Engineers working at the CSEM have developed a new machine-learning methodology that lays the foundation for artificial intelligence to be utilize in applications that until recently were considered too sensitive. The new method, which has been tested by running climate-control simulations on a 100-room building, is supposed to cut energy costs by about 20%.
Back in 2016, a supercomputer beat the world Go champion. Go is an extremely complicated board game that takes a lot of experience to master. So, how did the computer do it? By using reinforcement learning. Reinforcement learning is a type of of artificial intelligence where computers are able to train themselves after being programmed with just simple instructions. The computer can learn from its mistakes and over time can become highly powerful.
The main drawback with reinforcement learning is that it can’t really be used in several real-life applications. The reason for this is because in the process of training themselves, computers at first attempt just about everything that they can before eventually finding themselves on the right track. This initial ‘trial-and-error’ phase can be problematic or even dangerous for certain applications like climate-control systems, where abrupt swings in temperature wouldn’t be ideal.
The CSEM (Computer Science, Engineering and Mathematics) engineers developed such an approach that it overcomes this issue. The team was able to show that a computer can first be trained a more basic, simplified theoretical model before being set up to learn on real-world systems with live data. So, once the computers begins the machine-learning process on the real-life systems, they wil be able to draw on what they learned learned previously in the safer training session. This makes it so the computers can get on the right track more speedily without having to go through periods of extreme volatility. The research done by the engineers was just recently published in IEEE Transactions on Neural Networks and Learning Systems.
It’s like learning the driver’s manual before you start a car. With this pre-training step, computers build up a knowledge base they can draw on so they aren’t flying blind as they search for the right answer.Pierre-Jean Alet, head of smart energy systems research at CSEM
The engineering team tested their system on a HVAC (Heating, Ventilation and Air Conditioning) system for a complicated 100-room building using a multi-step process. To begin, the computer was trained on a ‘virtual model’ that was built from simple equations that describe a building’s behavior roughly. Next, the computer was fed actual data a building. Things like how long blinds were open, the temperature, what the weather conditions were, and other factors into the computer to incrase the accuracy of its training. Lastly, the computer runs its reinforcement-learning algorithms to find the best way to manage the HVAC system.
This discovery could open new opportunities machine learning and AI by expanding from its current limited use to applications where large deviations in operating parameters are important.