Multiscale vortex flows are found in many engineering applications. The principal method for studying turbulence is large eddy simulation (LES). However, the over-dissipative behavior of LES poses a special difficulty in the simulation of multiscale vortex flows that occur over strongly-curved surfaces. In order to overcome the over-dissipation behavior, and make the prediction of fluctuating characteristics with multiscale vortex flow over curved surfaces more accurate, the regularized variational model (RVM) was adopted to improve the sub-grid scale model in the wall-adapting local eddy-viscosity model (WALE), leading to the development of the modified wall-adapting local eddy-viscosity model (MWALE). The modification was implemented in C++ language in the OpenFOAM software. Using the classical case of viscous flow around cylinder, the influence of filter operator on predicting performance of massive and multiscale vortex shedding at subcritical Reynolds number ( Re = 4.0 \times 10^4 ) is considered in this paper. Computational results show that the improved model is sensitive to the order (n
) of the filtering operator. Fourth-order filter operator (n
= 2) can accurately predict the distribution of mean vorticity on the cylinder surface, and the instantaneous vorticity in the wake. Also well predicted are the recirculation zone length and the mean-velocity profiles. However, the second-order filter operator (n
= 1) cannot accurately predict the fluctuating velocity profiles in the reversed-flow and the wake region. It is found that the predicted pressure distribution in the strong pressure-gradient region, the fluctuating lift and drag forces and the multiscale vortex structures using the fourth-order filter operator are in close agreement with the experimental data.