How Google is boosting the value of wind energy by using statistics

Author: Carlos Grajales

A few months ago, the International Energy Agency reported record high global carbon emissions in 2018 [1]. Once again protesters turned to the streets; once again a debate of academics, politicians and businessmen started. Once again, bold pronouncements from parties all over the globe, yet in the end few concrete policies emerge from all of this. But this grim briefing (of all global warming related news of the last couple of years) don’t have to sadden you down. Not at all, because if you are reading right now it means you are the kind of people who can help the planet and contribute in relieving earth (a bit) from carbon emissions. That’s right, developments in statistical analysis are helping fight global warming, particularly some developments within a statistically minded company: Google.

By now, we should be used to hearing about Google using statistics for some weird, pet project that involves the use of data in not so obvious areas of research. I can remember writing about a collaborative effort of behavioural Science PhDs and statisticians who studied the effects of sugar consumption in Google employees, about six years ago [2]. It also brings to mind the high level of statistical analysis involved in the Google Translate app [3], which, by the way has been recently improved with newer statistical models. As such, it wouldn’t be that much of a surprise to know that Google has now used statistics to improve the capabilities and economic viability of alternative forms of energy [4].

For those unaware of it, Google has its own Carbon Free Energy Program [5]. The program is part of Google’s effort to depend completely on green energy, by relying on large-scale, long-term contracts to buy renewable energy directly from producers in the U.S. and abroad, mostly to fuel their huge, energy-hungry data centers [5]. Thanks to this program, by 2017 Google was the world’s largest corporate buyer of renewable power. It was as part of this initiative that last year, DeepMind and Google analysts started working on ways to improve wind energy production using statistical methods [4] [6].

The reason why wind power was chosen for such a project lies in the fact of its complexity. Unlike other forms of electricity generation, wind farms are an unpredictable energy source, as their production relies heavily on random weather conditions, thus becoming less useful and less economically viable [4]. To be profitable, energy generation depends heavily on distribution, ensuring that generated power can be sent to where it is needed. However, it is hard to ensure appropriate energy distribution if you don’t know how much energy you will have to distribute, a common problem of wind farms. Being dependent on natural weather conditions, electricity production in wind farms is highly variable. Here come the statisticians, which happen to be good with variation.

The team of analysts involved in the project developed a statistical model to predict power output of one of the Turbine farms that distribute power to Google, in the United States. For the model, they used as input all available weather forecasts and historical turbine data generation and fit a neural network, a non-linear model, that was able to predict energy output of the Turbines 36 hours ahead [4] [6]. This became a huge boost for the plant’s viability, as scheduled distribution is more profitable, and the hourly forecasts produced by the model were accurate enough to increase production value. Google claims their model boosted the value of wind energy in this farm by roughly 20 percent, compared to the baseline scenario of no forecasts available [4].

Making green energy sources more cost-effective is the most efficient way to promote the use of green energy. But this is not the only way statisticians can save the planet, as data analysis can also help in the other side of the problem: reducing energy consumption. And for that, we can also use Google as example.

In 2016, another statistically based project used a statistical model to define the optimal way to cool data centers [7]. As Google’s servers analyze data and produce results for their users, their facilities and computers tend to get hot, requiring extensive efforts to keep them cool and safe from overheating. A system of fans, windows and air conditioners have the job of ensuring these servers avoid dangerous temperatures. Google’s 2016 model used data that collects temperature from a variety of sensors to suggest optimal ways to keep a safe temperature while reducing energy consumption: when things get too hot, air conditioning systems should be set to max, but if the hardware is barely warm, just opening a few windows can be enough. This optimization helped Google reduce their overall power consumption in data centers by 15 percent [7].

So even if politicians can’t get along to define ways to help our world, it is encouraging to know that we, as statisticians, do have a helping hand to aid the planet.


[1] Chestney, Nina Global carbon emissions hit record high in 2018: IEA. Reuters (March, 2019)

[2] Grajales, Carlos. Why are Google using M&Ms in their latest statistics project? StatisticsViews (September, 2013) ]

[3] Grajales, Carlos. The statistics behind Google Translate. StatisticsViews (June, 2015)

[4] Withersppon, Sims et al. Machine learning can boost the value of wind energy. Google Official Blog (Feb, 2019)

[5] Hölzle, Ulz. 100% renewable is just the beginning. Google Sustainability Website (2018)

[6] Statt, Nick. Google and DeepMind are using AI to predict the energy output of wind farms. The Verge (Feb, 2019)

[7] James, Vincent. Google uses DeepMind AI to cut data center energy bills. The Verge (July, 2016)