Off-campus UNL users: To download campus access dissertations, please use the following link to log into our proxy server with your NU ID and password. When you are done browsing please remember to return to this page and log out.
Non-UNL users: Please talk to your librarian about requesting this dissertation through interlibrary loan.
Analytical and experimental comparison of model-based, model -free, and hybrid learning control of active and passive building thermal storage inventory
Unlike active thermal energy storage systems in central chilled water plants, most commercial buildings embody a substantial passive thermal capacitance in their structural mass and interior furnishings, which to date is not systematically utilized for cost savings, energy efficiency, load management, or demand response. This dissertation describes research efforts on optimal control for the simultaneous use of active and passive building thermal storage inventory. Model-based predictive optimal control, model-free learning control, and a hybrid learning but model-based control scheme were investigated successively. Numerical analysis and experimental study demonstrated that each of these approaches has advantages and disadvantages in terms of feasibility, robustness, and control performance. While model-based predictive optimal control offers the best control quality from the perspective of cost savings compared with conventional control strategies, it requires an accurate building model and accurate weather prediction, which is hard to achieve when the building is complex. Model-free learning control avoids the need for a physical model and forecasting by using a statistical summary of past operational experience. However, the training period is too long to be practical for commercial building applications, and the performance of the pure learning controller is sensitive to the selection of learning parameters. Finally, a hybrid learning control approach was investigated, which is based on simulated reinforcement learning and combines the merits of model-based and pure learning control. An experimental study confirmed the feasibility of the proposed hybrid learning control approach. Although operating cost savings were achieved compared with conventional control methodologies, they were lower than those for model-based predictive optimal control. The hybrid learning controller was significantly affected by the quality of the training model used to implant prior domain knowledge, and extensive real-time learning was required for the learning controller to eliminate false cues it received during the initial training period. Nevertheless, compared with pure reinforcement learning, the hybrid controller can be much more readily applied to a commercial building for implementation of supervisory adaptive optimal control. ^
Engineering, Civil|Engineering, Mechanical
Liu, Simeng, "Analytical and experimental comparison of model-based, model -free, and hybrid learning control of active and passive building thermal storage inventory" (2005). ETD collection for University of Nebraska - Lincoln. AAI3176793.