The new smart thermostat algorithm will learn optimal temperature thresholds within a week.


For both energy usage and user needs, a smart thermostat easily learns to optimize the microclimate of the house.
In the U.S., buildings account for around 40 percent of energy consumption and are responsible for one third of global emissions of carbon dioxide. Making buildings more energy efficient is not only a cost-saving measure, but also a vital solution to climate change mitigation. The rise of “smart” buildings, which are becoming the norm around the world, is therefore rising.

Systems like heating, ventilation and air conditioning (HVAC), lighting, electricity and protection are automated by smart buildings.

Sensory data such as indoor and outdoor temperature and humidity, concentration of carbon dioxide, and occupancy status are important for automation. In a mix of innovations, intelligent buildings use data that can make them more energy efficient.

Since HVAC systems account for almost half of the energy usage of a building, intelligent buildings use smart thermostats that can automate HVAC controls and learn the temperature preferences of the occupants of a building.

Researchers from the MIT Laboratory for Knowledge and Decision Systems (LIDS) collaborated with scientists from Skoltech in a paper published in the journal Applied Energy to create a new smart thermostat that uses data-efficient algorithms that can learn optimal temperature thresholds within a week.

The implementation of smart buildings is hampered by the time-consuming process of collecting data from buildings, despite recent advancements in Internet-of-Things technology and data analytics,”Despite recent advances in Internet-of-Things technology and data analytics, the implementation of smart buildings is hindered by the time-consuming process of collecting data from buildings” (IDSS). To learn how to work optimally, smart thermostat algorithms use building data, but gathering the data will take months.

The researchers used a technique called manifold learning to speed up the learning process, in which complex and ‘high-dimensional’ functions are represented by simpler and lower-dimensional functions called ‘manifolds.’

The researchers substituted a generalized control method using Manifold Learning and knowledge of building thermodynamics that can have several parameters with a set of “threshold” guidelines, each with fewer, more interpretable parameters.

Algorithms designed to learn optimal multiples need less information and are therefore more effective in terms of data.

The algorithms designed for the thermostat use a technique called reinforcement learning (RL), a sequential decision and control approach guided by data that has gained a lot of attention for mastering games such as backgammon and Go in recent years.
In order to learn a good game strategy, we have powerful simulation engines for computer games that can generate a wealth of data for the RL algorithms,”We have efficient simulation engines for computer games that can generate a wealth of data for the RL algorithms to learn a good game strategy,” “However, for microclimate control in buildings, we don’t have the luxury of Big Data.”

Hosseinloo is able to apply insights from statistics and modern computing to real-world physical structures, with a background in mechanical engineering and experience in methods such as RL. “My main motivation is to slow down or even prevent an energy and environmental crisis by improving the efficiency of these systems,” he says.

The latest RL algorithms of the smart thermostat are “event-driven,” meaning they only make decisions when those occurrences occur, rather than on a fixed schedule.

Certain situations that exceed a threshold – such as when the temperature in a room falls out of the optimum range – describe these “events” This makes less frequent learning updates possible and makes our algorithms less costly in terms of computing,”This allows for less frequent learning updates and makes our algorithms less computationally expensive,”
For learning algorithms, computing capacity is a possible limitation, and computational resources rely on whether the algorithms run in the cloud or on a computer itself – like a smart thermostat. “We need learning algorithms that are both computationally and data efficient,” Hosseinloo says.

Beyond lowering emissions and reducing costs, energy-efficient buildings give other advantages.

The “microclimate” and air quality of a building can directly impact building occupants’ efficiency and decision-making.

Microclimate management is an important problem for governmental control, considering the many far-reaching economic, environmental and social impacts.


Leave A Reply