How to make the intelligent temperature controller more efficient< br />
Building energy consumption accounts for a high proportion of overall energy consumption, and its carbon dioxide emissions are also high. Therefore, improving building energy efficiency is not only a cost saving measure, but also an important climate change mitigation strategy. Therefore, Architecture & quot; Smart & quot; Globalization is rising all over the world.
The intelligent building realizes the automation of heating, ventilation and air conditioning (HVAC), lighting, power and safety systems. Automation requires sensing data, such as indoor and outdoor temperature and humidity, carbon dioxide concentration and residential status. Intelligent building can make it more energy-saving by using data and combining various technologies. Since HVAC system accounts for nearly half of the energy consumption of buildings, intelligent buildings use intelligent thermostats to automatically control HVAC and automatically learn the temperature preference of building residents.
Researchers from MIT's information and Decision Systems Laboratory (lids) and skoltech scientists have designed a new intelligent thermostat, which uses a data efficient algorithm to learn the optimal temperature threshold in a week. The research results were published in the journal applied energy.
In order to speed up the learning process, researchers used a method called manifold learning, which is complex and & quot; High dimension & quot; The function of is called & quote; Manifold & quot; Represented by simpler and lower dimensional functions. By using manifold learning and building thermodynamics knowledge, researchers use a set of & quot; Threshold & quot; Strategy replaces a general control method, which can have many parameters, and each strategy has fewer and more interpretable parameters. Algorithms developed to learn the best manifold require less data, so they are more data efficient.
The algorithm developed for the thermostat adopts a method called reinforcement learning (RL), which is a data-driven sequential decision-making and control method. In recent years, it has attracted much attention in games such as backgammon and go.
The new RL algorithm of intelligent thermostat is & quot; Event trigger & quot; This means that they make decisions only when certain events occur, rather than according to a predetermined timetable. These & quot; Event & quot; It is defined by certain conditions reaching the threshold, such as the temperature in the room falling out of the optimal range. This reduces the frequency of learning and updating and reduces the computational cost of the algorithm.
Researchers believe that their methods and algorithms are also applicable to other physics based control problems, such as robots, autonomous vehicles and transportation, in which data and computational efficiency are very important.