Carbon-free technologies, such as renewable energy, are helping to combat climate change, but many of them have not reached their full potential. Consider wind power: Over the past decade, wind farms have become an important source of carbon-free electricity as the cost of turbines has fallen and deployment has increased. However, the fluctuating nature of wind makes it an unpredictable source of energy – less useful than one that can supply power reliably at a given time.
In search of a solution to this problem, DeepMind and Google began applying machine learning algorithms to 700 megawatts of wind power capacity in the central United States last year. These wind farms – part of Google’s global fleet renewable energy projects– To jointly produce as much electricity as a medium-sized city needs.
Using a neural network trained in widely available weather forecasts and historical turbine data, we configured the DeepMind system to predict wind power 36 hours before actual production. Based on these predictions, our model recommends how to make optimal hourly commitments to the power grid all day in advance. This is important because timed energy sources (i.e., can supply a certain amount of electricity at a given time) are often more valuable to the grid.
While we continue to refine the algorithm, the use of machine learning in our wind farms has yielded positive results. So far, machine learning has increased the value of our wind energy by about 20 percent compared to the baseline scenario with no time-based commitments to the grid.