Google develops a system that predicts the wind power generation volume of the next day by machine learning and provides efficient power supply

by Johan Bos

Google has developed " systems to forecast floods by machine learning " and " systems that cool down data centers with AI control ", and we are trying to improve the world using machine learning. Such Google is also expanding into renewable energy fields such as wind power generation and cooperates with Deep Mind of AI development company acquired in 2016, predicts the wind power generation amount of the next day using machine learning algorithm, and it is efficient Today announced that it has developed a system that enables power supply.

Machine learning can boost the value of wind energy
https://www.blog.google/technology/ai/machine-learning-can-boost-value-wind-energy/

Google optimizing wind farms w / DeepMind to predict output - 9to5 Google
https://9to5google.com/2019/02/26/google-deepmind-wind-farms/

Renewable energy that does not emit greenhouse gases during power generation is an important field to cope with global warming and the accompanying climate change. On the other hand, renewable energy is often seen as insufficient as a completely replacing conventional power generation system due to the nature of generating power using natural power such as solar power and wind power.

Along with the huge and low cost of turbines, wind power generation is already said to be a low cost category among a number of power generation technologies, and the number of hiring has also increased substantially. On the other hand, the wind blowing on the earth is not always constant, and it is considered not useful as compared with a power generation system that uses fossil fuel and the like that can produce a certain amount of electricity at a fixed time.

Google is constructing several wind farms in the project to promote renewable energy, among which wind power plants in the middle of the United States have facilities capable of generating 700 megawatts. Google and Deep Mind applied the machine learning algorithm to this power generation facility and conducted an experiment that predicts "how much wind power generation facility produces electricity before actually generating electricity".

by Bru-nO

Machine learning algorithm used normal weather forecast and data collected by past wind turbines. By using these data, it is said that a system that outputs wind power generation amount predicted 36 hours before actual power generation is constructed.

This is a graph comparing the amount of power generation predicted by DeepMind's machine learning algorithm and the actual amount of power generation. The blue line shows the predicted power generation amount by the algorithm, and the gray line shows the actual power generation amount. You can see that the actual amount of generated electricity fluctuates upward and downward, but the approximate result that the amount of electricity generation changes is consistent with the prediction.

Based on this prediction, DeepMind 's system will recommend the optimum amount of electricity to be transmitted to the grid of the next day every hour. A system that can send the power set at the set time to the power grid is very important and so far Google's wind farms have increased the value of wind power by 20% more than before . Google says it will continue to refine the machine learning algorithm, "Although we can not completely eliminate the influence of the wind, we can make wind power generation predictable and value by machine learning algorithm" I said.