Computational principles and applications of data-driven seismic dynamic response analysis

The seismic safety evaluation of high slopes in hydropower projects within strong earthquake regions of Southwest China is of great significance for disaster prevention and mitigation. Addressing the unreliability of traditional analyses relying on empirical constitutive models and subjective parameter calibration, this paper proposes a data-driven method for seismic dynamic response analysis. By computing directly from discrete rock stress-strain data, the method eliminates complex constitutive modeling and parameter calibration. Furthermore, by coupling viscous, free-field, static-dynamic unified, and seismic input boundaries, it accurately simulates wave propagation in semi-infinite media. After validating the framework via a layered site benchmark, a full-process simulation (gravity–excavation–earthquake) was conducted on a typical high slope at the Mengdigou Station. Results demonstrate the method's excellent stability in complex multi-step static-dynamic coupled analyses. Physically, excavation-induced vertical rebound and lateral relaxation are strictly controlled by weak faults, easily triggering uncoordinated shear slips. Under subsequent seismic action, the dynamic response exhibits significant spatial amplification effects dominated by elevation, slope surfaces, and local topography. This method offers a new paradigm for computational rock dynamics, providing robust support for the seismic design and safety evaluation of major hydropower slopes.