Solving the optimal power flow (OPF) problem on a large power system is computationally expensive. Network reduction and ac-to-dc network conversion can relieve this burden by simplifying the full system model to a smaller and mathematically simpler model. Traditional reduction methods, like Ward reduction, fractionalize generators when the buses they are attached to are removed, and scatters these fractions to topologically adjacent buses. In some OPF applications, this type of generator modeling is problematic. An improved approach is to keep generators intact by moving them whole to buses in reduced model and then redistributing loads to maintain base-case line flows. Determining generator placement using a traditional shortest electrical distance (SED) based method may result in cases where the OPF solution on reduced model is infeasible while the full model has a feasible solution. In this paper, an improved generator placement method is proposed. Tests show that the proposed method yields a better approximation to the full model OPF solutions and is more likely to produce a reduced model with a feasible solution if the unreduced model has a feasible solution.