A variable weight combination model for prediction on landslide displacement using AR model, LSTM model, and SVM model: a case study of the Xinming landslide in China

It is necessary to improve the accuracy of the prediction on landslide displacement owing to its danger to the local environment and residents. However, it is difficult for the constant weight combination models widely used now to apply to the actual situation because of the complexity of the coupling relationship between the actual displacement and prediction model. Therefore, we develop a novel combination model using variable weights. The variable weight combination (VWC) model is proposed using the autoregressive (AR) model, long short-term memory (LSTM) model, and support vector machine (SVM) model, and the weights of the three individual models are comprehensively analyzed by the errors between the actual displacements and their prediction results. The weights are continuously optimized as the periods increase to optimize the VWC model, and it retains the advantages of the individual models and useful information in the individual models. Taking the Xinming landslide as an example, displacements data of nine sites are collected. The prediction displacements are obtained using the AR model, LSTM model, SVM model, and VWC model and compared with monitoring displacements using nine performance measures. The comparison results show the prediction precision using the VWC model is more satisfactory than that of individual models, and the VWC model is, therefore, more applicable to the study landslide.