Model Predictive Control for Naturally Ventilated Buildings
Natural ventilation systems can reduce active cooling requirements for buildings while also improving occupant thermal comfort and indoor air quality. The performance of these systems rely on the design of robust and optimal control strategies to exploit the benefits of natural ventilation. Model predictive control is an optimal control technique in which a building model is used to predict the response of the building to certain control inputs under future operating conditions and select a particular control strategy that optimizes for two objectives: minimizing active energy use and maximizing thermal comfort. This project develops a framework for developing a model predictive control strategy of a naturally-ventilated building using the Yang and Yamazaki Environment and Energy Building (Y2E2) as a case study. We illustrate the potential for building energy efficiency improvements and progress towards real-time optimal control of the building.
We developed a building thermal model for Y2E2 by splitting each floor into four zones: atria (with actuated windows to provide natural ventilation), south offices (with active cooling), interior offices (with active cooling), and north offices. We solve the heat and mass balance in each zone at each timestep in the simulation, including the effects of natural ventilation, heat transfer between the internal air and surfaces (walls, floors, etc), and internal loads.
This thermal model was validated over a six day period. Overall, the model captures the diurnal temperature variation and agrees with the experimental observations. The root mean square error for the four zones are as follows: atrium, 0.07 K; north offices, 0.01 K; south offices, 0.04 K; interior offices, 0.03 K
The model predictive control system uses this physics-based model and weather forecast to predict the building's response to control actions. We simulate the building dynamics over a two day horizon (including an initial 24-hour period to spin-up the simulation), optimize a control strategy over that period, implement the first step of that optimal control input, and then repeat the process for the next point in time. The design variables are temperature set points (constant over 2-hour time intervals) and the times when the atrium windows open and close each day. There are two competing objective functions which we aim to minimize: active cooling energy and thermal discomfort.
We optimize using a genetic algorithm and wind up with a Pareto front of optimal solutions. This front represents different tradeoffs between minimized cooling energy and maximized thermal comfort. A key result is that the current building control (baseline) is suboptimal in this simulation period. For a penalty of only 0.2% in thermal comfort, the cooling requirement could be completely eliminated with one of these optimal control strategies. On the other hand, for only 1% more active cooling energy, a different control input could improve thermal comfort by 10%. These findings illustrate the potential for an optimal and robust control system to improve the performance of natural ventilation systems over simple rule-based controllers.
LoCascio, M., Sousa, J., Lall, S., & Gorle, C. A framework for model predictive control of a large naturally-ventilated building. (in prep)