Researchers at IBM's T.J. Watson Research Center and the U.S. National Renewable Energy Laboratory have developed a machine-learning algorithm that uses weather models and data to predict days or weeks in advance how much power solar and wind plants will generate for the U.S. power grid.
The researchers say the new system, called Self-Learning Weather Model and Renewable Forecasting Technology (SMT), is about 30 percent more accurate than the state-of-the-art weather forecasting systems used by the U.S. National Weather Service.
The algorithm relies on massive amounts of data, more than 1 terabyte every day, gathered from weather-monitoring stations, solar and wind plants, and weather satellites, to constantly improve its models.
"It's providing a forecast for solar, wind, and other environmental parameters," says Hendrik Hamann, research manager at the Watson Research Center. "It learns from solar plants [and] weather stations, and constantly adjusts and improves the forecast."
Renewable energy from solar and wind plants currently accounts for 5 percent of the electricity generated in the U.S., but that share is expected to increase to more than 25 percent by 2050. However, the amount of power wind and solar facilities can generate varies depending on the weather, making systems such as SMT crucial to maximizing their use.
View Full Article
Abstracts Copyright © 2015 Information Inc., Bethesda, Maryland, USA