Abstract: Ground vibrations caused by overloaded vehicles are influenced by factors such as vehicle weight, speed, road surface roughness, and soil parameters, making their attenuation patterns complex and difficult to accurately predict. This study characterizes and predicts the relationships between vehicle weight, speed, road surface roughness, soil parameters, and vibration attenuation performance. A prediction model for vibration response and attenuation at different sites under traffic loads is developed using a sparrow search algorithm (SSA)-optimized back propagation (BP) neural network. Additionally, a game-theory-based Shapley additive explanation (SHAP) algorithm is used for the model inter-pretability analysis. The results show that the model accurately simulates ground vibration propagation due to overloaded vehicles with high consistency in time- and frequency-domain characteristics when compared to field test data. Although the BP model exhibits prediction errors exceeding 10% for high-dispersity data, the SSA-BP model maintains high prediction accuracy across various datasets. In silty soil environments, the vehicle speed exhibits the strongest correlation with the vibration acceleration in the surrounding environment, whereas in clay and gravel environments, the distance from the vibration source exhibits the strongest correlation. The SHAP value analysis indicates that with the decrease in shear wave velocity, the effect of driving speed initially decreases and then increases, whereas the effects of vehicle weight and road surface grade gradually increase.

Key words: traffic load, vibration attenuation, sparrow search algorithm-back propagation (SSA-BP) prediction model, interpretable machine learning