among the models evaluated,BO-RF model demonstrated the highest predictive accuracy and generalization capability,achieving 92%testing accuracy.BO-RF model also ranked top in a multi-criteria evaluation framework.This devised ranking system underscores the importance of evaluating model performance on both training and unseen testing data to ensure robust generalization.The findings underscore the effectiveness of BO-RF in enhancing rockburst risk assessment and providing reliable predictive insights for underground engineering applications.
基金supported by the National Natural Science Foundation of China(Nos.42472351,42177140,52404127,and 42207235)the Natural Science Foundation of Hubei Province(No.2024AFD359)+1 种基金the Young Elite Scientist Sponsorship Program by China Association for Science and Technology(No.YESS20230742)the China Post-doctoral Science Foundation Program(No.2024T170684).
文摘The uniaxial compressive strength(UCS)of rocks is a crucial indicator for evaluating the bearing capacity of geological structures in rock engineering,and it holds significant implications for disaster management.However,direct measurement poses a significant challenge.Therefore,simpler alternatives such as Schmidt hammer rebound number(SRn),P-wave velocity(Vp),and point load index(Is)are frequently used to estimate UCS indirectly.In this study,we compiled a comprehensive dataset of 1168 samples that included SRn,Vp,Is,and UCS values.The dataset was refined using an isolation forest algorithm,which identified and removed 280 outliers,leaving a dataset of 888 samples for analysis.We developed and assessed an automated machine learning(AutoML)model for predicting UCS,introducing a novel approach to tackle this prediction challenge.Additionally,we compared models enhanced by Bayesian optimization,including multi-layer perceptron(MLP),support vector machine(SVM),Gaussian process regression(GPR),and K-nearest neighbor(KNN).Among these,the AutoML model demonstrated superior performance in UCS prediction,offering a rapid and efficient method for estimating UCS in engineering applications and enabling intelligent classification of rock masses.The study also evaluated the sensitivity and contribution of SRn,Vp,and Is in UCS estimation by various techniques,including permutation feature importance(PFI),SHapley Additive exPlanations(SHAP),and local interpretable model-agnostic explanations(LIME).The results underscore that the AutoML approach not only streamlines UCS modeling but also provides a robust and comprehensive solution,significantly enhancing the accuracy and efficiency of the prediction process.
基金supported by the National Natural Science Foundation of China under Grant No.42177140the Key Research and Development Project of Hubei Province of China under Grant No.2021BCA133.
文摘Engineering disasters, such as rockburst and collapse, are closely related to structural instability caused byinsufficient bearing capacity of geological materials. Uniaxial compressive strength (UCS) holds considerablesignificance in rock engineering projects. Consequently, this study endeavors to devise efficient models for theexpeditious and economical estimation of UCS. Using a dataset of 729 samples, including the Schmidt hammerrebound number, P-wave velocity, and point load index data, we evaluated six algorithms, namely AdaptiveBoosting (AdaBoost), Gradient Boosting Decision Tree (GBDT), Extreme Gradient Boosting (XGBoost), LightGradient Boosting Machine (LightGBM), Random Forest (RF), and Extra Trees (ET) and utilized Bayesian Optimization (BO) to optimize the aforementioned algorithms. Moreover, we applied model evaluation metrics suchas Root Mean Square Error (RMSE), Mean Absolute Error (MAE), Variance Accounted For (VAF), Nash-SutcliffeEfficiency (NSE), Weighted Mean Absolute Percentage Error (WMAPE), Coefficient of Correlation (R), and Coefficient of Determination (R2). Among the six models, BO-ET emerged as the most optimal performer duringtraining (RMSE ¼ 4.5042, MAE ¼ 3.2328, VAF ¼ 0.9898, NSE ¼ 0.9898, WMAPE ¼ 0.0538, R ¼ 0.9955, R2 ¼0.9898) and testing (RMSE ¼ 4.8234, MAE ¼ 3.9737, VAF ¼ 0.9881, NSE ¼ 0.9875, WMAPE ¼ 0.2515, R ¼0.9940, R2 ¼ 0.9875) phases. Additionally, we conducted a systematic comparison between ensemble andtraditional single machine learning models such as decision tree, support vector machine, and K-NearestNeighbors, thus highlighting the advantages of ensemble learning. Furthermore, the enhancement effect of BO ongeneralization performance was assessed. Finally, a BO-ET-based Graphical User Interface (GUI) system wasdeveloped and validated in a Tunnel Boring Machine-excavated tunnel.
