Research on Time-Frequency Parameter Prediction Models of Ground Motion
In order to make the parameter prediction equations in engineering applications such as seismic hazard analysis and seismic zoning reflect the time-frequency non-stationarity of ground motion,a new time-frequency parameter prediction model of ground motion is needed.Based on more than 30 earthquake re-cords in the western United States,the complementary ensemble empirical mode decomposition method was used to calculate the time-varying power spectrum,the time-varying power spectrum was used to calculate six time-frequency parameters,and the prediction equation and neural network topology of time-frequency parameters with magnitude,distance and site conditions were established.The nonlinear least squares method was used to fit the coefficient values of the prediction equation,and the neural network prediction model was trained.The results showed that the total energy increased with the increase of magnitude and decreased with the increase of distance.The centroid and standard deviation of the spectrum,the centroid and standard deviation of the time,and the time-frequency correlation coefficients decreased with the in-crease of the magnitude,but varied with the increase of the distance.The established neural network time-frequency parameter prediction model showed strong generalization ability and could better predict the time-frequency parameters of ground motion.
time-frequency parameters of ground motiontime-varying power spectrumcomplementary en-semble empirical mode decompositionartificial neural networksprediction model
NETL
NSTL
万方数据
