Abstract: The surrogate-assisted evolutionary algorithm can improve the solving efficiency of expensive optimization problems. When dealing with expensive multi-objective optimization problems, the number of objectives and the limitation of computational resources pose new challenges to the modeling of surrogate-assisted evolutionary algorithms and the balance between convergence and diversity. This paper proposes an ensemble secondary prediction surrogate-assisted expensive multi-objective evolutionary algorithm (ESPSEMEA). Firstly, the population is mapped to the transposed space and the radial space, respectively. Next, a two-level prediction model for radial space and transpose space is built by kriging model, which reduces the modeling complexity. In addition, according to the diversity information provided by the radial space and the convergence information provided by the transpose space, the diversity and convergence are taken as the two optimization goals, and the non-dominant sorting method is used to select the offspring, which can effectively balance the convergence and diversity. By comparing with the other six advanced algorithms, it is proved that the proposed algorithm can not only reduce the modeling time complexity but also effectively balance convergence and diversity.