optimal control, variable air volume, model-based, uncertainty analysis, sensitivity analysis
In the U.S. A Variable Air Volume (VAV) system is one of most commonly used air system for multiple-zone commercial buildings due to its capability to meet the varying heating and cooling loads of different building thermal zones. One of key component of VAV system is the terminal VAV box. There are an air damper and a reheat coil in the box. How to effectively and efficiently control the VAV box plays a significant role to reduce energy consumption and maintain acceptable indoor environment in buildings. Currently, there are two control logics used for controlling VAV box, namely, single maximum and dual maximum control logics. The single maximum logic is the most common, where the room temperature setpoint is maintained by only adjusting the reheat coil valve position in the heating model. The damper position is kept as the minimal to satisfy the ventilation requirement only. On the other hand, the more advanced dual maximum control logic realizes the room air temperature control by adjusting both damper position and reheat coil valve position in the heating model. For the cooling model, both control logics have the same action to maintain room air temperature setpoint through adjusting the damper position. Â In this study, a model-based optimal control is explored to minimize the energy consumption of the VAV box with a hot water reheat coil. Data driven approach based on an Autoregressive exogenous (ARX) model is investigated to represent dynamics of the room thermal response. The similar data-driven approach is used to develop an energy consumption model of the VAV box. Measured data for the VAV box from a real building is used to train and test data-driven model. Such data includes room air temperature, outdoor air temperature, supply air temperature, supply air flow rate, damper position, reheat coil valve position and VAV box energy consumption. A platform of AMPL (A Modeling Language for Mathematical Programming) is used to for mathematical modeling and links to different optimization solvers. Â In addition, uncertainty analysis and sensitivity analysis are conducted to help understand the model behaviors and performance. In this study, the Monte Carlo sampling method is applied to generate samples for model inputs including supply air temperature, outdoor conditions, etc. A quantified sensitivity index of Sobol is calculated to indicate the impact level from different inputs or disturbances.