A critical quality attribute of therapeutic monoclonal antibodies (mAbs) is the terminal sugar molecules of the N-linked glycan attached to the fragment crystalizable (Fc) region. There exists naturally-occurring heterogeneity in the N-linked glycan structure of mAbs, and such heterogeneity has a significant influence on the clinical safety and efficacy of mAb drugs. We previously proposed a constraint-based modeling method called glycosylation flux analysis (GFA) to characterize the rates (fluxes) of intracellular glycosylation reactions and applied the method to examine the N-linked glycosylation of immunoglobulin G (IgG) in fed-batch Chinese hamster ovary (CHO) fed-batch cultivations. In this work, we significantly improved the computational efficiency of the GFA, and employed the method to analyze the glycosylation of IgG in continuous perfusion CHO cultivations. Perfusion cell cultures have several advantages over the traditional (fed-)batch operation, including higher productivity per unit volume of reactor and more consistent product quality. The GFA showed that as in the fed-batch cultivation, the dynamical changes of IgG glycan heterogeneity in the perfusion culture are mainly attributed to alterations in the galactosylation flux activity. Furthermore, a regression analysis of the galactosylation flux activity using random forest regression linked the dynamics of galactosylation activity with the cell-specific productivity of IgG and the extracellular ammonia concentration.