2023 Volume 23 Issue 3 Pages 3_59-3_76
To acquire new knowledge through earthquake ground motion evaluation from a new perspective, the authors attempted to create a site-specific earthquake ground motion evaluation model by machine learning using earthquake ground motion records obtained in the past as training data. The epicentral direction and response duration time of the earthquake ground motion, which had not been addressed by conventional attenuation relations, were also examined. Overall, the observed values were evaluated and modeled well, within twice to half the observed values. The average ratio of the evaluated value to the observed value was approximately 1, and the common logarithmic standard deviation was slightly greater than 0.2 for the amplitude of ground motion and slightly greater than 0.1 for the response duration time. The impact of the epicentral direction on the response duration time was large, and in some cases, it was almost equal to or greater than the impact of each parameter in the conventional prediction equations.
