A novel hybrid model based on PCA-EEMD-LSTM neural network for short-term landslide prediction

Abstract

It is of great practical significance for landslide prediction, because its uncertainty has brought great harm to the safety of human life and property. This study proposes a new algorithm that combines Principal Component Analysis, Ensemble Empirical Mode Decomposition, Long Short-Term Memory Network which establishes a PCA-EEMD-LSTM combined model to predict the cumulative displacement of the landslide. Through PCA, it is found that temperature, relative humidity, and wind speed (east-west, north-south) are the important meteorological factors affecting the landslide in this case. Then we decompose the entire displacement into sub-sequences of different frequencies through EEMD, predict each sub-sequence separately through LSTM, and finally reconstruct all sub-sequence predictions to get the final prediction result. We not only compare the difference between the predicted value of the model and the actual measured value, the accuracy of the four models of PCA-LSTM, EEMD-LSTM, ELM, and BP after training under the same data set conditions are also compared. The results show that the short-term prediction effect of the PCA-EEMD-LSTM model is better than other models. Compared with the conventional landslide prediction model, this model has a shorter time span, and has higher accuracy and stability, which is of great significance for landslide prediction.