Ground motion inversion method based on generalized chaotic particle swarm optimization

Ground motion inversion is a core technology for revealing earthquake mechanisms and obtaining accurate source and site parameters. However, it has long been constrained by strong nonlinear coupling of parameters, reference station dependence, and the tendency of optimization algorithms to get trapped in local optima. To address these issues, this study constructs a robust ground motion inversion model based on Generalized Chaotic Particle Swarm Optimization and Generalized Inversion Technique (GCPSO-GIT). First, strong ground motion data are preprocessed by screening, baseline correction, and Konno-Ohmachi smoothing. Then, a two-step inversion strategy is employed: the path attenuation term is initially separated using linear inversion, followed by the introduction of a Chaotic Mechanism (CM) to enhance the global search capability of particle swarm optimization. Crucially, the model achieves parameter decoupling by minimizing the site effect Coefficient of Variation (CV) to enforce statistical stability across multiple seismic events. Results show that the peak value of the site effect variation coefficient inverted by this model is only 12%, significantly lower than the 35% of the particle swarm optimization-generalized inversion technology. The median source stress drop is approximately 42 bar, with the smallest dispersion compared to other models. In the seismic ground motion simulation, the peak ground acceleration of the synthesized acceleration time history is 2.7 m/s², with a very small deviation from the observed value of 2.8 m/s². At the application level, the correction error of the seismic design response spectrum for Class C sites is as low as 5.3%, and the deviation rate of ground motion parameters for major projects is only 6.2%−8.1%. Furthermore, computational cost evaluations confirm the high efficiency of the framework, while independent cross-verification using the KiK-net database demonstrates its robust generalizability under diverse regional propagation characteristics. This model successfully achieves accurate parameter decoupling without relying on an ideal reference station, providing reliable parameters for regional refined seismic fortification and probabilistic risk assessment.