Research On Landslide Deformation Prediction Based On Improved GM(1,1) Model

Landslide is a common and harmful geological disaster,which seriously threatens people’s life and property safety.Monitoring,forecasting and forecasting of slope deformation is one of the important means of landslide disaster prevention and control.This paper starts with the historical landslide monitoring data,and takes the traditional grey GM(1,1)as the basic model to carry out the related improvement research on the slope deformation prediction model.(1)The paper briefly introduces the modeling method of the traditional GM(1,1)model,and discusses the typical domestic landslides in my country,such as the Danba landslide,the Huangci landslide,the Gushuwu landslide,and the Xintan landslide.The prediction performance and scope of application of the GM(1,1)model show that GM(1,1)is more suitable for landslides with upward concave deformation curves such as the Huangci landslide,and the length of the modeling sequence cannot be too large,otherwise the model will be affected.prediction accuracy.(2)Aiming at the shortcomings of traditional GM(1,1)in the applicability analysis of landslide cases,the discrete gray prediction model ADGM based on adjacent accumulation gray series and the combination optimization of background value and initial value parameters are studied.Grey prediction model GM(1,1).In view of the influence of modeling sequence length on model prediction performance,particle swarm algorithm PSO is used to optimize the modeling sequence length and model parameters of the improved model,and finally build a PSO parameter-optimized,dynamic,and improved gray prediction model.Then,the improved model is exemplified by the landslide example.The experimental results verify the validity of the improved model.The improved model can improve the prediction ability of the slope deformation and expand the applicable scope of the slope deformation prediction.(3)In view of the uncertainty of the internal deformation mechanism of the slope and the limitations of a single prediction model,the paper uses the GM(1,1)model and the long-short-term memory neural network model LSTM as the basic model.From the perspective of perspective,two types of combined forecasting models are constructed:weight-based combined forecasting models and combined forecasting models that consider external factors.(1)According to different weighting methods,three combined prediction models of "optimal weight","grey comprehensive correlation degree weighting" and "information entropy weighting" are respectively constructed.The Gushuwu landslide and Xintan landslide are taken as the An example is used to compare the forecasting effect of the combined forecasting model under different fixed weighting methods.The results show that the combined forecasting model based on the fixed weight of grey comprehensive correlation degree has higher forecasting accuracy;(2)Taking Dashuitian slope,Wachangping slope and a landslide in the Three Gorges Reservoir area as verification examples,a combined prediction model based on time series decomposition is established.Method and wavelet analysis method decompose the deformation displacement into trend term and random term displacement,then use the GM(1,1)model to fit and predict the trend term,and use the LSTM algorithm to approximate the random displacement under the influence of the inducing factors.The predictions of the individual models are combined.The calculation example results show that this kind of combined model can effectively improve the prediction accuracy and adaptive performance of the original GM(1,1),and the selection of the decomposition method is also a factor that affects the prediction results of the combined model.Relatively speaking,the prediction accuracy is higher and the adaptation performance is better.