The Prediction Model of Seismic Variation in Complex Terrain based on the BP Neural Network with Cavities

Abstract

Surface irregularities and subsurface cavities have a significant impact on seismic wave propagation, leading to either amplification or reduction of ground motion. This study focuses on creating a ground motion prediction model using artificial neural network techniques driven by a synthetically generated database. In this study, we focus on the Erlang Mountain region in Sichuan Province, China, to simulate surface ground motion using the spectral element method, considering the presence of underground cavities in the research area. The classical back propagation neural network model is used to predict changes in ground motion. The model is designed to forecast the PGA influence coefficient, and 5% damped PSV amplification ratio (for periods ranging from 0.33 to 10 s). Input parameters include the buried depth of the cavity, the distance between the surface and the cavity in the mountain projection, elevation, the first gradient of the elevation, and the second-order gradient in two orthogonal directions. The model’s performance falls within acceptable error limits. Additionally, the significance of input features is analyzed, and the model’s applicability in other regions of the ErLang Mountain is validated.