A smart and less intrusive feedback request algorithm towards human-centered HVAC operation

• Published in: Building and environment Vol. 184; p. 107190
• Main Authors: Lee, Seungjae, Karava, Panagiota, Tzempelikos, Athanasios, Bilionis, Ilias
• Format: Journal Article
• Language: English
• Published: Oxford Elsevier Ltd 15.10.2020; Elsevier BV
• Subjects: Algorithms; Bayesian analysis; Bayesian modeling; Buildings; Data collection; Feedback; Human-in-the-loop; HVAC; Interfaces; Mathematical models; Occupant behaviour; Occupant centric; Office buildings; Performance evaluation; Personalized comfort; Personalized environments; Preferences; Thermal preferences; User interface; Occupant centric; Personalized comfort; Occupant behaviour; User interface; Bayesian modeling; Personalized environments; Thermal preferences; Human-in-the-loop
• ISSN: 0360-1323; 1873-684X
• DOI: 10.1016/j.buildenv.2020.107190

There is an increasing number of recent studies about personalized thermal preferences and controls in office buildings. Data collection from occupants in real buildings is necessary for training and updating models. However, sufficient quantity and quality of data are required for developing reliable models, along with optimal model complexity, efficient updating modes and robust evaluation metrics. Therefore, long-term collection of occupant feedback is often needed, which might be intrusive and impractical. This paper presents a Bayesian modeling approach which incorporates voluntary feedback data (comfort-related responses), collected via participatory interfaces, along with requested feedback data, into the personal thermal preference learning framework. This is achieved by explicitly considering occupants’ participation –a type of behavior –in the model structure, i.e., integration of thermal preference-related feedback and occupant behavior. The approach was evaluated with two different datasets collected from two experimental setups with human test-subjects. A smart feedback request algorithm was developed, which determines whether to request feedback at any given time based on the quantified value (i.e., information gain) of the request. The value was computed using the expected Kullback-Leibler divergence between the current and updated posterior parameter distributions. In addition, a simulation study was conducted to evaluate the performance of the algorithm. The results show that the new algorithm learns individual thermal preferences with reduced feedback requests, i.e., effective but less-intrusive. Requesting occupant feedback only when truly needed is important for smart and practical human-centered HVAC operation. •Integrated occupant comfort-related responses into personal thermal preference learning.•Conducted experiments to compare occupants' voluntary and requested thermal preference responses.•Incorporated occupants' participation –a type of behavior –in the model structure.•Developed a smart feedback request algorithm based on information gain.