Search

This paper presents the development of a self-tuned HVAC controller that provides customized thermal conditions to satisfy occupant preferences (i.e., online learning) while minimizing energy consumption, and the implementation of this controller in a real occupied office space. The evolution of personalized thermal preference models and the delivery of thermal conditions with model predictive control (MPC) form a closed-loop. To integrate these two parts, we propose a new method that always provides a set of lower and upper indoor temperature bounds. Different from ad hoc rules proposed in previous research, the control bounds are based on a decision-making method that minimizes the expected cost. We implemented the self-tuned controller in an actual open-plan office space conditioned with a radiant floor cooling system with eight independently controlled loops. Localized operative temperature bounds in each radiant floor loop were determined based on occupants’ feedback and personalized thermal preference models, developed using a Bayesian clustering and online classification algorithm. The self-tuned controller can decrease occupant dissatisfaction compared to a baseline MPC controller, tuned based on general comfort bounds. To generalize the findings of this work: (i) we integrated the self-tuned controller with local MPC into a building simulation platform using synthetic occupant profiles, and (ii) demonstrated a method for automatic system adjustment based on comfort-energy trade-off tuning. In this way, decisions resulting in energy waste or occupant dissatisfaction are eliminated, i.e., the energy is deployed where it is actually needed.

This paper presents a method to incorporate personalized visual preferences in real-time optimal daylighting control without using general discomfort-based assumptions. A personalized shading control framework is developed to maximize occupant satisfaction while minimizing lighting energy use in daylit offices with automated shading systems. Personalized visual satisfaction utility functions were used along with model-predicted lighting energy use to form an optimization framework using two approaches. In the multi-objective optimization scheme, the satisfaction utility and predicted lighting energy consumption are used as parallel objectives to provide a set of Pareto solutions at each time step. In the single-objective optimization scheme, the satisfaction utility is converted into a constraint when minimizing lighting energy use. A simulation study with two distinct visual satisfaction models, inferred from experimental data, was conducted to evaluate the implementation feasibility and optimization effectiveness. Daily and annual simulation results are presented to demonstrate the different patterns of optimal points depending on preference profiles, occupant sensitivity to utility function, and exterior conditions. Finally, we present a new way to apply the multi-objective optimization without assigning arbitrary weights to objectives: allowing occupants to be the final decision makers in real-time balancing between their personalized visual satisfaction and energy use considerations, within dynamic hidden optimal bounds. A slider is introduced as a dynamic user interface with mapped and sorted optimal solutions.

The objective of this paper is to understand the complex interactions related to visual environment control in private offices of perimeter building zones and to develop a new method for learning occupant visual preferences. In the first step of our methodology, we conduct field observations of occupants’ perception and satisfaction with the visual environment when exposed to variable daylight and electric light conditions, and we collect data from room sensors, shading and light dimming actuators. Consequently, we formulate a Bayesian classification and inference model, using the Dirichlet Process (DP) prior and multinomial logistic regression, to develop probability distributions of occupants’ preference, such as prefer darker, prefer brighter, or satisfied with current conditions. Based on field observations, we encode within the model structure that occupants’ visual preferences are influenced by a combination of measured physical and control state variables describing the luminous environment, as well as latent human characteristics. The latter represent hidden random variables used to determine the optimal number of possible clusters of individuals with similar visual preference characteristics in the studied office building population. In the final step, we learn the visual preferences of new occupants in the dataset, by inferring their cluster values, and we derive the personalized profiles, using a mixture of the general probabilistic sub-models.

In this paper, we present a hierarchical Bayesian approach to model human interactions with motorized roller shades and dimmable electric lights. At the top level of hierarchy, Bayesian multivariate binary-choice logit models predict the probability of shade raising/lowering actions as well as the actions to increase the level of electric light. At the bottom level, Bayesian regression models with built-in physical constraints estimate the magnitude of actions, and hence the corresponding operating states of shading and electric lighting systems. The models are based on a dataset from a field study conducted in private offices designed to facilitate a large number of participants and to collect data on environmental parameters as well as individual characteristics and human attributes governing human-shading and – electric lighting interactions. In this study, models were developed only for arrival periods due to the low frequency of actions during intermediate time intervals with continuous occupation. Our modeling framework demonstrates the advantages of the Bayesian approach that captures the epistemic uncertainty in the model parameters, which is important when dealing with small-sized datasets, a ubiquitous issue in human data collection in actual buildings; it also enables the incorporation of prior beliefs about the systems; and offers a systematic way to select amongst different models using the Bayes factor and the evidence for each model. Our findings reveal that besides environmental variables, human attributes are significant predictors of human interactions, and improve the predictive performance when incorporated as features in shading action models.