This blog post explains a new collaboration between the University of Connecticut and utility company Eversource to use artificial intelligence to improve how utilities prepare for power outages.
It summarizes how researchers are combining traditional physics-based weather models with machine learning to reduce errors in wind gust forecasts, with the goal of cutting unnecessary resource mobilization and improving outage readiness during extreme weather events in the Northeast.
Why better forecasts matter for utilities
Utilities make high-stakes operational decisions based on weather forecasts: when to pre-position crews, how many trucks to mobilize, and which distribution assets may need reinforcement.
Overpredicted wind gusts can lead to unnecessary expenses and inefficient use of personnel, while underprediction increases the risk to public safety and infrastructure.
The University of Connecticut–Eversource project focuses on the specific problem of wind gust prediction, which is notoriously difficult for conventional numerical weather prediction systems.
Improving these forecasts can directly reduce costs and improve resilience during hurricanes, nor’easters, and severe convective storms.
The UConn–Eversource collaboration and the hybrid approach
Researchers at UConn, led by Dr. Marina Astitha, and partners at Eversource are developing a hybrid model that blends physics-based numerical weather prediction with data-driven machine learning.
Traditional models solve complex physical equations but do not always correct for past biases.
The hybrid approach allows AI to learn from those biases and adjust forecasts accordingly.
Ph.D. candidate Israt Jahan has shown that AI can significantly reduce the tendency of traditional models to overpredict wind strength.
By correcting systematic errors, the hybrid system improves confidence in forecasts and helps utilities make more precise operational plans.
How the AI-enhanced system works and what it improves
The core idea is to keep the solid physical foundation of conventional models while adding an AI layer that learns from historical errors.
This results in forecasts that are both fast and more accurate, especially for difficult-to-predict variables like short-duration wind gusts.
Key advantages of this approach include improved reliability, speed, and practical utility for outage management teams.
The AI does not replace meteorology; instead, it augments it by identifying and correcting predictable biases.
Practical benefits for utilities and the public
Adopting AI-enhanced forecasts can yield measurable benefits for utilities and customers by aligning resources more tightly with actual risk.
Challenges and next steps
Although promising, hybrid AI-weather systems require careful validation and continual monitoring.
Machine learning models must be trained on representative historical data and tested across diverse storm types to avoid overfitting.
Close collaboration between meteorologists and utility operators is essential to translate improved forecasts into actionable procedures.
Conclusion
The collaboration between the University of Connecticut and Eversource illustrates how targeted AI can enhance weather forecasting for utility operations.
By reducing overprediction of wind gusts and improving forecast reliability, hybrid models offer a pragmatic route to lower costs and faster responses.
Here is the source article for this story: UConn researchers use AI in weather forecasting to predict power outages