Research On Displacement Prediction Method Of Step Type Landslide Based On VMD And Hybrid Deep Learning Model

In recent years,due to the influence of reservoir water level and rainfall,landslide disasters occur frequently in the Three Gorges area of China.At present,there are 4200 landslides.The establishment of an efficient landslide displacement prediction model can more intuitively reflect the landslide deformation trend,which is an effective means to solve the problem of landslide disaster.Facing the step landslide with complex displacement curve,the variational modal decomposition(VMD)method can adaptively decompose the total displacement into trend term displacement and periodic term displacement.The prediction of each component displacement alone can reduce the disturbance of the sharp increase of displacement curve caused by environmental factors in the process of overall displacement prediction,and make the prediction results more accurate.Facing the landslide displacement prediction problem with complex causes,deep learning can mine the deep-seated characteristics in the actual monitoring data,and train the model based on a priori knowledge,so as to realize the accurate prediction of displacement.Therefore,this study combines VMD with depth learning algorithm,and takes Baijiabao Landslide in the Three Gorges Reservoir Area as an example to study the landslide deformation.The main work is as follows:(1)Analyzing the basic general situation of Baijiabao landslide,and aiming at the difficulty of fitting the displacement curve of the step landslide in this area,the VMD method is used to decompose the total displacement of the landslide into periodic displacements caused by the trend displacement evolved by the internal structure of the slope and external induced factors.Because VMD can clearly set the decomposition quantity,it can simplify the process of displacement decomposition and make each displacement component more physical.At this time,the characteristics of each displacement curve are obvious,and subsequent experiments can predict each displacement separately to obtain more accurate prediction results.In view of the reasonable selection of landslide inducing factors,the influencing factors of landslide are analyzed in depth from both qualitative and quantitative angles,and the key inducing factors of landslide deformation are screened by gray correlation analysis method,which provides effective data support for accurate prediction of landslide displacement.(2)Aiming at the limitation of single depth learning model in multi inducement step landslide displacement prediction,a parallel TCN-GRU model is proposed for landslide single point displacement prediction.The model is composed of TCN module and Gru module with parallel structure.Combined with the efficient extraction ability of TCN for data features and the memory ability of GRU for long-term or short-term information of time series data,the model can make full use of the existing landslide measured data,extract higher-order trigger factor features and time series data features from the input data,and effectively improve the displacement prediction performance.The simulation results show that TCN-GRU model is superior to other benchmark models in prediction accuracy and model fitting ability.(3)Aiming at the problem of low prediction efficiency of multi-point multi-time model of landslide in the region,a multi-point joint displacement prediction model based on MCSGRU is proposed.The model extracts the spatial distribution characteristics of different receptive field sizes between monitoring points through multi-scale convolution,so as to mine the internal relationship between adjacent landslide monitoring points.Through the channel attention mechanism,the channel weight of features is dynamically adjusted to improve the feature expression ability of the network.The GRU module is used to identify the time dependence of the input information to realize the temporal feature extraction.Finally,the extracted multiple features are spliced and fused,and multiple displacement prediction values are obtained through linear regression calculation.The displacement prediction results show that MCSGRU multi-point joint prediction model has higher prediction accuracy and prediction efficiency.