Abstract:

The noise interference in the displacement monitoring data of the open-pit mine slope seriously affects the accuracy of landslide prediction. To address this issue, a hybrid prediction model integrating data smoothing and deep learning was proposed in this paper. Firstly, the WT was used to denoise the original time series to extract the real deformation trend component. Then, based on the smoothed high-quality data, the key parameters of the LSTM network were optimized, and an optimized LSTM prediction model was constructed to learn the complex long-term dependence of slope displacement. The results show that compared with the single LSTM benchmark model that directly uses the original data, the prediction accuracy of the optimized combination model is effectively improved. The root mean square error (RMSE) is reduced by 86.3%, and the coefficient of determination (R²) is increased from 0.51 to 0.76, which proves that it can effectively improve the accuracy and timeliness of the prediction and has high engineering practice value.

Key words: wavelet transform (WT), long short-term memory (LSTM) network, slope displacement prediction, monitoring data, prediction model