MOUNTAIN VIEW, Calif. — Google announced that it has applied its DeepMind machine learning algorithms to its own data centers to reduce the amount of energy used for cooling by up to 40 percent.

One of the primary sources of energy use in the data center environment is cooling. However, Google said dynamic environments like data centers make it difficult to operate optimally for several reasons:

1. The equipment, how it is operated, and the environment interact with each other in complex, nonlinear ways. Traditional formula-based engineering and human intuition often do not capture these interactions.

2. The system cannot adapt quickly to internal or external changes (like the weather). This is because we cannot come up with rules for every operating scenario.

3. Each data center has a unique architecture and environment. A custom-tuned model for one system may not be applicable to another. Therefore, a general intelligence framework is needed to understand the data center’s interactions.

To address this problem, Google began applying machine learning two years ago to operate its data centers more efficiently. And over the past few months, DeepMind researchers began working with Google’s data center team to improve the system’s utility. Using a system of neural networks trained on different operating scenarios and parameters within its data centers, the team created a more efficient and adaptive framework to understand data center dynamics and optimize efficiency.

This was accomplished by taking the historical data that had already been collected by sensors — data such as temperatures, power, pump speeds, setpoints, etc. — and using it to train the neural networks. Since the objective was to improve data center energy efficiency, neural networks were trained on the average future power usage effectiveness (PUE), which is defined as the ratio of the total building energy usage to the IT energy usage. Google then trained additional neural networks to predict the future temperature and pressure of the data center over the next hour. The purpose of these predictions is to simulate the recommended actions from the PUE model, to ensure that the data center does not go beyond any operating constraints.

The model was tested by deploying on a live data center, including testing with the machine learning recommendations turned on and with them turned off.

Google said the machine learning system was able to consistently achieve a 40 percent reduction in the amount of energy used for cooling, which equates to a 15 percent reduction in overall PUE overhead after accounting for electrical losses and other non-cooling inefficiencies. It also produced the lowest PUE the site had ever seen.

Because the algorithm is a general-purpose framework to understand complex dynamics, Google said it plans to apply this to other challenges in the data center environment and beyond in the coming months.

For more information, visit

Publication date: 8/9/2016