Landslide Prediction Based Ondistributed Hydrological Model And Deepneural Network

Geological disasters,such as landslides and mudslides,have caused enormous damage to social economic development and the safety of people's lives and property.Yunyang County in Chongqing is located in the southwest of my country and northeast of Chongqing,whose terrain is in the first ladder.The terrain of Yunyang County is characterized by multiple mountains and steep slopes.In summer,affected by heavy precipitation and the water level of the Three Gorges Reservoir area,geological disasters frequently occur in the region.In this paper,the area around Xuejiagou,Yunyang County,Chongqing City,is selected as the research area,and the slope stability is studied by using the limit equilibrium model of landslides.In this paper,a distributed hydrological model is used as the calculation model for the soil moisture content of the underlying surface;a deep neural network(DNN)is used to study the relationship among soil moisture content,terrain slope,rock & soil distribution and landslides.Therefore,a landslide prediction model is established.The main research contents of this paper are as follows:(1)A distributed hydrological model is built based on the precipitation data and runoff coefficient of the study area.According to this distributed hydrological model,the change process of soil water content in the study area under the condition of time series precipitationl data can be simulated and calculated.The study basin belongs to the area without runoff data,so this article first calculates the runoff coefficient of the study area according to the distributed hydrological model,and compares the model parameters with the measured runoff coefficient to correct the model parameters.The calculated runoff coefficient in the study area in the past 10 years is close to 0.40,while,the measured runoff coefficient 0.41,indicating that the model constructed in this paper has a strong ability to simulate runoff generation in the basin.(2)A limit equilibrium model of landslide is constructed based on the data of geotechnical mechanics in the study area and the existing research results.This model is used to analyze the slope stability under three different precipitation conditions in Xuejiagou.With the increase in precipitation,the number of unstable grids in the area began to increase.When the precipitation in the study area is 0,the proportion of unstable grids in the area is less than 0.2%;when the precipitation is 35 mm,the unstable grids in the area increase to 1.08%;when the precipitation is 70 mm,The unstable grids in the region increase to 1.38%;when the precipitation continues to increase,the unstable area inside the region will not increase continuously,but maintain at a certain amount.There is a certain relationship between landslide and slope.The larger the slope,the greater the proportion of unstable areas is.There are all in an unstable state for the extremely steep slope areas,and the unstable areas account for the largest proportion;most of the steep slope areas are in an unstable state.The ratio exceeds 2/3 of the steep slope area;and there are no unstable grids on the micro slopes with a slope of less than 15°,indicating that when the slope is small,some slopes will not have landslides under the influence of heavy precipitation.(3)to predict regional landslides separately.The optimal number of hidden layer neurons of the two model algorithms are discussed separately.The optimal number of hidden layer neurons of the BP-ANN model is 21,and the optimal number of hidden layers of the DNN model is 6.The nodes for 6 layers are 20,32,30,30,30,and 10,respectively.During the training process,the minimum Nash coefficient of the BP-ANN model is 0.521 and the maximum is 0.758.When the DNN model adjusts the 6th hidden layer,the minimum Nash coefficient is 0.610 and the maximum is 0.805.After comparing and analyzing the results of the two neural networks,it is found that the Nash coefficient of the DNN model is better,indicating that this model is more capable of mining feature information among data,and has a better effect in predicting landslides.