Making more efficient thermostats that quickly acclimatize to the ambiance of buildings, is today’s need and is increasingly gaining popularity. The sedentary work culture has created a need for a climate tolerant workspace. And in a country like the US, buildings specifically consume nearly 40 percent of the energy produced. Not only this, but they also account for the world’s one-third carbon-di-oxide exhausts which is huge. Thus, making more efficient thermostats that control and stabilize the microclimate of the buildings well, shall do the needful. And will also minimize energy consumption to a great extent.
HVAC in Smart Buildings
Now smart buildings, regulate heat, ventilation, and air conditioning i.e. HVAC to maintain an optimum level of internal microclimatic conditions. This is required to battle the temperature extremes and achieve the best working and living conditions for humans. Now to automate such conditions, temperature, and environment-sensitive data is required. Factors such as humidity conditions, internal and external temperature, level of carbon-di-oxide, and also the number of persons or occupants, make an important part of the data. And as far as the smart buildings are concerned, it is expected for them to have an intelligent combo of data and technology. And thus designing and making more efficient thermostat systems for buildings.
MIT and the LIDS
A team of researchers at the MIT Laboratory for Information and Decision Systems with a team from Skoltech developed a thermostat system. The system works on data-efficient smart algorithms that learn more and more about the temperature thresholds in not more than a week, which is a commendable speed. But, at the same time, a professor Munther Dahle, the director of the Institute for Data, Systems, and Society (IDSS) reveals a relevant problem that comes along with making more efficient thermostats. It is the time-incurring process of gathering data that hampers the overall implementation of such smart buildings. The collection of data many a time takes months together, which is huge and not very cost-effective.
Manifold Learning (ML)
Thus, to rev up the speed of data collection, researchers rely on a process known as “manifold learning”. Now manifold learning is a process where higher-dimensional complexities are presented through lower-dimensional functionalities. And with the combination of manifold learning and thermodynamics knowledge, the team was able to replace the existing method with a new one, that satisfies fewer parameters. The new system used better and more data-efficient algorithms and thus is capable of saving an ample amount of time.
The methodology of Reinforcement learning (RL)
The reinforcement learning methodology has gained immense popularity in the recent past. As it is a data-centric approach that aids decision making, it is also used in the making of more efficient thermostats. Games like backgammon and Go are good examples of data-driven reinforcement learning. The smart thermostats work based on an event-trigger mechanism i.e. they take decisions only when an event occurs. And do not work on pre-decided mechanisms.
What do researchers say about making more efficient thermostats?
Mr. Ashkan Haji Hosseinloo is the lead author of the original paper and also a postdoc at the LIDS. As the paper was published in the journal ScienceDirect, Mr. Akshan reveals how they have utilized the simulation engines that work for computer games, to even work for such thermostats. Coming from a mechanical engineering background too, he is adept at applying modern computer technologies to real-world systems. And he further adds that a strong combination of computational-efficient and data-efficient algorithms is required to draw closer to the desired goal. His co-author Henni Ouerdane adds to it and says that any system that has minimum user intervention is always a promising one.
Also Read: Device that detects plants’ stress
Conclusion:
The pursuit to achieve smart buildings seems to come to a glorious finish. Scientists have worked hard towards utilising the versatility of reinforcement methodology and manifold learning. And they say that these can be efficiently applied to a huge array of physics-driven systems like autonomous vehicles, robotics, and transportation, etc. Thus, this reinforcement methodology coupled with a data-driven algorithm seems promising for the future of smart buildings. The large-scale economic activities that are carried out in large buildings have many economic and societal impacts. And many a time become problematic for the governments, managers, and even for the property owners. Thus mitigating and regulating microclimatic conditions within these buildings and making more efficient thermostats was required to maximize the efficacies of occupants.
