Researchers at North Carolina State University have demonstrated techniques that improve day-ahead solar power forecasts by up to 13% over the most consistently performing individual forecasting model.
The research team tested seven forecasting models using weather and power generation data from two California utilities — the Imperial Irrigation District and the Los Angeles Department of Water and Power (LADWP) — covering more than 20,000 hours of operational data from 2019 to 2022.
The team found that no single model performed best in every case, and that combining forecasts from multiple machine learning models through ensemble approaches yielded the most significant accuracy improvements.
“Our results demonstrate that combining multiple machine learning-based models can provide more robust predictions and help improve the reliability of solar integration into power systems,” said Anderson De Queiroz, associate professor of civil, construction and environmental engineering at NC State, in a statement.
Two ensemble approaches were tested: weighted averaging, which combines forecasts from separately trained location-specific models and assigns greater weight to better-performing ones, and multi-input, which allows each model to draw on weather data from multiple locations simultaneously.
Weighted averaging produced improvements of up to 11% for the Imperial Irrigation District, while the multi-input approach yielded improvements of up to 13% for LADWP.
A key finding of the research is that no universal forecasting strategy performs equally well across all regions, underscoring the importance of testing and fine-tuning models for specific geographic areas.Â
Read more here.
