Abstract: Inversion parameters are used to calculate and predict long-term deformation of high concrete face rockfill dam (CRFD). Since the construction process and deformation mechanism of rockfill are complex, it is difficult to separate the instantaneous deformation and rheological deformation, and hence it is necessary to carry out the integrated inversion of static constitutive and rheological model parameters. Based on the measured displacement data, static constitutive and rheological model parameters sensitive to the deformation of rockfill dam are selected for inversion. Additionally, a modified particle swarm optimization algorithm (MPSO) based on particle migration and radial basis function neural network (RBF-ANN) are adopted. What is more, by using the trained neural network to describe the relationship between model parameters and displacements, shorts computing time of parametric inversion. The inversion result of Shuibuya CRFD shows that the calculated settlements agree well with the measured data; and dam deformation is within a reasonable range and tending towards stability.

Key words: rockfill dam, parameter inversion, deformation prediction, MPSO, RBF-ANN, double-yield-surface model