Adit deformation prediction based on ICEEMDAN dispersion entropy and LSTM-BP

Abstract

Changes in the overburden load and dynamic compaction disturb the upper part of the mine’s adit. Some positions of the adit’s surrounding rock are cracked, and the roof, floor, and two sides are deformed to varying degrees. Accurate and effective adit deformation prediction is helpful to the underground safety decision. Considering the accuracy of adit deformation measurement and the complexity of the underground environment, fiber Bragg grating (FBG) advantages are combined for monitoring the adit deformation. The adits’ deformation during coal mining is predicted by combining the improved adaptive noise complete ensemble empirical mode decomposition (ICEEMDAN) and dispersion entropy analysis with the GA-optimized short and long-term memory network and back-propagation neural network (GA-LSTM-BP) prediction model. This study compares the No. 5 FBG strain gauge to seven other prediction models using it as an example. Compared to other comprehensive algorithms, absolute error (MAE), relative error (MSE), and root mean square error (RMSE) decrease by 0.05–3.07, 0.05–26.96, and 0.13–4.53, respectively, and linear regression correlation coefficient (R2) reaches 0.99. This shows that the model has higher accuracy in predicting adit deformation. The model is applied to the No. 3 FBG strain gauge and No. 1 FBG proximity gauge data for verification and comparative analysis. It also indicates that the accuracy of this model for adit deformation prediction is higher than that of other comprehensive ones.