| With the rapid development of national economy,higher requirements have been put forward for the convenience and development of railway and highway transportation system,which also makes more Crossing-Mountain tunnels in continuous development and construction.Many cross-ridge tunnels will have multiple hazards from Karst erosion when crossing karst development zones,such as high-pressure water gushing,mud gushing and sand gushing.In the southwest of China,when the tunnel crosses the karst area,most of the water inrush disasters occur.A lot of scholars have studied the prediction of water inrush in tunnels and made great achievements.At present,the methods used in the prediction of water inrush in tunnels are mainly based on the theory of water balance and groundwater dynamics and numerical simulation.Because the geological structure of most mountain-crossing tunnels is too complex,and they pass through karst zones and regional watersheds,it brings great difficulties to traditional theoretical calculation methods and numerical simulation.In practical engineering applications,because of the complex karst geological structure and the inaccuracy of hydrogeological parameters,some actual hydrogeological conditions are often simplified or removed artificially,and often replaced by empirical values.Limited by personal knowledge,some important information is forgotten,which makes the prediction accuracy of tunnel water inflow very low,and can not provide credible prediction value for actual tunnel excavation.From the point of view of complex factors affecting water inrush in tunnels,the actual cases of water inrush disaster in Karst Crossing-Mountain tunnels built in southwest China are collected,and the support vector machine model of water inrush is established by using machine learning method,which adds luster to the traditional water inrush prediction method.Based on the data collection of karst mountain-crossing tunnel and the actual field exploration practice I participated in,this paper uses machine learning method to analyze and predict the construction of Chongqing-Kunming high-speed railway line project based on the case of water inrush disaster of karst mountain-crossing tunnel in southwest China.By collecting and sorting out a large number of data samples,we can find out the factors affecting the inrush disaster and carry out grading control processing,which provides a great convenience for the establishment and testing of the model,greatly improves the accuracy of the model,and lays a solid foundation for better prediction of the inrush level.The following are the main research results:(1)Based on the data of water inrush in 105 karst mountain-crossing tunnels in southwest China,the disaster-causing factors of water inrush in Karst mountain-crossing tunnels in southwest China are compared and explored.Referring to the academic research of predecessors,this paper makes a statistical analysis and Discussion on the influencing factors of water inrush tunnel in southwest China,and analyses the influencing factors of karst water inrush disaster from seven aspects:rock solubility,geological structure type,surrounding rock classification,rock thickness,rock dip angle,hydrodynamic zoning,topographic gradient and rainfall infiltration coefficient.(2)Based on the actual data of water inrush in Karst mountain-crossing tunnels in southwest China,the relatively complete and accurate water inrush sections in 45existing tunnels are selected as the sample set for calculation,and the influencing factors with high frequency in the collected data samples are selected.Thus,eight indexes of support vector machine are constructed,which are rock solubility,geological structure type,surrounding rock grade and rock.Thickness of stratum,dip angle of stratum,hydrodynamic zoning of tunnel,topographic gradient and rainfall infiltration coefficient.(3)The selected influencing factors are processed by hierarchical control,and a hierarchical system of influencing factors is established.On the index control level,the principle of evaluating the risk of water inrush from Karst Tunnels in southwestern mountainous areas established by Yang Yanna is used for reference.The missing data are filled up by K-neighboring method,and the problem of unbalanced data samples is solved by SMOTE method to ensure more accurate prediction results.Except for the interference of missing data on the prediction results.(4)Model comparison:Random forest model has a F1 score of 60%,Naive Bayesian model has a F1 score of 46%,and Support Vector Machine(SVM)model has a F1 score of 66%.This shows that Support Vector Machine(SVM)model is more accurate than Random Forest and Naive Bayesian model in forecasting water inflow,which has a strong guiding significance for practical work.(5)Through data collection and analysis,the correlation between influencing factors and water inrush is explored from the perspective of machine learning,and the mathematical statistical model of multi-factors is obtained by using support vector machine(SVM).The F1 scores of linear-svm model,polynomial kernel support vector machine(poly-svm)model and Gaussian kernel support vector machine(rbf-svm)model are 0.66,0.62 and 0.69 respectively.Gaussian kernel support vector machine(rbf-svm)model is better than the other three models,so this paper chooses Gaussian kernel function to construct support vector machine(SVM)model.(6)The genetic algorithm(GA)is used to optimize the parameters of penalty factor C and gamma in the kernel function,and the support vector machine(GA-SVM)model optimized by GA is obtained.The final prediction result of GA-SVM model is9%higher than that of SVM model without GA treatment.(7)The GA-SVM model is used to predict the water inrush grade of the proposed Yukun high-speed railway.The results show that the water inrush grade of a few sections of the Yukun high-speed railway is 2(100-1000m~3/d),and that of most sections is 4(10000-100000m~3/d).This provides a good reference for the excavation of Chongqing-Kunming high-speed railway. |
PDF Full Text Request