The identification results of excitation system parameters are often unstable for the reason that partial parameters are correlated into implicit functions when the measurements are insufficient

thus a method is proposed to identify these associated parameters. Firstly

the singular value decomposition approach for subset selection is adopted to divide the excitation system parameters into the associated set and the non-associated set. Secondly

in order to reduce the errors of remaining parameter estimate caused by the deviations between typical values and true values

when the correlation is removed by means of assigning typical value to representative parameters

the parameters with low sensitivity are assigned values. In order to avoid introducing the associated parameter during the selection process

the relevance between all parameters should be clarified further. In this respect

the discriminant coefficients are calculated by the trajectory sensitivity matrix and parameters could be separated into several implicit function parameter sets accordingly. The experimental results reveal that the proposed method can not only effectively judge the identifiability of excitation system parameters but also reduce the bias of identification results because of inaccuracy of typical values.