Risk assessment of mountain tunnel entrance collapse based on pso-LSTM surface settlement prediction

Purpose – Predicting surface settlement at mountain tunnel entrances during construction is increasingly crucial for risk analysis, as the accuracy of these predictions directly impacts collapse risk assessments and personnel safety. Design/methodology/approach – This study introduces a novel approach using a particle swarm optimization (PSO)-optimized long short-term memory (LSTM) neural network for surface settlement prediction. The PSO algorithm optimizes key hyperparameters of the LSTM model, including the number of hidden layer neurons, the learning rate and L2 regularization, while the Adam optimizer refines LSTM iterations. Dropout is used in combination with adaptive L2 regularization parameters to avoid overfitting situations, and sensitivity analysis of the remaining variables ensures the identification of the optimal solution. Findings – The model, based on monitoring data from the Aketepu No. 1 Tunnel's left tunnel, establishes evaluation criteria incorporating error margins and root mean square error (RMSE). By examining the range of maximum (minimum) settlement rates for the cumulative settlement values, the study determined that the section is exposed to an average risk of collapse with slow deformation, which is consistent with actual observations. Originality/value – This suggests that construction can proceed normally, with appropriate monitoring to mitigate the risk of collapse. The PSO-LSTM forecast model presents a promising approach for predicting collapse risks at mountain tunnel entrances.