Abstract: A back analysis model considering interaction and radiation damping effects is proposed for dynamic parameters of earth-rock dams. In the model, the viscous-elastic artificial boundary combined with the equivalent node force method of ground motion input is used for achieving the wave motion input of semi-infinite site, and the multiple population genetic algorithm (MPGA) and the radial basis function neural network (RBF) are adopted. Using MPGA to optimize the dynamic parameters of the dam can effectively avoid the prematurity of the traditional genetic algorithm. The trained RBF is used to describe the mapping relationship between the model parameters and the response spectra of acceleration so as to reduce the computing time of parametric inversion. The inversion results of Liyutan dam in China show that the calculated settlements agree well with the measured data. The proposed model is feasible in back-analysis of dynamic parameters of earth-rock dams. The study shows that the dynamic shear modulus coefficient of the dam materials obtained from the rigid foundation model inversion and the indoor dynamic triaxial tests is small. The value of the indoor tests is much smaller and should be revised.