An Approach to Data-Driven Sensing and Predictive Supervisory Control for Commercial Buildings with In-Situ Evaluation

Abstract

Modern building automation systems offer virtually limitless programmability, sensing, and connectivity, but often do not achieve the energy use or occupant comfort goals of building stakeholders. One major source of the shortfall of modern building automation systems is room-level and zone-level control. Given the lack of perceived importance of zone-level controllers compared to system-level and building-level controllers, comprehensive and efficient zone-level programming is overlooked at the cost of quality of implementation. This thesis addresses these obstacles to improve comfort and energy performance by developing a suite of interrelated approaches and technologies and extensively field testing them in 27 highly instrumented (but typical) offices. One approach to reducing sensor infrastructure (and associated installation and maintenance costs) for advanced control applications is to apply proxy sensing using data-driven modelling. By comparing indirect sensors for energy use estimation (i.e., proxy sensors) to direct energy use estimation sensors, this thesis develops an approach to rank all physically reasonable sensors at the zone level for energy use estimation usefulness. From this approach, it was found that outdoor air temperature, hydronic flow, and hydronic temperature sensors were the most useful for proxy sensing. With the ability to accurately estimate room-level energy use, the performance of novel supervisory control approaches can be quantified. To improve zone-level control, this thesis first developed a simplified model-based predictive control approach for the heating season. The simplified MPC achieved 94% of the energy savings of MPC, but the implementation in a typical building automation system (BAS) controller was much simpler. The simplified MPC approach was modified to remove the underlying indoor air temperature model and scaled to operate in both the heating and cooling seasons in 27 rooms of an institutional building at Carleton University. The MPC and RE control reduced energy use by 42% and 27% in the cooling season and 18% and 33% in the heating season when compared over four weeks of testing for each algorithm and season. Thus, the RE approach was experimentally verified to have comparable energy savings to detailed MPC while using only a few dozen lines of BAS controller-embedded code.