Aiming at the aggregation of coherent generator groups, based on an artificial fish swarm algorithm, a method is proposed to optimize parameters of the equivalent generator through the simulation of fish group behavior under the principle that the aggregate functions of generator group should be consistent with the transfer functions of equivalent generators in frequency domain. The proposed method can effectively solve the limitations of traditional aggregation methods in initial value setting, optimization direction constraint, local extremum, etc. The validation results of generator aggregation in New England 39bus system show the optimal parameters derived by the new method can better ensure that the frequency response and system transient characteristics of the generator before and after aggregation are basically the same and improve the accuracy of generator aggregation.