To address the problems of low computational efficiency and difficulties in multi-objective optimization during the inversion of constitutive parameters of the Duncan E-B model for gravel pile composite foundations, an intelligent inversion method integrating a radial basis function (RBF) surrogate model and a multi-objective rime (MORIME) optimization algorithm was proposed. In this method, Latin hypercube sampling was adopted to construct the parameter sample set; an RBF surrogate model was established to replace finite element computations; the MORIME optimization algorithm with a local exploitation strategy was introduced to achieve rapid inversion. A single-pile static load test at a pumped storage power station was taken as an example to verify the parameter inversion of the Duncan E-B model for the gravel pile and the surrounding soil. The results show that the inversion method has high inversion accuracy; the coefficients of determination of curves at different monitoring points obtained by substituting the optimal inverted parameters back into the finite element computations are all above 0.99; the dual-mode local exploitation strategy can significantly optimize key objectives and reduce the number of iterations for convergence. The SHAP sensitivity analysis further shows that the surrounding soil parameters K and K b have a significant influence on the pile-top displacement and the soil stress near the pile top.