Landslide Prediction Based On Long And Short-Term Memory Network

Landslide disaster has always had an irreversible impact on the safety of people’s lives and property,and the state has paid more and more attention to the occurrence and protection of landslide disaster.Under this background,the prediction of landslide has become an urgent problem to be solved.This paper mainly focuses on the prediction of cumulative displacement of landslide.The main research contents and conclusions are as follows:1.The cumulative displacement data of landslide has high complexity,which is embodied in high nonlinearity and instability.To solve this problem,this paper decomposes the cumulative displacement sequence of landslide by using ensemble empirical mode decomposition,obtains the modal characteristic information of the cumulative displacement sequence of landslide under different frequencies and time scales,establishes LSTM model for each IMF component after decomposition for prediction,and establishes EEMD-LSTM model for verification.2.This paper analyzes the impact of 10 meteorological factors on landslide,because there is great redundancy between each component,which will lead to the decline of prediction accuracy.In order to solve this problem,principal component analysis is used to reduce the dimension of meteorological factors,which not only reduces the dimension of data,but also avoids the loss of relevant information.Therefore,the LSTM neural network model of principal component optimization is established.3.There is no specific formula for the parameter selection of LSTM model,which can only be set according to experience,which will affect the final result.In this context,this paper selects whale algorithm to optimize the weight and bias of LSTM neural network,and establishes WOA-LSTM model.In order to further improve the prediction accuracy of the model,multi-point joint prediction is added to optimize the model,and the relevant models are compared and verified.4.In order to obtain a more stable landslide cumulative displacement prediction model,this paper combines the above models and extracts the multi-point PCA-EEMD-WOA-LSTM algorithm model.The model comprehensively considers the respective advantages of several algorithms,and solves the problems of high dimension of meteorological factors,complex cumulative landslide displacement data and difficult parameter selection of LSTM network.In the prediction process of the model,the other points with high correlation with the monitoring points and the principal components after PCA dimensionality reduction are respectively input into the WOA-LSTM model corresponding to each IMF component after EEMD decomposition for prediction.Finally,each result is reconstructed to obtain the final prediction value.In this paper,the model is applied to the short-term landslide displacement prediction,and the prediction results are compared with the other 10 models.It is found that the RMSE of the predicted values of the model at the two monitoring points are 2.6140 and 3.0783,MAPE are 0.6304 and 0.6447,and R~2 are 0.9731 and 0.9726 respectively.The evaluation indexes of the model are better than the other 10 models,which proves the superiority of the model in this experiment.