Google puts an Artificial intelligence system responsible for cooling its data center after the system proved it could reduce power consumption. Now, Google and its DeepMind amplification company are going further in the realization of the project. Instead of implementing recommendations by staff, the IA system directly controls the cooling of data centers that run services such as Google Search, Gmail, and YouTube.
“This cloud-based control system, the first of its kind, now offers safe energy savings. several Google data centers”Said Google. Data centers use a lot of energy and cloud computing demand even small adjustments in areas such as cooling can result in significant time and cost savings. Google’s decision to use its own DeepMind system is also a good tool for its AI business.
Every five minutes, the AI takes a snapshot of the data center cooling system from thousands of sensors. This data feeds deep neural networks, which predict how different choices will affect future energy consumption.
The AI system then identifies the changes that can reduce the power consumption, which is then returned to the data center, verified by the local control system and implemented. Google said that giving responsibilities to the AI came at the request of its data center operators who said that implementing the recommendations of the AI system required too much effort and effort. supervision.
“We wanted to save energy by reducing overhead costs for operators.” The automation of the system allowed us to implement more specific actions more accurately while making fewer mistakes. ” explains Dan Fuenffinger, head of the Google Data Center.
Google has implemented security measures to ensure that the AI will behave as expected. For example, for each potential share, the AI must calculate its confidence in the fact that it is a good deed. Unreliable actions are eliminated.
The potential actions calculated by the AI are compared to an internal list of security constraints and local data center operators can take control when they need it.
Google has used different forms of AI in its data centers for several years. He began by exploring how to use AI to make these mega-data more efficient by using a neural network system. formed on different scenarios. Deep neural networks have been trained on historical data already collected by thousands of sensors within the data center, such as temperatures, power, and pumping speed.
The neural networks have been trained to determine the future average power utilization efficiency (PUE), the ratio between the total energy consumption of the building and computer energy consumption. Two other neural networks were trained to predict future data center temperature and pressure over the next hour, to ensure that any adjustments did not exceed the operational limits of the data center.
Google found that machine learning systems were able to consistently achieve a 40% reduction in the amount of energy used for cooling, a 15% reduction in overall CPUE. The AI standalone control system initially resulted in a 12% improvement, which improved in nine months to reach about 30%, with further improvements expected with more training data. Google said that in the long run, there is a potential to apply the technology in other industrial environments.