Develop and evaluate a transferable model predictive control based group occupant-centric control strategy

Yicheng Li

doi:10.17918/00011094

Occupants' well-being is significantly influenced by how buildings are designed and operated. In recent years, occupant-centric control (OCC) strategies have gained growing attention for their potential to enhance occupant comfort and energy efficiency. Among advanced control approaches, model predictive control (MPC) has been increasingly applied to OCC, as it leverages predictive models and optimization algorithms to reduce HVAC energy consumption while improving thermal comfort. Based on the literature review, there are many well-established methodologies for modeling building and heating, ventilation, and air conditioning (HVAC) systems, but the thermal comfort model to be used in MPC still requires further exploration. Specifically, (1) there is a lack of publicly available and comprehensive virtual testbed that integrates buildings, HVAC systems, and occupant models; (2) there is a lack of thermal comfort models that address individual variability and resolve comfort conflicts among a group of occupants; (3) there is a high cost to collect sufficient occupant data to train reliable models; and (4) evaluation of MPC performance in occupant-centric applications is not widely reported. To this end, the objective of this study is to develop a transferable MPC-based group OCC strategy that enhances occupant thermal comfort and saves energy without requiring large amounts of occupant comfort data. This study aims to bridge these gaps through the following contributions: · Virtual testbed development: A calibrated virtual testbed (representing a single zone with two-stage air source heat pump system in typical commercial building) was developed. Results indicate that the virtual testbed can accurately simulate the building thermal environment, HVAC system energy consumption, and occupant thermal comfort. · Data-driven prediction model development: Data-driven prediction models needed in the proposed MPC framework were developed, specifically group thermal comfort model, HVAC system power model, and zone environment (air temperature and humidity) models. Among them, the primary focus is on the definition and development of group thermal comfort model. Results show that the random forest-based model provides accurate predictions for group thermal comfort, and the RMSE values for other models are less than 0.5 °C, 5%, and 90W, respectively. · Transfer learning for group thermal comfort model: To address data scarcity, an instance-based transfer learning framework was introduced to improve the performance of group thermal comfort models. Evaluation results demonstrate that this method can enhance model predictive performance and is robust across varying occupant groups. · Development of a Transferable MPC-Based Group Occupant-Centric Control (TMPC-GOCC) strategy: By integrating the developed prediction models, a TMPC-GOCC strategy was developed and evaluated using the virtual testbed. The results show that compared to a conventional MPC that uses zone temperature as the constraint, the proposed strategy achieves better control performance in occupant comfort, reducing occupants' discomfort votes by up to 50%. Finally, this thesis presents the overall achievements, identifies key limitations, and proposes future works.

Develop and evaluate a transferable model predictive control based group occupant-centric control strategy

Files and links (1)

Abstract

Metrics

Details

Develop and evaluate a transferable model predictive control based group occupant-centric control strategy

Files and links (1)

Abstract

Metrics

Details

Drexel University Social media