For Model-Predictive Control (MPC) to be implemented in real application, data driven inverse models are advantageous since they are easily constructed as well as relatively fast and accurate, compared to first principle based models (simplified calculation [ISO 13790], dynamic simulation [EnergyPlus, ESP-r, TRNSYS, etc.], state space models, etc.). Gaussian Process Model (GPM), one of the inverse methods, can be beneficially used for real time stochastic optimal control of nonlinear building systems, since the GPM consumes much less computational time and does not require significant efforts. The GPM is a black-box model based on Bayesian approach based on measured in-output dataset. For real-time optimal control of chiller operation, this paper presents a coupling between the GPM and an optimization routine in MATLAB optimization toolbox. The two control parameters studied in the paper are the outlet temperatures of chilled water loop and cooling tower loop. In particular, Genetic Algorithm (GA), one of the meta-heuristic methods, was applied to find optimal control strategy. It is elaborated in the paper that GPM produces reliable control results reflecting probabilistic natures of the chiller system.