Session Chair: Jonghwan Kwon, ANL
This session examines the application of artificial intelligence to power system planning, with an emphasis on improving the scalability, efficiency, and fidelity of planning studies. Speakers will present advances in AI-enabled transmission expansion planning, resource adequacy, reliability assessment, renewable integration, and uncertainty quantification. Topics include foundation models, surrogate modeling, agentic AI, optimization, physics-informed machine learning, and hybrid AI-physics approaches for large-scale planning and decision support.
Graph-Aware Quantum Feature Extraction for AC Power Flow Prediction
Speaker: Giorgio Cortiana – E.ON Digital Technology GmbH
We present a graph-aware quantum feature extraction framework that encodes power-grid topology and electrical interactions into quantum circuits to generate representations for downstream AI models. We evaluate these representations for predicting transmission-line currents from nodal power injections. Quantum-derived features are integrated into two complementary learning architectures: a fully connected neural network (MLP) without explicit topology awareness and GENCO, a domain-specific, physics-informed, grid-aware model. Across IEEE 39-, 57-, and 118-bus benchmark systems, the inclusion of quantum-derived features improves prediction accuracy and data efficiency relative to the corresponding classical model configurations, with error reductions of up to one order of magnitude for the MLP and up to 50% for GENCO. The results indicate that graph-aware quantum representations can provide complementary information to classical features, especially relevant for topology-unaware model architectures. Experiments on IBM quantum processors reproduce the main trends observed in simulations, demonstrating the feasibility of generating such features on quantum hardware. Overall, the results provide evidence for quantum feature extraction as a hybrid quantum-classical approach in which quantum processors serve as representation engines for graph-structured learning tasks.
Agentic Grid Planning on AWS: Accelerating Interconnection Studies using Agentic AI
Speaker: Osama Ansari – Amazon Web Services (AWS)
Generation and load interconnection queues continue to grow, and utilities face rising study volumes with constrained engineering teams and legacy workflows. Agentic AI offers a way to scale this work, but applying it to power system studies raises problems that general-purpose AI applications do not face. This talk shares an industry perspective from building Agentic Grid Planning on AWS, a program that couples AI agents with existing grid simulation tools to help planners run interconnection studies faster and more thoroughly. We discuss how agents reason about study criteria, orchestrate power flow and contingency analysis, diagnose violations, and propose mitigation alternatives, while planners stay in the lead at every step. We also cover the practical challenges around validating agent outputs against physics-based results, working with commercial tools and data, and earning engineers’ trust.
