Brain Canada Foundation

Occupant behaviour (OB) centric controls have significant potential in advancing next-generation HVAC systems. Many OB-centric control studies solicit feedback from occupants to tackle the thermal preference learning problem. Behaviour nudging was also implemented in various systems to influence occupant behaviour to be more energy efficient. This study addresses the gap of using behaviour nudging and unsolicited occupant thermostat overrides to learn their thermal preferences. A multi-armed bandit (MAB) reinforcement learning (RL) was used to learn occupant thermal preferences from their thermostat interactions. The reward signal of the algorithm was designed to reward energy savings and penalize discomfort. The occupants were continuously nudged by slowly reducing the zone setpoint during the heating season, to encourage them to override the thermostats. The algorithm was implemented in two zones with multiple occupants in an academic facility in Ottawa, Canada, achieving energy savings of up to 12.7% compared to static setpoints.

Bibliometrics methods have allowed researchers to assess the popularity of brain research through the ever-growing number of brain-related research papers. While many topics of brain research have been covered by previous studies, there is no comprehensive overview of the evolution of brain research and its various specialties and funding practices over a long period of time. This paper aims to (1) determine how brain research has evolved over time in terms of number of papers, (2) countries' relative and absolute positioning in terms of papers and impact, and (3) how those various trends vary by area. Using a list of validated keywords, we extracted brain-related articles and journals indexed in the Web of Science over the 1991-2020 period, for a total of 2,467,708 papers. We used three indicators to perform: number of papers, specialization, and research impact. Our results show that over the past 30 years, the number of brain-related papers has grown at a faster pace than science in general, with China being at the forefront of this growth. Different patterns of specialization among countries and funders were also underlined. Finally, the NIH, the European Commission, the National Natural Science Foundation of China, the UK Medical Research Council, and the German Research Foundation were found to be among the top funders. Despite data-related limitations, our findings provide a large-scope snapshot of the evolution of brain research and its funding, which may be used as a baseline for future studies on these topics.

Thermal models of buildings are helpful to forecast their energy use and to enhance the control of their mechanical systems. However, these models are building-specific and require a tedious, error-prone and time-consuming development effort relying on skilled building energy modelers. Compared to white-box and gray-box models, data-driven (black-box) models require less development time and a minimal amount of information about the building characteristics. In this paper, autoregressive neural network models are compared to gray-box and black-box linear models to simulate indoor temperatures. These models are trained, validated and compared to actual experimental data obtained for an existing commercial building in Montreal (QC, Canada) equipped with roof top units for air conditioning. Results show that neural networks mimic more accurately the thermal behavior of the building when limited information is available, compared to gray-box and black-box linear models. The gray-box model does not perform adequately due to its under-parameterized nature, while the linear models cannot capture non-linear phenomena such as radiative heat transfer and occupancy. Therefore, the neural network models outperform the alternative models in the presented application, reaching a coefficient of determination R2 up to 0.824 and a root mean square error down to 1.11 °C, including the error propagation over time for a 1-week period with a 5-minute time-step. When considering a 50-hour time horizon, the best neural networks reach a much lower root mean square error of around 0.6 °C, which is suitable for applications such as model predictive control.