IBM’s quantum GENCO neural solver unifies three grid tasks

IBM Research and its partners are releasing GENCO, a neural solver designed to unify three core grid-analysis tasks, alongside the GridFM Development Framework. This development addresses a critical challenge, as approximately 2.6 terawatts of mostly clean generation and storage await approval to connect to the United States power grid, nearly double the existing capacity. “Operators often face a trade-off between computational speed and model accuracy,” said Etienne Vos, a manager at IBM Research, as GENCO aims to deliver speedups orders of magnitude greater than traditional methods while recovering key grid variables.

GENCO Unifies Power Flow, OPF, and State Estimation

Approximately 2. This consolidation addresses a critical challenge for grid operators facing increasingly complex demands from electrification and renewable energy integration. GENCO differs from conventional approaches by delivering substantial speed improvements without sacrificing accuracy; for grids up to 10,000 buses, it is up to 30 times faster than traditional Newton-Raphson AC solvers when calculating power flow, achieving comparable residual levels to simpler DC approximations. Beyond speed, the unified architecture of GENCO streamlines complex assessments that often require chaining together multiple specialized solvers, reducing development costs and accelerating analysis timelines.

“For years, teams built a separate model for every grid task. GENCO shows you don’t have to,” said paper co-author Thomas Brunschwiler, principal research scientist and manager at IBM Research.

The solver’s design leverages a graph representation of the power grid, utilizing a heterogeneous graph transformer as its core, coupled with corrective layers to ensure physical consistency and physics decoders to translate predictions into viable solutions for each task. This approach not only enhances computational efficiency but also improves the quality of results, recovering key grid variables often omitted by DC approximations.

In optimal power flow calculations, GENCO is up to 85 times faster than interior-point solvers at an optimality gap of 0.3% or less, while also demonstrating superior accuracy in state estimation, particularly under conditions of sparse measurements and noisy signals. Validation of GENCO’s performance extends to real-world operational data, with successful testing conducted on a 1,200-bus transmission network operated by Hydro-Québec, using a full year of supervisory control and data acquisition (or SCADA) data.

The development of GENCO and the GridFM framework is rooted in a collaborative, open-source approach, involving a community of organizations and members from industry, academia, and government. The team anticipates future advancements, including integration with quantum computing to tackle complex combinatorial decisions, while GENCO rapidly solves continuous optimization problems.

“We see a promising future in pairing GENCO with quantum computing,” Hamann added. “GridFM with GENCO matters because it brings together the grid community to build foundation AI models that can evaluate those futures at unprecedented speed and scale — giving us the foresight to operate the grid more securely today and plan it more reliably for decades to come,” Hamann concluded.

GridFM Framework Enables Reproducible Neural Solver Benchmarking

The demand for rigorous, reproducible benchmarks in neural network-based power grid analysis is being met with the release of the GridFM Development Framework alongside IBM Research’s GENCO neural solver. This framework addresses a critical need for standardized testing as the power grid undergoes rapid transformation, facing increasing pressure from electrification and the influx of renewable energy sources. Currently, roughly 2.6 terawatts of mostly clean generation and storage, nearly double the entire capacity installed across the United States today, sits in interconnection queues, waiting several years to connect.

GridFM is not simply a testing environment; it’s designed to facilitate fair, rapid, and reproducible benchmarking of neural grid solvers. The framework includes gridfm-graphkit, used for training and evaluating the solvers, and gridfm-datakit, which generates the synthetic data necessary for robust testing.

This focus on reproducibility is particularly vital given the increasing complexity of grid modeling and the potential for subtle errors to propagate through simulations. The collaborative nature of the GridFM project distinguishes it from many other AI initiatives in the energy sector. The GridFM community involves a broad range of participants spanning industry, academia, and government. This broad participation ensures that the framework reflects the diverse needs and priorities of grid operators and planners.

“This work is distinguished by the breadth of the community behind it: Many dozens of organizations representing the full spectrum of grid operations and power systems, collaborating closely to develop foundational AI capabilities for the power sector,” explained Hamann. The project’s open-source approach, hosted by LF Energy, encourages contributions and fosters innovation, moving away from proprietary models with limited transparency.

The development of GENCO and GridFM also responds to the escalating demands placed on the grid by emerging technologies. Global data-center electricity use is projected to more than double by 2030, creating a rapidly increasing and time-sensitive load that traditional analysis methods may struggle to accommodate.

DOE Study Highlights Need for Faster Grid Analysis

Hydro-Québec’s transmission network, a 1,200-bus system, served as the proving ground for GENCO, a new neural solver developed by IBM Research and its partners and released through the OpenGridFM project at Linux Foundation Energy. Validating the solver’s steady-state capabilities on a real network represents a critical step toward wider industry adoption, according to researchers involved in the project.

The release addresses a growing need for faster and more accurate grid analysis as the power system undergoes a dramatic transformation driven by renewable energy and increasing demand. The Department of Energy’s National Transmission Planning Study highlighted a significant bottleneck in current grid planning methods; historically, planners have relied on a limited number of system snapshots for analysis.

With the influx of variable renewable generation, evolving demand patterns, and the need to interconnect roughly 2.6 terawatts of new clean energy and storage, nearly double existing U.S. capacity, planners now require the ability to analyze thousands, or even millions, of power-flow cases. This escalating computational burden is pushing traditional numerical methods to their limits, often forcing operators to compromise between speed and accuracy.

This consolidation streamlines complex assessments that previously required separate solvers and pipelines, reducing both development effort and computational time. It’s also up to 85 times faster than interior-point solvers at finding the least-cost generator settings, at an optimality gap of 0.3% or less.

GENCO Achieves 30x Speedup Over AC Solvers

With approximately 2. This design allows it to process grids of up to 10,000 buses, delivering substantial performance gains over traditional methods. This improvement is crucial as detailed AC models are often too slow for comprehensive grid analysis, forcing operators to rely on less accurate DC models. The unified design of GENCO also reduces development costs and time. Because the same foundational architecture and hyperparameters are used for all three grid analysis tasks, the extensive tuning typically required for each individual solver is performed only once.

This efficiency extends to the development framework, GridFM, which facilitates building and benchmarking neural grid solvers with fairness, speed, and reproducibility. The team demonstrated GENCO’s robustness by testing it under extreme conditions, including scenarios with up to 20 component failures, where it outperformed DC solvers with a reduction in median residuals of about 3 times.

The team also demonstrated the model’s adaptability by pre-training it on 100 decomposed grids, which doubled data efficiency. “Today, grid operators and planners are forced to make critical decisions using only a small fraction of the possible futures the system may face,” he said.

Hydro-Québec Data Validates GENCO on Real Grid

The demand for increasingly detailed grid analysis is rapidly outpacing traditional computational methods, a challenge IBM Research and its partners sought to address with GENCO, a novel neural solver. This validation process wasn’t merely academic; access to operational grid data remains limited within the energy sector, making this demonstration a significant milestone for the industry. The Hydro-Québec dataset allowed researchers to move beyond theoretical performance and assess GENCO’s capabilities under realistic conditions. GENCO’s ability to deliver substantial speedups without sacrificing accuracy positions it as a potentially crucial tool for managing this complex landscape.

The team’s approach involved rigorous testing, including scenarios operating outside nominal conditions and simulating high-order contingencies, up to 20 simultaneous component failures, where GENCO consistently outperformed traditional direct current solvers, exhibiting median residuals about 3 times lower. Beyond speed and accuracy, the unified design of GENCO offers significant developmental advantages. “GENCO shows you don’t have to.” This consolidation streamlines development, reducing the time and resources required to build and maintain complex grid analysis tools.

“That shouldn’t be the price of doing good science,” Hamann said, emphasizing the importance of open collaboration and shared resources in advancing grid technology. The successful validation on Hydro-Québec’s network represents a step toward realizing that vision and deploying AI-powered solutions for a more resilient and efficient power grid.