A machine learning approach to solar forecasting
ACEP summer intern RJ Michael gives a presentation on his project.
September 24, 2025
Can improving the accuracy of solar forecasts help energy planners utilize energy resources more effectively?
This is the question RJ Michael wanted to find the answer for during his internship with the Alaska Center for Energy and Power this summer.
If tomorrow, a sunny day with a cool breeze, is expected to produce a high number of kilowatt-hours, for example, energy forecasters could determine the offset from distributed solar energy and plan accordingly, Michael thought.
鈥淒istributed鈥 energy refers to energy generated at or near the location where it is consumed, rather than energy being produced at a large, centralized power plant and then transmitting over long distances.
ACEP summer intern RJ Michael shows a heat map of distributed solar energy (or energy generated at the location where it is consumed) in Golden Valley Electric Association鈥檚 service area in Fairbanks, Alaska. Areas in red represent higher concentrations of solar panels, while areas in blue represent fewer.
Under the mentorship of Phylicia Cicilio with the but formerly with ACEP, Michael set to work on real-time distributed solar energy forecasting.
Michael applied a machine learning approach, using models such as XGBoost, LSTM and Prophet. He studied historical distributed solar energy generation in conjunction with weather data, including solar irradiance (energy received from the sun), temperature and cloud coverage. Using this information, Michael developed machine learning models to forecast the next 24 hours of solar generation. As a next step, he planned to connect these forecasts to other applications through an application programming interface, which is a set of rules and tools that allow different software applications to communicate information with each other. Once integrated, this would enable GVEA to access reliable short-term solar forecasts directly within its operations.
ACEP summer intern RJ Michael stands by as Golden Valley Electric Association engineer Adam Saunders checks the transformer meters at GVEA鈥檚 Gold Hill substation.
As distributed solar generation in GVEA鈥檚 system grows, so will the importance of accurate forecasting. Michael鈥檚 work provides energy forecasters at GVEA with insight into the amount of solar energy generated from distributed solar sources within the GVEA service area, which covers Interior Alaska, including Fairbanks, North Pole and Delta Junction. Solar energy helps meet summer demand while providing a clean energy source.
Michael, a Gwich鈥檌n Athabascan who grew up in Bethel, Alaska, a small town in Southwest Alaska off the road system, is a senior at the University of Alaska Anchorage studying computer science and mathematics.
Michael appreciates the opportunity to explore data science within computer science in a way that directly benefits Alaska communities.
He worked as a summer intern with ACEP last year, designing and organizing a cybersecurity awareness training series. This year鈥檚 internship added another dimension to his interests in computer science.
ACEP summer interns tour Golden Valley Electric Association's battery energy storage system.
鈥淭hrough research and mentorship, I learned about solar energy systems, forecasting methods and what it鈥檚 like to work at a utility in Alaska,鈥 he said.
After graduating this winter, Michael plans to pursue graduate studies in artificial intelligence or data science.
This internship was funded by the program, an initiative supported by the Office of Naval Research, through the ACEP Undergraduate Summer Internship program. View the on ACEP鈥檚 YouTube channel. For more information on this project, contact Phylicia Cicilio at pcicilio@alaska.edu.