210 / 2024-03-13 13:54:51

基于二次模态分解与多模集成的滑坡位移多步滚动预测

滑坡位移预测；,模态分解,多模集成；,滚动预测；

The landslide disaster seriously endangers the safety of people's lives and property, and the landslide early warning based on radar monitoring slope displacement has been widely used. However, the current early warning often only uses historical displacement, and the warning time and accuracy are lower, If the multi-step rolling accurate prediction of slope displacement can be made, it is expected to effectively increase the early warning duration and accuracy. However, the slope displacement data monitored by radar contains noise and has strong fluctuation, so the accurate prediction of multi-step rolling is a great challenge. Aiming at this, this paper proposes a multi-step rolling prediction algorithm of landslide displacement based on quadratic mode decomposition and multi-mode integration. Firstly, empirical mode decomposition (EMD) is used to decompose the radar landslide displacement once to obtain the trend term and the periodic term, so as to remove the measurement noise. For the mode aliasing of its periodic terms, the fully integrated empirical mode decomposition (CEEMDAN) of adaptive noise is further designed to reduce the fluctuation of the periodic terms; Then, the cubic spline interpolation (spline) model was used to predict the trend term of the primary decomposition of EMD, and the periodic term after the quadratic decomposition was reconstructed into a sequence containing strong, medium and weak fluctuation terms, and proposed a multi-mode integrated prediction mechanism based on long and short term memory network (LSTM), gated cycle unit (GRU) and spline model. Finally, the update criteria of each model are designed for the period term and trend term respectively, and the multi-step rolling prediction is realized based on the new monitoring data. The proposed algorithm is applied to actual monitoring data, and the results show that the proposed method can effectively increase the prediction time and improve the prediction accuracy.