Researchers have developed an AI forecasting approach that improves electricity demand predictions, helping grid operators reduce reserve costs, strengthen reliability and minimise blackout risks.
Researchers at Florida State University and the FAMU–FSU College of Engineering have developed GridFusionX, an artificial intelligence-based forecasting system designed to improve electricity demand predictions and help reduce blackout risks. The research addresses the growing complexity of modern power grids, where increasing renewable energy generation makes balancing electricity supply and demand more challenging.
The AI model treats the electricity grid as a connected network rather than analysing regions independently. By combining historical electricity demand, renewable power generation, energy market prices and other relevant datasets, the system generates more accurate forecasts alongside confidence intervals. This enables grid operators to make better-informed decisions while reducing unnecessary reserve power and associated operating costs.
Unlike conventional forecasting methods, GridFusionX uses a graph neural network to capture relationships between interconnected regions. The multimodal approach incorporates multiple sources of information, including weather conditions, traffic patterns and electricity usage, allowing the model to respond more effectively to sudden fluctuations in demand or renewable energy output. Researchers said the system provides both spatially connected predictions and measures of uncertainty, helping operators plan ahead with greater confidence.
In real-world evaluations across 10 European regions, the model improved forecasting accuracy by up to 56% while reducing reserve costs by as much as 66%, according to the study. The researchers noted that the improved precision could help utilities balance electricity generation more efficiently, maintain reliable service and potentially reduce costs for consumers by matching energy supply more closely with actual demand.
The team also highlighted the educational impact of the project, with doctoral researchers contributing alongside faculty members from multiple disciplines. The findings, published in IEEE Transactions on Network Science and Engineering, demonstrate how advanced AI forecasting could strengthen grid resilience as renewable energy adoption continues to increase, while supporting more reliable and cost-effective power system operations.
