As many studies prefer to focus on the accuracy of prediction models in model-based problems of chiller plants, the practicality and feasibility of the models used for field applications are often compromised for a superior performance. We herein present an online predictive control method based on linear regression model for a typical chiller plant from the perspective of real-time application to optimize its overall performance and energy consumption. A multi-input–single-output (MISO) controller is developed for predicting the global coefficient of performance (COP) of the chiller plant to reflect integral characteristics and interactions of components based on the multivariable linear regression (MLR) model. The input variables are much easier to obtain in a practical system compared to variables such as water flow and part load ratio, thereby rendering an excellent model capable of easy implementation and duplication. The performance of this approach is tested and evaluated in a real chiller plant system by comparing it to that obtained using a local control strategy. The results of this study indicate that the online predictive control strategy can enhance the global COP values by 5.35％ on average and reduce electricity consumption by 2.70％ daily compared to the local control strategy.