Session Chairs: Francois Miralles – HQ and Jochen Cremer – TU Delft

This session focuses on extending Grid Foundation Models beyond static network representations toward the modeling and understanding of power-system dynamics. Speakers will examine how AI and foundation models can learn dynamic behaviors from simulations, synthetic and operational data, enabling applications such as renewable integration, dynamic security assessment, dynamic state estimation, stability monitoring, accelerated simulations, and control support. Topics include time-series foundation models, physics-informed neural networks, hybrid AI-physics approaches, uncertainty-aware forecasting, and the integration of dynamic models into next-generation AI for power systems.

Revisiting data-driven dynamic security assessment with a tabular foundation model

Speaker: Arowolo Olayiwola – TU Delft

Data-driven pre-fault dynamic security assessment (DSA) rapidly evaluates the dynamic risk of credible contingencies on a power system using machine learning. Existing approaches face two limitations. First, they require a large labelled database for training, with a separate model trained, tuned, and maintained for each contingency in a potentially long list of credible contingencies. Second, the trained models generalize poorly to unseen contingencies. This work addresses these limitations by using a tabular foundation model (TFM) that assesses stability through in-context learning, requiring no retraining or hyperparameter optimization. A single TFM can assess many contingencies at once, removing the need for one model per classifier.
We also characterize when the use of electrical distance coordinates (EDC) as continuous features enables generalization of TFM to unseen contingencies and when they do not, demonstrating how a few labelled samples can reliably improve generalization.
Through comprehensive case studies on the IEEE 68-bus system, we show that a single TFM attains an average Macro F1 score of about 90% with only 120 labelled samples per contingency, roughly two orders of magnitude fewer than conventionally assumed, without any model retraining or hyperparameter tuning. For new/unseen contingencies, we show that using just 10 labelled samples of the new contingency with EDC encoding matches the best achievable transfer learning oracle model, which requires fully labelled data and is not deployable in practice.
Overall, this initial study paves the way for developing and deploying foundation models for power system operations, with possible applications across multiple operational tasks.

Neuro-Symbolic AI for Automatic Control Design: Pairing Control Graphs with LLMs

Speaker: Hamweemba Grainger – INESC TEC

The increasing penetration of renewable energy sources has heightened the importance of grid-forming (GFM) converters, which are expected to play a key role in supporting stability in converter-dominated power systems. At the same time, the growing complexity of modern grids makes controller design more challenging, particularly when both the controller structure and its parameters must be determined. Conventional approaches still rely heavily on predefined architectures and engineering experience, with optimization often carried out only after the control structure has been chosen. This work presents a neuro-symbolic framework for automated controller design in power systems by combining control graphs with large language models (LLMs). The control system is represented as an interpretable graph-based model that preserves the structure of conventional block diagrams while allowing the topology to evolve. The design follows a bi-level optimization scheme in which simulated annealing (SA) handles structural evolution and particle swarm optimization (PSO) tunes the numerical parameters of each candidate structure. The LLM is embedded in the SA search and uses control-domain knowledge, together with information from previously evaluated solutions and their dynamic performance, to suggest structural changes. This makes the search less dependent on purely random mutations while retaining the SA acceptance mechanism. Candidate controllers are assessed through stability analysis of the system state matrix and nonlinear time-domain simulations. The resulting framework provides a systematic and interpretable approach to automated controller design in power systems, with GFM control used to demonstrate the methodology.

PowerAgentBench-Dyn: A Benchmark for Agentic AI in Power System Dynamic Studies

Speaker: Costas Mylonas – UBITECH

Large Language Model (LLM)-based agents are increasingly being used to automate multi-step engineering work flows by interacting with software tools, interpreting intermediate results, and autonomously planning subsequent actions. Power system dynamic studies represent a particularly promising yet largely unexplored application domain for these agents. Unlike static computational tasks, dynamic studies often require more time on model parameter calibration, engineering judgment, and decision making under constrained action spaces. This paper introduces PowerAgentBench-Dyn, a benchmark designed to evaluate Agentic AI systems on power system dynamic-analysis tasks. The benchmark targets problems that cannot be reduced to a single optimization or coding task, but instead require a type of reasoning, tool usage, and iterative experimentation routinely performed by experienced power system engineers. The proposed framework includes two initial benchmark tasks. The first, the Dynamic Model Quality Review Benchmark, evaluates agents’ ability to validate and diagnose dynamic models based on model-quality compliance criteria specified by system operators. The second, the Dynamic Security Risk Screening Benchmark, assesses agents’ capability to leverage semantic memory and a limited simulation budget to identify, rank, and analyze the most critical short-circuit contingencies from an unseen fault dataset, as well as propose and evaluate possible mitigation measures. For each task, we define the simulation environment, observation and action spaces, and evaluation metrics. The benchmark is reproducible in a metric-based sense: released cases and simulator settings define a deterministic evaluator, while stochastic agent behavior is assessed over repeated runs using success rates and other metrics. The benchmark supports the development of future Agentic AI for power system operation and planning.

Enabling Time-Domain Grid Foundation Models with PowSyBl and Dynawo

Speaker: Nicolas Lair – Artelys

The development of foundation models for power systems requires access to large-scale, diverse, and physically consistent training datasets. This presentation introduces the integration of Dynawo and PowSyBl into the Open GridFM DataKit, extending its data generation capabilities from static power system analysis to dynamic simulation and time-domain datasets.
Leveraging industrial-grade power system simulation tools provides a robust basis for generating high-quality synthetic data representative of realistic grid behavior. Such datasets can support the training of more reliable models and help reduce the sim-to-real gap when transferring models towards real-world applications.
Beyond introducing temporal dynamics, dynamic simulation also considerably expands the possibilities for data augmentation. Grid disturbances and dynamic events, together with their characteristics and operating conditions, can be systematically generated and randomized to create richer and more diverse training distributions.
This integration therefore represents a key enabler for time-domain Grid Foundation Models, paving the way towards models capable of learning and representing complex power system dynamics.