Uncertainty inverse analysis of dam material parameters based on relevance vector machine and stochastic finite element method

It is difficult to control the dam settlement of face rockfill dam because of uncertain factors in the design, construction and operation periods, so an uncertainty inverse analysis model of the material parameters was proposed based on relevance vector machine (RVM) and stochastic finite element method. First, stochastic field discrete method based on the Cholesky decomposition was used in Monte Carlo stochastic finite element method. Hence, the variability of the material parameters can be considered in numerical calculation, and the examples were generated by the stochastic finite element method to train the RVM. Then, the complex nonlinear relationship between the material parameters and the dam settlement was simulated by the RVM which is an advanced machine learning algorithm. The RVM can establish the uncertainty connection between the input and output of the model, so the stochastic finite element method with time-consuming was replaced by the RVM. Finally, with the measured data of dam settlement, the variation coefficient of the material parameters was optimized by genetic algorithm. The application example of Gongboxia rockfill dam shows that the variation coefficient of the material parameters can be quickly and accurately determined by the uncertainty inverse analysis model. This model considers the uncertainty of the the numerical calculation and input-output, so it has good application prospect and promotion value.