Vertical ground motions acceleration response spectrum is an important part of the structural seismic design, and the domestic and foreign seismic design codes use the H/V response spectrum ratio to calculate the vertical response spectrum indirectly, which is a simple and feasible method, but the current seismic codes are still too rough for the vertical ground motions response spectrum, which will underestimate the role of vertical component in the engineering application or there is redundancy in the application. In addition, the geological structure of the subduction zone is complex and earthquakes occur frequently, and the interface earthquakes have the characteristics of high magnitude and high intensity, so it is necessary to study the interface earthquakes in the subduction zone separately.

In this paper, based on the surface observation data obtained from the KiK-net and K-NET in Japan, the data are first preprocessed to improve the accuracy of the model and to eliminate the seismic records that affect the fitting effect, which involves filtering the magnitude and the source distance of the data, checking the acceleration time history, then correcting the instrumental response and filtering the waveforms. Finally, 3 048 seismic records from 59 subduction zone interface earthquakes are selected for the study, and the acceleration response spectra are calculated for each record with 5% damping ratio. Random-effects algorithms and mathematical statistics were used to analyze the effects of spectral period, earthquake type and site conditions on the H/V response spectral ratio, based on the site period （T_s） division of site classes, the H/V response spectrum ratio model for non-scenario earthquakes （without magnitude and source distance parameters） from the subduction interface is established, the model coefficients are smoothed and adjusted to facilitate engineering applications, and the standard deviation of model residuals is evaluated and analyzed to explore the magnitude and source of random residuals and improve the accuracy and reliability of the model. The model can be used to adjust the response spectrum of engineering seismic design when there is no earthquake source and no path information.

The study shows that: ① T-test method verifies that the type of earthquake and the site classes have a significant effect on the H/V response spectral ratio, so it is necessary to establish the corresponding response spectral ratio models according to different types of earthquakes and site classes; ② The peak period of the site term coefficients is close to the average period for each type of site, indicating that the site term coefficients are reasonably well taken to elicit a resonant site response. In the prediction model, the peak period of the H/V response spectral ratio is 0.2 s for class I sites, 0.27 s for class II sites, and 0.5 s for class III sites. The peak period of the H/V response spectral ratio in the prediction model has a certain correlation with the site predominant period, which can prove that the H/V response spectral ratio model is well fitted by using the random-effects method; ③ The standard deviation analysis shows that the dispersion generated by the source effect is smaller than that generated by the path effect and the site effect, the random residuals of the path effect term and the site effect term contribute more to the model variability, indicating that the random residuals of the horizontal and vertical components of the seismic record have strong correlation. Different site classes produce the largest dispersion in response spectral ratios at their corresponding mean site period. The increasing dispersion of the H/V response spectral ratio due to path effects with increasing spectral period demonstrates that the correlation between the horizontal and vertical component random path residuals decreases with increasing period.

The H/V response spectral ratio model for non-scenario earthquakes established in this study has the characteristics of simplicity and convenience for engineering application, and the source and path parameters are not added into the model. Residual analyses show that the between-event residuals are correlated with the magnitude and fault depth, and the effects of magnitude and fault depth on the between-event residuals are clearly segmented at magnitude 7.1 and fault depth of 25 km. There is a correlation between the within-site residuals and the fault distance. Moreover, the effects of magnitude, source depth and fault distance on the residuals change with the spectral period, the seismic source characteristics and propagation paths still affect the H/V response spectral ratio. Therefore, when there is clear seismic information, the introduction of magnitude, source distance, fault depth and other factors can further improve the model prediction accuracy, and will be committed to constructing a more accurate H/V response spectral ratio model in the future research. The results of this paper are of reference value for the seismic fortification of offshore engineering.