In order to solve the problem of large parameter identification error caused by nonlinear links of excitation system being triggered easily when transient stability is under fault state, an improved differential evolution algorithm for system parameter identification is proposed by using the characteristic of artificial intelligence algorithm that the nonlinear link is approximated infinitely through optimization. The improvement of the algorithm solves the problems of slow convergence speed, poor fine optimization ability and easily to produce local optimum when classical artificial intelligence algorithm identifies the parameters of non-linear links. At the same time, in order to solve the problem of inaccurate parameters in the whole identification, a decomposition link identification strategy is proposed. The example analysis shows that the algorithm improves the convergence speed, avoids local optimum and improves the convergence accuracy. According to the proposed parameter identification strategy, the excitation system is decomposed and identified, which improves the accuracy of generator excitation system parameter identification, and provides an accurate model and method for power system stability analysis