Alphabet Company DeepMind Uses AI To Boost Wind Energy Value By 20%
- February 27, 2019
- 593 views
How effective is artificial intelligence in integrating renewable energy into electric grids? Alphabet subsidiary DeepMind may have an answer.
The Alphabet subsidiary, which specializes in applying artificial intelligence techniques to problems, sourced data from a 700 MW wind farm in Oklahoma and trained its neural networks to predict power output generated using wind power upto 36 hours in advance.
For those who are unfamiliar with machine learning models, it involves training AI systems to detect patterns in the data and make predictions based on past inferences. DeepMind’s networks used weather data and wind turbine data as inputs and were able to predict optimal hourly grid commitments a day in advance. “Machine learning has boosted the value of wind energy by roughly 20%,” the post’s authors wrote. “We can’t eliminate the variability of the wind, but our early results suggest that we can use machine learning to make wind power sufficiently more predictable and valuable.”
Google acquired London-based DeepMind in 2014. Since then, DeepMind has applied and tested machine learning technology to disparate endeavors, from an analysis of the strategy game Go to building 3D models of proteins with AlphaFold to research protein folds.
A Positive Development
To be sure, there are several parameters to consider while forecasting wind power. These parameters range from the technical details of physical infrastructure used to generate energy to complex and granular weather data.
DeepMind states that it used weather data and wind turbine data as inputs to generate predictions. But they haven’t provided any detail regarding the quality or source of their data. This is important because quality of data is important in making numerical weather predictions (NWP) or using statistical models. Typically, performance of wind turbines decreases by an average of 1.6% per year. Performance, in this case, refers to the capacity factor of wind turbines. What’s more, forecast performance varies with several factors from distribution of wind speeds relative to the curve to shape of plant-scale power curve.
That said, this is a promising development and could have utility in the future. Variations in power generated using renewable energy have increasingly emerged as an area of concern because they affect reliability and operational aspects of the grid. Advance notice of power output could help electric utilities make provisions for additional power, such as in the case of dispatchable power for baseload generation.
Interestingly, Germany’s Wind Power Management System (WPMS) already uses artificial intelligence to generate day-ahead and short term energy forecasts. A 2006 research report reinforced the importance of data quality in generate accurate forecasts for the grid. Siemens Energy started work on the Simulation Environment for Neural Networks (SENN) more than 20 years ago and the software is now being used to make renewable energy and demand forecasts in places as diverse as South Africa and Switzerland.