Research On Collapse Risk Assessment For Mountain Tunnel Construction Based On Machine Learning

The drilling and blasting method is widely used in mountain tunnel construction because of the simplicity,economy and efficiency of the process.However,mountain tunnels always face many uncertainties in drilling and blasting construction,resulting in frequent collapse disasters with adverse consequences such as casualties,property losses and schedule delays.In order to address the shortcomings of the current tunnel collapse risk assessment,such as low accuracy,lack of targeted content,and weak prediction,this study adopted various methods such as machine learning,multisource data fusion,and numerical simulation to conduct the research work of collapse risk assessment based on the Jinzhupa Tunnel Project of the Fujian Puyan Expressway.The main research work and conclusions are as follows:(1)A fusion intelligent risk assessment model for tunnel collapse was developed to realize the automation,intelligence and high accuracy of the assessment process.First,the literature analysis method was used to collect tunnel collapse risk assessment cases.Then,the main influence factors were identified and the corresponding risk index system was established.Next,four intelligent assessment models were developed based on commonly used machine learning classification algorithms and the collected cases.After comparing the performance of the above models,a fusion model consisting of the SVM model and the RF model was developed by the improved DempsterShafer(D-S)evidence theory,which has a higher accuracy than a single model in risk assessment.(2)A novel calculation method for collapse risk probability of tunnel was proposed.By introducing the idea of reliability analysis method,the proposed method integrates numerical method,support vector regression(SVR)and Monte Carlo method,using the safety factor of the surrounding rock as the random output variable and the reliability index,can quantify the collapse likelihood under specific construction behavior.The reliability and efficiency of the proposed method was verified by the collapse analysis of the Jinzhupa Tunnel.(3)A dynamic assessment model for collapse risk was developed based on intelligent prediction of surrounding rock deformation.First,an intelligent prediction model for the deformation of the surrounding rock after tunnel excavation and support installation was proposed based on the rolling forecasting method and multi-output least-squares support vector regression.Compared to standard SVR,the proposed model has higher accuracy and efficiency.Then,the collapse risk was dynamically predicted using the cloud model and improved D-S evidence theory based on the deformation prediction results.The proposed deformation prediction based dynamic risk assessment method is prospective.