In this paper, we address the interrelated challenges of predicting user comfort and using this to reduce energy consumption in smart heating, ventilation and air conditioning (HVAC) systems. At present, such systems use simple models of user comfort when deciding on a set point temperature. Being built using broad population statistics, these models generally fail to represent individual users' preferences, resulting in poor estimates of the users' preferred temperatures. To address this issue, we propose the Bayesian Comfort Model (BCM). This personalised thermal comfort model uses a Bayesian network to learn from a user's feedback, allowing it to adapt to the users' individual preferences over time. We further propose an alternative to the ASHRAE 7-point scale used to assess user comfort. Using this model, we create an optimal HVAC control algorithm that minimizes energy consumption while preserving user comfort. Through an empirical evaluation based on the ASHRAE RP-884 data set and data collected in a separate deployment by us, we show that our model is consistently 13.2 to 25.8% more accurate than current models and how using our alternative comfort scale can increase our model's accuracy. Through simulations we show that using this model, our HVAC control algorithm can reduce energy consumption by 7.3% to 13.5% while decreasing user discomfort by 24.8% simultaneously.

Smart thermostats offer impressive scope for adapting to users' thermal comfort preferences and saving energy in shared work environments. Yet human interactions with smart thermostats thus far amount to an assumption from designers that users are willing and able to provide unbiased data at regular intervals; which may be unrealistic. In this paper we highlight the variety of social factors which complicate users' relationships with smart thermostats in shared work environments. These include social dynamics, expectations, and contextually specific factors that influence motivations for interaction with the system. In response we outline our framework towards a Smarter Thermostat: one which better accounts for these messy social inevitabilities, is equipped for a decline in user feedback over time and one which augments rather than attempts to replaces human intelligence-thereby ensuring a smarter thermostat does not create dumber humans.

Space-heating accounts for more than 40% of residential energy consumption in some countries (e.g., the UK and the US) and thus is a key area to address for energy efficiency improvement. To do so, intelligent domestic heating systems (IDHS) equipped with sensors and technologies that minimise user-input, have been proposed for optimal heating control in homes. However, a key challenge for IDHS is to obtain sufficient knowledge of the thermal dynamics of the home to build a thermal model that can reliably predict the spatial and temporal effects of its actions (e.g., turning the heating on or off or use of multiple heaters). This challenge of learning a thermal model has been studied extensively for decades in large purpose-built buildings (such as offices, educational, commercial or communual residential buildings) where machine learning is used to infer suitable thermal models. However, we believe that the technological gap between homes and buildings is fast vanishing with the advent of home automation and cloud computing, and the techniques and lessons learned in purpose-built buildings are increasingly applicable to homes too; with necessary modifications to tackle the challenges unique to homes (e.g., impact of household activities, diverse heating systems, more lenient occupancy schedule). Following this philosophy, we present a methodical study where stochastic grey-box modelling is used to develop thermal models and an extended Kalman filter (EKF) is used for parameter estimation. To demonstrate the applicability in homes, we present the case-study of a room in a family house equipped with underfloor heating and custom-built .NET Gadgeteer hardware. We built grey-box models and use the EKF to infer the thermal model of the room. In doing so, we use our in-house collected data to show that, in this instance, our thermal model predicts the indoor air temperature where the 95th percentile of the absolute prediction error is 0.95 • C and 1.37 • C for 2 and 4 hours predictions, respectively; in contrast to the corresponding 2.09 • C and 3.11 • C errors of the existing (historical-average based) thermal model.

This paper details our work towards designing a system for crowd-sourcing responses on thermal comfort in naturally ventilated office buildings. We provide preliminary qualitative findings on the deployment of this system. Specifically, we explore the different human factors that led to our system being used as something akin to a digital complaints box and how we intend to adapt and leverage the system as a thermal comfort alerts system to better inform building managers.