Abstract: A partial least squares regression principle is employed to construct a compound spectral acceleration that accounts for the effects of multiple vibration periods. Taking the Xi’an region as research subject, a ground motion prediction model for the compound spectral acceleration is developed. The probabilistic seismic hazard analysis for the compound spectral acceleration is performed using Monte Carlo simulation. Based on the deaggregation of probabilistic seismic hazard, scenario earthquakes for fundamental natural vibration periods of 0.2 s, 1.0 s, and 3.0 s are determined using the weighted method. A target response spectrum based on the compound spectral acceleration is proposed, and the ground motion records are selected using a greedy optimization algorithm. The results show that the hazard curve of the compound spectral acceleration represents the summation of hazard curves from three typical periods with different weighting ratios, providing a comprehensive assessment of probabilistic seismic hazard, compared to single spectral acceleration. The median values and logarithmic standard deviations of spectral accelerations from the selection of ground motion records through the conditional spectrum and greedy optimization algorithm show a good agreement with the conditional mean spectrum, and the selected ground motion response spectra all fall within the 2.5% and 97.5% percentiles of the conditional mean spectra.