How a Google blog post inspired us to bring Big Data to the resources & energy industries

Around the middle of last year, I came across a blog post by Deepmind, a London-based artificial intelligence company owned by Google.

As a long-time energy efficiency engineer, the prospect of anyone gaining a 40% improvement in their energy use sounded incredible. I had to know more.

Google, like all companies relying on large data centres, is a heavy user of electricity. As you’d expect, the racks of servers consume lots of power — but it’s often not appeciated that they also produce a lot of waste heat.

Computer systems don’t like to get too hot. Just ask anyone with a gaming PC how much heat they produce. So data centres require significant industrial cooling systems to remove the waste heat and keep the servers cool.

Every watt of power that is used in a computer is converted into heat. With demand for data centres rising relentlessly, the demand for cooling will also continue to rise. The cooling systems themselves are also big electricity users, running refrigeration compressors, cooling water pumps, fans and other heavy industrial equipment.

The engineers at Deepmind recognised that their data centre cooling systems had room for improvement in their energy use. There are many operating parameters in cooling systems that can be adjusted: cooling water flow rates, number of pumps on or off, number of cooling tower cells operating, and the distribution of cooling water and refrigerant throughout the cooling system.

The Deepmind engineers applied the tools of machine learning to model the operation of one of Google’s data centre cooling systems. Using the historic cooling system plant data — temperatures, flows, pressures and so on — as well as data on data centre power usage and weather data, they built neural network (sometimes called “machine learning”) models. These models allowed Deepmind to estimate the energy consumption of the cooling system as a function of these operating parameters.

Using these neural networks, they used computer-based optimisation to find a new set of operating settings that enable their cooling system to use significantly less energy. How much less? Take a look at this:

Source: Deepmind

The graph above shows a machine learning (“ML control”) improvement in energy use of 40% for the cooling system, or a 15% improvement in a power use effectiveness (ratio of the total building energy usage to the IT energy usage). It seems Google are not above exaggeration in their headlines.

Data centres are one source of potential energy improvements, but many other big industries — oil & gas, mining, chemicals, cement, and so on — use even more energy in the form of gas, electricity and liquid fuels and have not yet taken advantage of their plant data in this way.

I was so impressed by the results of Deepmind’s study, I started searching for a capable data scientist to help me move into the use of Data Science, including machine learning, for big energy users across industry. Since early this year I have been working with my co-founder Tim Maher, an accomplished engineer and data scientist, to build Sustainable Data.

I’ve worked for almost 20 years across heavy industry, and I can see the huge potential of the wealth of historic plant data available to improve energy efficiency, reduce emissions, increase reliability and minimise waste.

We look forward to bringing you more info about big data, analytics, measurement, machine learning and other techniques to help big industry save emissions, reduce energy use, improve reliability and reduce waste.

Sign up to our newsletter to keep up to date.

One clap, two clap, three clap, forty?

By clapping more or less, you can signal to us which stories really stand out.