In this work, we use Computational Fluid Dynamics (CFD) to generate the distributed dynamic responses of temperature and moisture in a restaurant that correspond to perturbations of input variables. The CFD model is validated with experimental data of the roof top units return temperatures. A Reduced Order Model (ROM) is developed by approximating the responses to these perturbations using a linear time invariant model. The resulting indoor-air model is coupled to a dynamic envelope model with longer time scales. Assuming the outside temperature distribution and the occupants' loads are known throughout the day, optimal control is applied to our coupled model to minimize the cooling power subject to local and global comfort constraints. The results show that the applied method is more efficient and comfortable than the results given by the experimentally measured response of a conventional set-point control strategy. Published by Elsevier Ltd.