The steady advance of artificial intelligence (AI) continues with Google and reveals the latest innovation that supports data center cooling and industrial control.
In 2016, Google and DeepMind collaborated on the development of an AI-powered recommendation system with the goal of improving the energy efficiency of Google's data centers.
And now Google has used and expanded this same AI system to remove recommendations that have been implemented by people and instead arrange the data center cooling itself – under expert supervision, of course.
So how does it work?
"Every five minutes our cloud-based AI takes a snapshot of the data center cooling system from thousands of sensors and carries it into our deep neural networks, which predict how different combinations of possible actions will affect future energy consumption," Google & # 39; s Amanda Gasparik and DeepMind & # 39; s Chris Gamble and Jim Gao reported in a release.
"The AI system then identifies which actions minimize energy consumption while complying with a robust set of safety restrictions that are returned to the data center, where the actions are verified by the local control system and then implemented."
The idea came from a trial and error approach to the previous AI recommendation system, while Google's data center operators praised the system for revealing new best practices (such as spreading the cooling load over more equipment instead of less), where the recommendations actually exercise required too much effort from the operator and supervision.
"We wanted to save energy with less overhead for the operator, and automating the system allowed us to perform more detailed actions faster and make fewer mistakes," said Dan Fuenffinger, Google's data center company.
That is why Google has implemented the new AI system to remove part of the manual implementation.
Google has thousands of servers and it is crucial that they all work reliably and efficiently. In light of this, the company claims that it has adjusted the AI agents from the outset to safety and reliability, using eight different mechanisms to ensure reliable system behavior.
For example, a simple step that Google has taken is estimating the uncertainty. There are billions of actions involved in the data centers and for each of these the AI agent determines his confidence whether it is a good step – low-confidence actions are taken out of the treatment.
Another example is two-level verification, in which optimal actions calculated by the AI are screened on the basis of an internal list of security restrictions set by the data center operators. In addition, the operators always have control and can exit the AI control mode at any time.
Although the AI system has the ability to determine the actions of data centers, Google says that it deliberately limited the optimization limits of the system in an attempt to give priority to safety and reliability.
After having been in operation for a few months, the system has proven itself with a consistent energy saving of about 30 percent on average. Furthermore, Google expects that this will improve over time, because the system will have access to more data and the boundaries will be expanded as the technology ages.
"It was great to see that the AI learns to take advantage of winter conditions and produce colder than normal water, reducing the energy needed for cooling within the data center." Rules do not get better in time, but AI does it ", says Fuenffinger.
Google claims that it is enthusiastic about the technology and that data centers are just the beginning because it believes that the AI system can be implemented in various other industrial environments.
Interested in this topic?
We can put you in touch with an expert.