Research On Prediction Of Landslide In Xishan Mountain Area Of Taiyuan Based On Recurrent Neural Network

Landslides have always ranked fifth in the frequency of global geological disasters,and are more frequent in China,ranking first.In 2019,landslide disasters caused a total of 211 deaths,13 missing,75 injured,and direct economic losses.2.77 billion yuan,causing huge economic losses.At present,the mechanism of landslide occurrence is not fully understood.The relationship between various factors and landslide deformation cannot be simply described by algebraic formulas,and the displacement of landslide deformation has complex nonlinear characteristics.For this,researchers have established many The prediction accuracy of the model is high or low,and it cannot meet the final effect in the actual disaster prediction in a changing and complex environment.Based on the above problems,this paper aims to find the hidden relationship between the amount of landslide deformation and each factor based on deep learning's ability to extract and recognize data features through machine learning,so as to achieve the prediction effect.This article mainly takes two typical landslides in Xishan mountain area of Taiyuan as examples,and introduces the application process of landslide prediction based on recurrent neural network in detail.The main research work is as follows:1.Combining the processing and analysis of remote sensing images,field surveys and field survey data,combined with previous theoretical research on landslide mechanism,the formation mechanism of landslide geological disaster body in the study area is analyzed,and the analyzed landslide.formation mechanism is Based on the theoretical basis,the stability analysis of landslide geological disasters No.1 and No,2 were carried out.2.through the analysis of the landslide mechanism in the study area,combined with the previous research on the landslide sensitivity factors.The relevant factors were extracted and regularized on the deformation displacements of No.1 and No.2 landslides in the study area.The main input factors of the landslide prediction model were obtained,and the factors were input into the machine learning model for prediction.3.In this paper,after data preprocessing and dimension construction of model factors and landslide deformation displacements,important parameters such as batch_size,epoch,hidden_size,time_step and other important parameters of the model are analyzed by trial and error method,aiming at the overfitting phenomenon of LSTM model on the research data Perform L1,L2 regularization and function gradient optimization,and point out the selection of relatively optimal regularization and gradient optimization functions;similarly,the support vector machine is simply optimized,and the performance of the main different kernel functions is compared and compared.select.Finally,the quasi-optimal parameters of the prediction model for the deformation displacement and related factors of No.1 and No.2 landslides in the study area are obtained.The final predicted value of landslide deformation displacement is obtained through the model,and the predicted values of two different models of shallow machine learning SVM and deep learning LSTM are compared and analyzed.Finally,a more reliable prediction accuracy is obtained.The more reliable prediction accuracy obtained through the above research,the combination of landslide deformation variables and landslide impact factors through machine learning for predictive analysis provides a certain reference value and scientific basis for research in related fields,and adds one to the disaster prevention in Shanxi Province.To provide a scientific basis for the realization of a cloud platform for geological disasters across the country.