Abstract: Rheological parameters of rockfill are important for long-term safety analysis of high concrete face rockfill dams (CFRDs). Parameter inversion can accurately obtain rheological parameters to meet the practical long-term deformation law. This paper uses Backpropagation (BP) neural network and Radial basis function (RBF) neural network to construct the response between the parameters to be inverted and the measured displacement, and introduces the root mean square error (RMSE), the average absolute percentage error (MAPE), and the linear regression determination coefficient (R2) in the statistical regression prediction model to comprehensively compare the mapping capabilities of different neural network response surface. They can improve the efficiency and accuracy of parameter inversion. Results show that the evaluation indexes of RBF neural network response surface are better than those of BP neural network response surface. Therefore, we adopt RBF neural network response surface and multi-population genetic algorithm (MPGA) to obtain the rheological parameters after inversion and use them for finite element calculation. It is found that the obtained settlement values of the Xujixia concrete face rockfill dam agree well with the measured ones both in magnitude and in distribution.

Key words: parameter inversion, rockfill rheology, neural network response surface, multi-population genetic algorithm