The problem with photovoltaic facilities and wind farms is that their operators can never predict how much power they will feed into the grid at a set time. Google and its AI subsidiary DeepMind now plan to change this. In a pilot project , the two companies are using a neural network to evaluate data and reports from several widely available weather services. Applying a series of machine learning algorithms, they can predict the wind conditions 36 hours earlier than with other available solutions.
But that’s not all: Google can use the data thereby acquired to predict the energy output of several wind farms in the US Midwest around one day in advance. The farms have a wind power capacity of 700 MW, as much electricity as is needed by a medium-sized city. This approach enables operators to configure their power plants and turbines accordingly, to better adapt power output to meet energy demand. Google and DeepMind are currently in the process of optimizing their machine learning software’s algorithms to more accurately predict the amount of power delivered by wind farms.