A novel data-driven hybrid intelligent prediction model for reservoir landslide displacement

Abstract

Accurate and reliable displacement prediction is crucial for landslide monitoring and early warning. Landslide displacement data is complex nonlinear time series. Although some studies have employed dynamic models to predict landslide displacement, they have only focused on point displacement prediction, inevitably compromising the prediction credibility due to the inherent uncertainties in landslide prediction. This paper proposes a novel hybrid intelligent prediction model to enhance the prediction accuracy of point displacement in reservoir landslides and construct reliable displacement prediction intervals. Specifically, PSO-SVM is adopted to predict the trend displacement, while CNN-GRU-Attention is designed to predict the periodic displacement. Furthermore, the hybrid model allows for the direct construction of required displacement prediction intervals based on the landslide time series. The superior performance of the proposed model is proven by using the Baishuihe and Shuping landslides as case studies. The results demonstrate that the developed model achieves higher prediction accuracy and enables the construction of reliable displacement prediction intervals. Additionally, the proposed model can predict the time series of unknown displacement and provide an early warning of landslides at the early stage of displacement mutation. This research contributes to the improvement of landslide risk assessment and disaster early warning capabilities, providing reliable scientific guidance for landslide disaster prevention and mitigation.