The University of Michigan CLaSP Solar Forecast Team claimed the final prize in the American-Made Solar Forecasting Competition held in March. Sponsored by the U.S. Department of Energy, the competition is designed to better enable solar industry stakeholders with state-of-the-art solar forecasting capabilities.
The team, which consists of Prof. Xianglei Huang and Dr. Xiuhong Chen, employed a hybrid forecast method: a machine-learning forecast scheme trained with past observations and a bias correction method based on numerical weather prediction.
“The team has never done such a forecast competition before; it is an excellent experience for us, and we certainly have learned something from it,” said Huang. “Some other winners have already provided commercialized solar forecast services to their customers. It is exciting to see the algorithm fresh out of our research can outperform theirs.”
According to the website, the American-Made Challenges are a series of prize competitions, in partnership with the National Renewable Energy Laboratory (NREL), that are designed to incentivize the nation’s entrepreneurs to reenergize innovation and reassert American leadership in the energy marketplace. The Solar Forecasting Competition aims to increase the use of the Solar Forecast Arbiter, an open platform developed by the University of Arizona, to allow for the transparent, rigorous, and consistent analysis and evaluation of solar forecasts.
As for this Solar Forecasting competition, participants were required to make a probabilistic hourly forecast, with a lead time of 24-48 hours and for four consecutive weeks, for solar radiation reaching the ground at ten different sites across all climate zones in the U.S. The forecast performance was evaluated against a baseline forecast algorithm.
The team ranked first for the forecast performance, as announced on March 25. Thirty-four teams participated in the competition, including the U-M CLaSP team, with the anonymized team name of Prompt Molly. It was nearly a close tie with the second team, but the margin over the other leading teams was significant: the team in the third place gave a forecast performance only 68% as good as the U-M CLaSP team. Each winner was awarded $50,000 in cash prizes.
According to the National Renewable Energy Laboratory, probability-based forecasts are regarded as a critical tool for cost-effective reserve allocation and unit scheduling. Considering a future with high-penetration weather-bound renewable energy generation is becoming increasingly important. Independent system operators agree that solar forecasting is an important part of system operations, and forecasting becomes even more important as the fraction of variable weather-dependent generation increases.
They maintain that the deployment and integration of these forecasts into energy management systems has not been widely adopted.
The Solar Forecasting Prizeaims to address this by:
- Increasing stakeholder awareness of the state of the art in solar forecasting.
- Incentivizing the participation of a broad range of competitors from the solar forecasting industry and research and development space.
- Growing industry knowledge of the Solar Forecast Arbiter (SFA) platform and its potential.
- Promoting the adoption of uniform and transparent metrics and specifications for solar forecasts using SFA (or similar platforms) by forecast end-users.
- Identifying algorithms that perform better than a baseline probabilistic forecast.
For more information, visit: https://www.herox.com/SolarForecasting/updates
