A new artificial intelligence system developed by researchers at the FAMU-FSU College of Engineering and Florida State University’s Center for Advanced Power Systems could give electric-grid operators a clearer view of what is coming next—and help reduce the cost of keeping the lights on. Called GridFusionX, the model forecasts electricity demand and renewable-energy generation while also estimating how confident it is in those predictions. That combination could become increasingly important as solar and wind power expand across modern grids, where uncertainty can quickly translate into higher operating costs or greater reliability risks.
Electric utilities must constantly balance electricity production with consumption. If demand rises unexpectedly or renewable generation falls because of changing weather, operators need to activate reserve power, often from more expensive or less efficient sources. Keeping excessive reserves available provides a safety margin but increases costs. Holding too little reserve, however, can leave the system vulnerable to instability or outages. GridFusionX is designed to improve that balance by producing more precise forecasts and identifying the level of uncertainty surrounding them.
The system’s central innovation is its ability to combine multiple forms of information, a capability known as multi-modality. Instead of relying solely on historical electricity use, the model can analyze past demand, renewable-energy output, energy-market prices and other relevant signals at the same time. These data streams provide different perspectives on the grid’s behavior. A period of high demand, for example, may be influenced not only by past consumption patterns but also by weather, market conditions and renewable resources available in neighboring areas.
To understand those connections, GridFusionX uses a graph neural network. In this type of artificial-intelligence architecture, a complex system is represented as a collection of connected nodes and relationships. For an electric grid, nodes can represent regional power systems or locations, while connections represent electrical links and other interactions. Electricity flows across physical transmission lines, but regions can also influence one another through shared weather conditions, market prices and changes in demand. The network structure enables the model to incorporate those relationships into its forecasts.
“Every factor, from different energy sources to shifting demands in cities and changing prices, fits together,” said Tuy Nguyen, an assistant professor in the Department of Electrical and Computer Engineering and a co-author of the study. “What happens in one area can affect neighboring areas because they share power lines, weather patterns and energy markets.” By continuously analyzing these linked factors, the model is intended to capture patterns that conventional forecasting methods may overlook when they examine each region independently.
GridFusionX also addresses a major limitation of many forecasting tools: they often provide a single predicted value without clearly indicating how uncertain that estimate may be. The FSU-developed system produces probabilistic forecasts, including confidence intervals that show a range of likely outcomes. Such information can help operators distinguish between a stable prediction and one that carries a higher risk of error. The result is not simply a forecast of how much electricity may be needed, but a more detailed picture of possible future conditions.
In tests involving ten European regions, the model improved forecasting accuracy by as much as 56 percent and reduced reserve costs by up to 66 percent while maintaining reliable service, according to the researchers. Those results suggest that network-aware forecasting could have a direct effect on how utilities schedule generation and reserves. More accurate predictions may allow operators to avoid committing unnecessary backup capacity while still preparing for sudden demand spikes or steep declines in renewable output.
The researchers say the potential impact extends beyond utility control rooms. More efficient forecasting could eventually support lower operating costs throughout the energy system, with possible benefits for consumers. Olugbenga Moses Anubi, an associate professor and study co-author, said that more precise predictions could help align electricity bills more closely with actual consumption rather than relying on overly cautious estimates. The broader objective, he explained, is to reduce uncertainty so that energy systems can make smarter and more adaptive decisions.
The project also reflects a collaborative training model at the FAMU-FSU College of Engineering. Doctoral student Quoc Bao Phan led the study, working with Nguyen, Anubi, associate professor Ravikumar Gelli and assistant professor Abdulrahman Takiddin. Faculty members are incorporating related concepts into courses focused on cybersecurity for electric grids and artificial intelligence for power systems. The researchers describe the project as both an engineering advance and an opportunity for students to work across several areas of energy research.
GridFusionX was developed by treating the power grid as a connected, dynamic system rather than a collection of isolated locations. As renewable energy grows and electricity networks become more dependent on rapidly changing data, that perspective could prove essential. The study, published in IEEE Transactions on Network Science and Engineering, presents probabilistic, multi-regional forecasting as a potential pathway toward grids that are more reliable, more economical and better prepared for the uncertainty of a renewable-powered future.
Subject of Research: Artificial intelligence for network-aware probabilistic forecasting in multi-regional electric power systems
Article Title: GridFusionX: Network-Aware Probabilistic Forecasting for Multi-Regional Power Systems
News Publication Date: 27-Mar-2026
Web References: FAMU-FSU College of Engineering — https://eng.famu.fsu.edu/ ; Florida State University Center for Advanced Power Systems — https://www.caps.fsu.edu/ ; IEEE article — https://ieeexplore.ieee.org/abstract/document/11457347
References: IEEE Transactions on Network Science and Engineering; DOI: 10.1109/TNSE.2026.3678493
Image Credits: Scott Holstein/FAMU-FSU College of Engineering
Keywords
Artificial intelligence, GridFusionX, electric grids, renewable energy, probabilistic forecasting, graph neural networks, power systems, energy demand, grid reliability, reserve costs
