Surrogated-assisted multi-objective evolutionary algorithms (SAEAs) have been increasingly studied in recent years to solve expensive real-world problems, where expensive refers to the significant amount of time required for a single function evaluation. However, when the dimensions of the decision variables in the problem are extremely high, the predictive ability of the surrogate model and the search efficiency of SAEAs face great challenges. Therefore, this paper proposes an adaptive sensitive variable extraction strategy, reducing the original high-dimensional space to a low dimensional space. Moreover, we propose novel infill sampling criteria that include both convergence-based-criterion and diversity-based-criterion, in order to simultaneously improve convergence and maintain diversity. The proposed algorithm is compared with the other five state-of-the-art SAEAs on benchmark problems. The experimental results demonstrate the promising performance of the proposed algorithm in tackling high-dimensional expensive multi-objective optimization problems.