Researchers from the laboratory of DeepMind, spoke about the neural network that can predict the performance of wind turbines over 36 hours based on historical data about their performance and the weather in a particular location. According to them, the use of this algorithm allowed to improve the performance of wind farms Google’s 20 percent.
Wind power is one of the best and most used ways of alternative energy production. Their work is highly dependent on wind speed: rotor of the wind turbine needs to be at the height with a fairly high average wind speeds (approximately 4.5 meters per second). However, to reliably predict the pattern of wind strength at the location of the station can be difficult, because of which its performance may be reduced.
The developers at DeepMind (part of the holding Alphabet) was proposed to solve this problem with an algorithm that could reliably predict the energy generated by the wind farm in a particular area based on available data. They trained the neural network on data on wind force in a certain region, weather and energy produced at a specific wind turbine.