| The planning new railway line from Dujiangyan to Siguniangshan(short as Du-Si railway)in the transitional area between the first and the second staircase.Affected also by Longmenshan Fault zone,the geological structure in this area which combined with deep gullies,high moutains and steep slopes is very complicate and unsteady.Intense neotectonics movements and earthquakes are highly active there,making debris flows inevitably the most common outcome.Loose soils formed after earthquakes in Wenchuan and Lushan make these debris flows in much larger scales and more destructive once taking place.And extreme weather condition and engineering activities make debris flow thrive in quantity.To monitor these large-numbered gullies,traditional methods are severely insufficient,and sticking to them will greatly slow the process of Du-Si railway project and more importantly threaten lives and properties in this area.This paper takes 1599 square kilometers of the area where the proposed railway is located as the research area,A total of 96 debris flow in the area were identified through field surveys,remote sensing interpretation,data collection and other research methods,At the same time,a detailed database of 96 debris flow was established.Based on the database,the characteristics of the debris flow were counted and analyzed,and it was recognized that the area is dominated by medium-sized debris flows with small watersheds and steep slopes,which are generally distributed along the banks of the river,which are widely expressed on both sides of the mountain range and around the mountain basin,which has the characteristics of obvious siltation and densely developed with rainfall.Based on the mature theoretical basis of neural network and a wide range of application cases,combined with the development characteristics of debris flow in the study area,A total of 11 eigenvalues are used as the evaluation factors of debris flow susceptibility in the research area,including the watershed area,main gully length,average slope drop,maximum elevation,minimum elevation,maximum elevation difference,static reserves,dynamic reserves,rain intensity.At the same time,a debris flow susceptibility evaluation model based on a multi-layer perceptron was constructed.The model input and output layers represent 9 evaluation factors and 3 levels of susceptibility.The middle layer uses Sigmoid,tanh,Re LU and Leaky Re LU as activation functions and configures a different number of hidden layers.By using 66 debris flow data to train the model,four optimal debris flow susceptibility evaluation models under different activation functions are obtained.In order to verify the applicability of the obtained model,the training model was used to evaluate the susceptibility of the remaining 30 debris flows in the study area.The Leaky Re LU activation function model obtained 19 trenches with moderate susceptibility and 11 trenches with mild susceptibility.As a result,the agreement with the result obtained by the quantitative scoring method is as high as 0.8889.In addition,the verification results of mean impact value method show that the selection of evaluation factors is reasonable.It is finally concluded that the Leaky Re LU activation function model has the best stability and high accuracy for the debris flow susceptibility in the study area,and it is not easy to disappear the gradient and there is no overfitting,the neural network model based on the Leaky Re LU activation function can quickly and accurately determine the susceptibility of debris flow in the study area,and then the debris flow with a higher susceptibility level can be further investigated in detail or the monitoring and early warning program with "one ditch and one strategy" can be formulated in a targeted manner.It not only saves a lot of manpower and material resources,but also guarantees the life and property safety of the people in the research area. |
PDF Full Text Request