Back to Bigger Picture

BIGGER PICTURE Energy & climate

Making tomorrow’s wind power easier to promise

The turbine stays the same. The innovation is knowing enough about its next day to make a more useful commitment to the grid.

AI-generated illustration: Conceptual wind turbines connected to abstract future power-supply blocks.
Wind turbines connected to abstract future power-supply blocks.AI-generated illustration of the idea, not a photograph of the actual people, equipment or event.
Forecast horizon
36 hours
Reported gain
About 20% more economic value
Deployment studied
700 MW of US wind capacity

How the idea works

  1. Read wind and turbine history

    Weather forecasts join historical generation data.

  2. Forecast hourly output

    The model estimates generation up to 36 hours ahead.

  3. Schedule delivery

    Operators can make day-ahead hourly commitments.

Electricity has an appointment

A megawatt-hour is more useful when a grid knows when to expect it. Wind generation makes that promise difficult: the fuel arrives with the weather. In work reported in 2019, Google and DeepMind applied machine learning to 700 megawatts of wind capacity in the central United States to make the next day’s supply more predictable.

Sources: [1]

First predict, then commit

A neural network learned from weather forecasts and historical turbine data to estimate output 36 hours ahead. A second decision layer recommended hourly delivery commitments a day in advance. The distinction matters: an accurate prediction alone does not tell an operator what to promise. The useful product connects a forecast to an actual market and scheduling decision.

Sources: [1][2][3]

A better schedule, not a faster turbine

Google reported roughly 20% more economic value compared with a baseline without time-based commitments. That number describes the value of the electricity, not a 20% increase in physical generation. The potential human benefit is a renewable source that is easier to integrate economically into electricity supply; the company did not report a corresponding reduction in household bills.

Sources: [1][3]

The idea travelled beyond one owner

DeepMind subsequently described work to turn the approach into a Google Cloud product piloted by ENGIE. The IEA also identifies forecasting and scheduling as practical applications of AI in renewable power. Those developments establish wider interest, not independent verification of Google’s original savings. The 2019 result remains a provider-reported comparison with a specific baseline.

Sources: [2][3]

The original sources

  1. Google DeepMindMachine learning can boost the value of wind energy
  2. Google DeepMindUsing AI to fight climate change
  3. International Energy AgencyEnergy and AI in East Asia

What to explore next

A small change in flight path can leave less cloud behindA green box that helped doctors spot more polyps
Explore all 99 applications