STOCKHOLM — An artificial intelligence system designed to dynamically balance electricity supply and demand across municipal power grids reduced measurable energy waste by an average of 34 percent in a two-year pilot program spanning four mid-sized cities in northern Europe, according to findings released Friday by the Nordlund Energy Research Consortium.
The system, called GridMind, uses a combination of real-time sensor feeds, numerical weather prediction models, and machine learning algorithms trained on historical consumption patterns to forecast demand fluctuations at 15-minute intervals. It then automatically adjusts the routing and output levels of generation sources — including solar photovoltaic arrays, wind turbines, battery storage banks, and natural gas peaker plants — to minimize overproduction and reduce the frequency of emergency load-balancing actions that burn energy without delivering it to end users.
“Traditional grid management operates on fixed dispatch schedules and reactive adjustments triggered after imbalances are already occurring,” said consortium director Ingrid Halvorsen. “GridMind is constantly running probabilistic scenarios and repositioning generation assets ahead of demand shifts rather than after them. That predictive posture is where the majority of the efficiency gain originates.”
The pilot ran across Vellemark, Osterhavn, Brennstad, and Korsby — four cities with populations ranging from 140,000 to 310,000 — selected because they each have a substantial mix of renewable generation sources alongside conventional backup capacity, making real-time balancing a persistent operational challenge. Across the four sites, the GridMind system managed approximately 4.7 gigawatt-hours of electricity per day during the test period.
Over the 24-month pilot, measured grid losses — defined as electricity generated but not successfully delivered to end consumers due to imbalance, curtailment, or transmission inefficiency — fell from a baseline average of 9.2 percent to 6.1 percent. Carbon dioxide emissions attributable to wasted generation dropped by an estimated 218,000 metric tons cumulatively across the four sites over the two-year period, the consortium said.
Grid operators in all four participating cities reported that GridMind required human override on less than 2 percent of its automated generation dispatch decisions, a figure that the consortium said compared favorably to published industry benchmarks for autonomous industrial control systems operating in critical infrastructure settings. No significant service disruptions were attributed to system errors during the pilot.
Researchers noted that the cities selected for the initial deployment have relatively modern grid sensor networks and communication infrastructure, which simplified the integration of the AI platform and provided the clean, high-frequency data streams the system needs to perform optimally. Applying GridMind to older grids in lower-income regions — where sensor coverage is often incomplete, metering systems may be analog, and data transmission is intermittent — could significantly complicate deployment and reduce performance gains.
“The results in these four cities are strong, and the methodology is credible,” said Dr. Pieter Vanhoeck of the Brussels Center for Energy Systems, who reviewed the consortium’s study design but was not part of the research team. “The harder and more important question is what happens when you bring this into grids that were not built with digital integration in mind. That is where the real proof of concept needs to happen next.”
The Nordlund Consortium said it is in active discussions with energy regulators in five additional countries — including two in Southeast Asia and one in West Africa — about expanded deployments targeting aging grid infrastructure. A commercial licensing arrangement has been finalized with Praxis Grid Solutions, a privately held infrastructure technology firm headquartered in Amsterdam, to manage installations outside the consortium’s direct research program.
Energy economists said broad adoption of AI-driven grid management tools could materially reduce the incremental cost of integrating variable renewable generation at high penetration levels — one of the central logistical and financial obstacles facing ambitious decarbonization targets in Europe, North America, and rapidly industrializing economies elsewhere.