Stochastic Assessment for Model Predictive Control of a Variable Refrigerant Flow System

It has been widely acknowledged that technical building performance can be influenced by many uncertain factors such as weather, scenarios, occupant behavior, simulation parameters, and numerical methods. For objective and reproducible performance assessment, the aforementioned uncertainties must be reflected in the performance simulation analysis. With this in mind, the authors present a stochastic assessment of model predictive control (MPC) performance of a variable refrigerant flow (VRF) cooling system for an office space. The office space was modeled using EnergyPlus and surrogated models were employed for MPC studies. It is found that the energy savings by MPC can be highly stochastic, ranging from 0.3% to 20.4% depending on weather data. In addition, it is noteworthy that MPC intelligently takes different control strategies (high COP vs. drifting) under different weather conditions.