Abstract
Nowadays people spend 90% of the time in indoor. To provide a comfortable and healthy environment for occupants, buildings consume 40% of the total energy in the world. Due to the inappropriate design of the indoor environment, the problems related to bad indoor air quality caused over $20 billion loss in the US. Then it raises a question on how to improve the indoor environment and decrease the energy consumption in the buildings. One of the strategies available is to utilize the stratified airflow distributions such as mixed mode ventilation. Previously, the coupled simulation between building energy simulation program and computational fluid dynamics (CFD) models was used to study energy and comfort performance of those systems while putting the control dynamics aside. This research develops the coupled simulation model to study the dynamic systems of non-uniform airflows, HVAC, control, and building envelopes. Fast fluid dynamics (FFD) is chosen to simulate non-uniform airflows since FFD is computationally fast than FFD in simulating the non-uniform airflows. Modelica language, which is equation-based and object-oriented, is used to model HVAC, control, and building envelopes. Then, the coupled simulation model is further ameliorated by adding the multizone models to expand the application scope of the model from a single zone to a building with multi zones. To further improve the model for design optimization study, this research improves the computation speed of FFD by parallelizing it using open computing language (OpenCL). We systematically evaluated on the feasibility of using OpenCL to accelerate the airflow simulation using FFD as an example. Though FFD programmed in OpenCL running on different graphics processing units (GPU) may generate different results due to different interpretation of IEEE-754 standards, the difference is minor to some extents that are negligible in airflow simulation. Running FFD in parallel on a, up to 1139 times speedup is achieved, which is promising to dramatically reduce the time cost for design optimization of the dynamic systems. Regarding the operation optimization, it would be preferable to increase the computation speed of non-uniform airflow simulations by using reduced order models (ROM). We proposed to use in situ adaptive tabulation (ISAT), which differentiates from other conventional ROMS in that it can call a full-scale FFD simulation when the prediction is deemed not accurate. This is critical in the optimization. ISAT executes a FFD simulation if interpolation is deemed inaccurate. In this study, ISAT is trained by using FFD running in parallel on a GPU and once well trained ISAT can answer query points both inside and close to training domain using retrieve actions within a time less than 0.001s for each query. This shows that ISAT can be used to further improve the coupled simulation model to realize operation optimization, such as model predictive control using a non-gradient based optimization.