Researchers from North Carolina State University have demonstrated techniques that can improve day-ahead solar forecasts by up to 13% over the most consistently performing individual model. The work also emphasizes the importance of taking regional variables into account when creating forecast models.

“Solar power generation has expanded rapidly because it is both renewable and widely available,” says Yen-Hsi Chou, a postdoctoral research scholar at NC State and corresponding author of a paper describing the work. “But increasing use of solar can also make forecasting supply and demand more challenging because the availability of sunlight isn’t always consistent. Utilities need accurate day-ahead solar forecasts to be able to plan ahead.”

“As solar penetration continues to increase, forecasting uncertainty becomes a key consideration for energy planners and grid operators that need to balance supply and demand,” says Anderson De Queiroz, associate professor of civil, construction and environmental engineering at NC State and paper co-author. “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.”

The research team first looked at two types of models – statistical models and artificial neural networks – to identify a single model to use as a baseline. Statistical models identify historical patterns, whereas artificial neural network-based approaches are better suited to understanding nonlinear temporal relationships. They selected seven models to test.

The models were tested using weather and power data from 2019 – 2022 from two California utilities: the Imperial Irrigation District (IID) and the Los Angeles Department of Water and Power (LADWP).

“We wanted to use the models to find the relationship between weather data and solar power generation,” Chou says. “But the most interesting result was that no single model performed best in every case. We chose the most consistently performing model, BiLSTM, as a baseline and saw that by combining the forecasts from different individual models, we could further improve forecasting performance by over 10% in some cases, which was pretty impressive.”

The two ensemble approaches were weighted averaging, which combines forecasts from separately trained location-specific models and gives greater weight to better-performing models; and multi-input, which allows each model to use weather data from multiple locations.

The results showed that the effectiveness of each ensemble approach differed depending on the region. Weighted averaging provided improvements of up to 11% for IID, whereas multi-input yielded forecasting improvements up to 13% for LADWP.

“One of the key takeaways from this work is that there is no universal forecasting strategy that will perform equally well everywhere,” De Queiroz says. “Understanding the characteristics of each region and leveraging information from multiple models and locations is extremely important for developing forecasting tools that are both accurate and useful to support decision-making associated with real-world grid operations.”

“Overall, our findings indicate that ensemble methods do have the potential to further enhance predictions compared to an individual model, but the methods need to be tested and fine-tuned for the specific region where they will be used,” Chou says.

The work appears in the Journal of Cleaner Production and was supported in part by the National Science Foundation under award 2412711. Other NC State contributors were Arundhuti Haldar, a Ph.D. student in the department of electrical and computer engineering; and Shubh Nisar, a former graduate student in computer science.

-peake-