When you think of Google, what do you associate it with?
If you’re like most folks, you might say, The internet. Search engines. Or even the Google Pixel.
That said, Google is now doing something drastically different – it’s using machine learning to improve its electricity production (more specifically, wind energy production). Curious to learn exactly how machine learning can help improve wind energy production? Read on to find out more!
Google’s 100% renewable energy purchasing
Here’s some background context: last year, Google hit its goal of offsetting its energy usage with 100 percent renewable sources. If you’re wondering how they did this, they’ve been:
- Investing in and purchasing energy from solar and wind farms to power their data centres and facilities, and
- Buying Renewable Energy Certificates (RECs) to offset their standard power grid usage in other markets.
Where possible, Google tries to power its facilities with clean energy. However, it acknowledges that it’s difficult to source every unit of electricity from a renewable source.
Obviously, it’s not feasible for Google to build enough wind and solar farms to power all its facilities, especially when you take into consideration that there are local utility monopolies in various parts of the world.
Bearing this in mind, Google also purchases RECs, which are “tradable, non-tangible energy commodities that represent proof that 1 megawatt-hour of electricity was generated from an eligible renewable energy resource and fed into the shared system of power lines which transport energy.”
The problem with using wind farms to generate power
One of the wind farms that Google is currently using is Minco-II, a 100-megawatt farm that’s located about 50 miles outside of Oklahoma City. The farm is helmed by NextEra Energy Resources, which is the largest operator of wind and solar-generating sites in the US.
Since 2016, Google has been using all 64 turbines at Minco-II. All in all, these turbines are capable of generating 438,000 megawatt hours of electricity in a year – this can power the equivalent of 30,000 homes.
In using wind farms to generate power, though, Google encountered a major issue. As DeepMind (Google’s AI subsidiary) notes in a blog post, the variable nature of wind itself makes it an unpredictable energy source, and one that’s “less useful than an alternative energy source that can reliably deliver power at a set time”.
Simply put: it’s tough to tell how much energy a given wind farm will generate, and the amount of energy that farm has to store or deliver to the grid changes from day to day.
In order to solve this problem, Google and DeepMind are now using machine learning and artificial intelligence to predict wind output.
How Google is using AI to predict wind output
Here, Google sat down with DeepMind to train a neural network to accurately predict wind power output 36 hours ahead of the power being generated.
How does this work? First, feeding weather data and historical wind turbine performance data into the network allows it to predict wind power output. Using these predictions, DeepMind’s computer model can then recommend “optimal hourly delivery commitments” to the grid 24 hours in advance.
This, in turn, lets Google schedule set deliveries of energy output, which means that Google can now rely on its wind farms to deliver a set amount of electricity at a particular time.
While Google and DeepMind researchers note that it’s impossible to eliminate the variability of the wind, they also say that their early results suggest that they can use machine learning to make wind power sufficiently more predictable and valuable.
Both companies are hoping that this new approach can help “bring greater data rigour to wind farm operations”, and allow wind farm operators to make “smarter, faster, and more data-driven assessments” of how their power output can meet electricity demand.
20% boost in Google’s wind value
Curious to learn about the results Google achieved with all this experimenting?
According to Will Fadrhonc, the Carbon Free Energy Program Lead at Google, machine learning has boosted the value of Google’s wind energy by 20 percent. This is compared to the baseline scenario of no time-based commitments to the grid.
Now that Google has shown that wind energy can be a reliable source of energy, they’re hoping that this will “strengthen the business case for wind power” and increase adoption rates of carbon-free energy worldwide.
Want to go green, and do your part for the environment?
Unfortunately, wind farms aren’t a thing in Singapore – we have ONE wind turbine offshore at Semakau Landfill, but that’s about it.
That said, you can still do your part for the environment by purchasing your electricity from a retailer that offers green electricity plans, such as iSwitch.
Keep in touch and join our mailing list to receive our monthly newsletter!