The Correlation Study Between Rainfall And Water Inrush In Sangzhi Tunnel Of Qianjiangzhangjiajie-Changde Passenger Line

It has the most extensive and the most developed karst geologyinsouthwestern China.A large number of mountain tunnels are inevitably built to cross these forbidden areas.The frequent occurrence of flood water disasters not only leads to the extension of tunnel construction period,but also threatens the safety of construction personnel,and may affect the quality of the tunnel,which is a hidden danger for the safety operation of the railway.Therefore,how to accurately predict the inflow of water in tunnel construction has become a key problem in tunnel engineering.This paper is based on the newly constructed Sangzhi tunnel of Qianjiang-Zhangjiajie-Changde passenger line,discussed the hysteresisrelationship between the rainfall and water gushing of the tunnel from April 1 to July 14 by thecorrelation analysis and other methods,predicted the tunnel water gushing by the model that combinated the single multivariate linear regression model,BP neural network model and BP-multiple linear regression,has the following results:(1)The water inflow from April 1 to July 14 was highly correlated with the daily rainfall in Sangzhi tunnel.Under the regulation of karst aquifer the rainfall of 16~24 hours on that day had little effect on the water flow of the tunnel,and the precipitation was highly correlated with the rainfall at the time of 0~8,8~16,0~8,8~16,and 16~24 by examing the correlation between precipitation and water inflow.(2)A single multivariate linear regression model can not fit into the nonlinear relationship of rain-gush water in tunnel,and the overall prediction accuracy is difficult to meet the requirements.On the prediction results,the strong nonlinear approximation ability of BP neural network can reflect the tendency of changes in tunnel water gushing.And overall prediction accuracy is higher,at 15.65%,which can be used as a forecasting model to warn and to provide reference for the development of short-term drainage plan of tunnel.(3)The advantages of linear partial fitting and the characteristics of nonlinear arbitrary approximationof BP neural network are analyzed,and combined them.BP neural network was used to modify the residual error of multiple linear regression model,and it was applied to the prediction of tunnel inflow that the multiple linear regression model to correct the residual error of BP neural network was established.The results showed that The prediction results are better than a single prediction model,both in the single-day prediction error and the overall mean absolute error.The model Can be better applied to practical engineering.(4)Taking the cumulative rainfall of the current day and the previous day as variables,Using the BP-Multiple Linear Regression Combination Model to predict the gushing water risk level corresponding to the predicted value,forecasting the gushing water risk at different rainfall levels during the rainfall-intensive period of Sangzhi Tunnel.