Abstract:

A comprehensive and sophisticated multi-algorithm coupled dynamic prediction model is proposed to address the intricate reality and stringent accuracy requirements of predicting tailings dam displacement. Firstly, by employing a time series decomposition model, the cumulative displacement is disaggregated into its trend and cyclical components. The trend term displacement is then forecasted using a Gaussian regression time series prediction model. Secondly, various Copula functions are employed to investigate the overall correlation between the inducing factors and the cyclical term displacement. Owing to the diverse influencing factors and strong nonlinearities associated with the cyclical term displacement, the MISSA-CNN-BiLSTM model is utilized for prediction. Lastly, the predicted trend term displacement from the Gaussian regression model and the predicted cyclical term displacement from the MISSA-CNN-BiLSTM model are merged. The results demonstrate a high degree of consistency between the predicted cumulative landslide displacements and the measured values, with a correlation coefficient of 0.996 and a root mean square error (RMSE) of 0.13 mm. The multi-algorithm coupled model, based on MISSA-CNN-BiLSTM, exhibits remarkable prediction accuracy and effectively captures step changes in tailings dam displacements.

Key words: multi strategy improved sparrow search algorithm(MISSA), convolutional neural networks(CNN), Bi-directional long short-term memory(BiLSTM), tailing dam, displacement prediction, deep learning model