Risk Prediction of Water Inrush in Karst Tunnels Based on DNN and Attribute Mathematics
Water inrush is the most common disaster during karst tunnel construction,and it causes tunnel mud collapse,support,equipment damage,and substantial construction personnel casualties;therefore,predicting tunnel water inrush disasters is necessary.In this study,a method for predicting tunnel water inrush disasters based on a deep neural network(DNN)and attribute mathematical theory was proposed.First,based on an investigation of 103 karst water inrush tunnels worldwide,seven karst disaster indices and three disaster consequence indices were screened,and an extensive database comprising 489 datasets was established.Subsequently,the K-means clustering algorithm was introduced to cluster the tunnel water inrush data.Based on the clustering results and parameters of disaster consequences,the risk levels were classified into four classes,and reasonable distribution intervals of the calamitous parameters corresponding to each class were presented.Based on this,four algorithms,which were DNN,random forest,K-nearest neighbors,and AdaBoost,were employed to predict the levels of karst tunnel water inrush disaster.Furthermore,based on the attribute mathematical theory,the probability prediction of the karst tunnel water inrush disaster levels was achieved.Finally,the proposed method was applied and validated in engineering projects,such as the Jigongling Tunnel in Hubei Province.Compared with previous methods for predicting karst water inrush disasters,this study combines DNN and attribute mathematics theory to realize the automatic classification of karst tunnel water inrush disaster levels and obtain the probability,which is more objective and comprehensive for predicting actual karst tunnel water inrush disasters.
NETL
NSTL
万方数据
