The Forecasting of Surface Displacement for Tunnel Slopes Utilizing the WD-IPSO-GRU Model

Abstract

Continuous displacement prediction of tunnel slope deformation can serve as a basis for evaluating slope stability. For this purpose, a fusion optimized prediction model based on wavelet decomposition (WD), particle swarm optimization with genetic algorithm enhancement (IPSO), and gated recurrent unit (GRU) termed WD-IPSO-GRU is proposed. Initially, WD preprocesses noise and features in field displacement monitoring data; subsequently, IPSO dynamically sets learning factors and weights, optimizing the number of neurons and iteration times in GRU hidden layers L1 and L2, and introduces Dropout technique to prevent overfitting, enhancing GRU model performance in long-term sequence prediction tasks. Finally, leveraging the optimal solution enables prediction of GNSS displacement of tunnel slope surfaces. Results indicate that compared to GRU, recurrent neural network (RNN), and long short-term memory (LSTM) models, the WD-IPSO-GRU model demonstrates higher prediction accuracy. The root mean square error (RMSE), mean absolute percentage error (MAPE), and coefficient of determination (R²) for site 02 are 0.16, 0.18%, and 0.95 respectively, providing a new approach for tunnel slope displacement prediction.