Abstract. Aiming at the problem of large amount of calculation and solving inefficiency when using finite element software to analyze the "whole model" of complex space steel structure, the mapping model method built on the basis of SVM (Support Vector Machine) theory is proposed to reduce computation and enhance the computing efficiency. In this paper, taking tower parking system structure as the object, the mapping relationship is established among parking layers, conditions and the von Mises stress, maximum displacement which the system is subjected to. According to the mapping relationship, it can realize the prediction of von Mises stress and maximum displacement in different layers and conditions to avoid calculating "whole model". The research shows that the maximum relative error of stress is 1.69% and the maximum relative error of displacement is 3.16% when compared results that predicted by SVM with results that calculated by finite element. At the same time, according to the comparison of mapping results which predicted by SVM and neural network, it comes the conclusion that results of SVM have a goodness of fit.