Research On Simulating Ground-motion Fields Based On Strong Ground Motion

Based on the real ground motion recordings,this paper uses the parametric prediction model and the non-parametric prediction model to regress the ground motion model.The principal component analysis(PCA)and particle swarm optimization(PSO)algorithm in machine learning are used to extract mother wave of the ground motion.Through the linear combination of the motion mother wave,the prediction response spectrum of the target site is matched and the ground motion time history of the target site is achieved.The main work of this paper is as followed:(1)Based on the actual ground motion data in Sichuan-Yunnan area,the relationship between peak ground acceleration(PGA)(spectral acceleration(Sa))and magnitude,site and distance is established by using random effect model.A ground motion model is given,which conforms to the local ground motion characteristics.The variance of regression is close to 0.7,and the model is in good agreement with the local ground motion data characteristics.Compared with NGA and other models in China,the established model is suitable for Sichuan-Yunnan area.(2)A framework of least squares support vector machine(LSSVM)regression ground motion prediction model(GMM)is proposed,and the method of determining the parameters C and r of LSSVM model is given.Based on the ground motion in SichuanYunnan region,the least squares support vector machine(LSSVM)ground motion model is established.The proposed GMM can reasonably predict the PGA and Sa at a given field point.Compared with the prediction models,the results show that the proposed model is reasonable and has good generalization ability for ground motion prediction.(3)The principal component algorithm in machine learning is introduced to extract the effective ground motion mother wave information from the ground motion database.Combined with the ground motion prediction model of the target area,the ground motion time history conforming to the response spectrum of the specific site is given.The proposed combination of local actual ground motion and prediction model to synthesize new ground motion time history can reasonably contain the characteristics of regional actual ground motion,and can match the spectral characteristics of target ground motion.