Artificial neural network-based ground motion prediction equations for Sichuan-Yunnan area of China

Abstract This study presents a novel ground motion prediction equation (GMPE) for the Sichuan-Yunnan region of China based on artificial neural networks (ANNs). Utilizing data from 207 earthquake events and 3537 ground motion recordings collected since 2007, the ANN-based GMPEs predict peak ground acceleration (PGA) and pseudo-spectral acceleration (PSA) for periods ranging from 0.04 to 6.0 s. The model incorporates five key parameters: magnitude, epicentral distance, site conditions (represented by the predominant period $${T}_{0}$$ <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML"> <mml:msub> <mml:mi>T</mml:mi> <mml:mn>0</mml:mn> </mml:msub> </mml:math> derived from horizontal-to-vertical spectral ratios HVSR), hypocentral depth, and styles of faulting. A weighted loss function is employed during ANN training to address data imbalances, particularly the scarcity of near-field recordings. Residual analyses indicate that both inter-event and intra-event variabilities fall within acceptable limits. Comparisons with observed ground motions and existing GMPE commonly used in China confirm that the proposed ANN-based GMPE effectively captures key ground motion characteristics.