Unmanned aerial vehicle (UAV) linear parameter-varying (LPV) models can capture the possible nonlinear dynamics. When pursuing an LPV formulation, one ends up with models of relatively large dimension. Accordingly, finding control syntheses usually requires substantial computation. Therefore, an LPV model reduction theory based on
balanced truncation method is proposed. First, the concepts of well-posedness, stability and balanced realizations in a linear fractional transformation framework are given. Then a generalization of the balanced truncation procedure for the LPV model reduction is presented. Finally, LPV model is built with linear time-invariable (LTI) models of UAV lateral system by polynomial fitting. The proposed method is applied to reduce the lateral system LPV model. Simulation results show that the step response of the reduced-order model meets the accuracy specification of output response.