Session Chairs: Tianqiao Zhao, UTA and Jonghwan Kwon, ANL

This session examines how agentic AI can support grid planning, operations, and engineering decision-making through tool use, workflow orchestration, and validation-in-the-loop. Speakers will review emerging applications such as scenario analysis, contingency screening, study automation, operator support, and coordinated use of simulation, optimization, and data tools. Topics include human oversight, provenance, reliability, safety, evaluation, bounded autonomy, and pathways from research prototypes to utility deployment.

Role of AI in climate adaption of power systems

Speaker: Jin Zhao – Trinity College Dublin

The presentation explores the role of artificial intelligence (AI), represented machine learning techniques, in enhancing the resilience of power systems under extreme weather conditions. Power systems with high shares of non-dispatchable renewables, such as wind and solar, are increasingly vulnerable to storms and sequential extreme events that can trigger cascading failures and large-scale outages. Using high-renewable, stand-alone systems as motivating examples, the talk highlights challenges related to low inertia, weather exposure, and renewable variability. The presentation introduces AI-driven decision-making approaches based on DL and DRL to support fast and adaptive operational responses. Overall, the talk demonstrates how machine learning can serve as a powerful tool for resilient, climate-adaptive power system operation.

Auditable AI for Physical Grid Resilience: Multi-Agent Automated Reporting

Speaker: Yangmin Ding – NEC Labs America, Inc

The convergence of power grids and their underlying infrastructure requires comprehensive physical-layer observability across the entire network. This presentation introduces an auditable, schema-constrained agentic AI architecture designed for broad, live physical situational awareness in resilient grid operations. Shifting from passive offline analysis to active real-time processing, the framework continuously ingests live physical telemetry—such as acoustic signals and visual camera streams—over TCP. A major bottleneck in real-time continuous monitoring is background clutter generating false alarms, which heavily degrades event localization accuracy. Using physical substation security as a case study, we demonstrate how our framework employs deterministic AI agents to cross-validate acoustic anomalies (e.g., localized impacts or real-time gunshot events) against concurrent visual data. Once an event is securely localized and triaged, an efficient, structured-output language model automatically generates actionable operational reports for control room operators. By prioritizing multi-agent orchestration and multi-modal sensor fusion over legacy hardware diagnostics, this approach minimizes alert fatigue and provides a reasoning engine for safe, autonomous grid operations.

Agentic AI For Grid Applications

Speaker: Adrian Kelly – EPRI Europe

The presentation will give a summary of the applications for agentic AI in grid planning and operations.

Agentic Systems for the Power Grid: Orchestration, Provenance, and Trust

Speaker: Adrian Maldonado – Argonne National Laboratory

As power systems become more complex, analytical workflows increasingly need to combine heterogeneous data, physics-based simulations, optimization, machine learning, and human expertise. Agentic AI offers a promising way to coordinate these capabilities by acting as an orchestration layer that can interpret goals, invoke specialized tools, manage multi-step workflows, and synthesize results. For power-grid applications, however, effective agents must do more than reason and call tools: they must also track provenance, manage intermediate artifacts, preserve reproducibility, and determine when previous results remain valid or must be recomputed. This talk explores these challenges and opportunities, using GridMind as one example of a broader class of trustworthy agentic systems for power-grid analysis and decision support.