Analyzing and managing the electric power grid are complex tasks. They involve many different types of simulations, from reliability checks to scheduling and contingency planning. Each simulation requires specialized tools, programming skills and deep technical knowledge. These disconnected workflows can slow down decision-making when fast, clear actions are most needed.
Recent advances in artificial intelligence (AI), especially in systems that can act independently (called “agentic AI”), offer a way to simplify this complexity. These AI agents can orchestrate multiple tasks at once, understand and reason across different types of analysis, and support decision-making — all through natural conversations in plain language. They also use proven tools and methods to ensure that results are accurate and reliable.
To harness this potential, researchers at the U.S. Department of Energy’s (DOE) Argonne National Laboratory have created GridMind, an agentic AI system that brings these capabilities together to serve as a reasoning co-pilot for power system operators — a step toward the control room of the future.
Although AI and natural language tools have been used in technical fields before, applying them in highly technical, cross-disciplinary engineering tasks — like those in power grid operations — is still new. GridMind integrates them into a coherent reasoning engine for power system operations. It does not simply display numbers but interprets them, connects them across tasks and explains what they mean.
“GridMind is designed to be a reasoning partner for grid operators,” said Kibaek Kim, a computational mathematician. “It keeps the analysis rigorous but allows operators to interact with it using natural language — essentially turning technical analysis into conversational, explainable support for their decisions.”
At the heart of GridMind is a multi-agent system — a group of AI agents each specializing in a different task. For example:
- One agent handles power system scheduling, making sure power is produced and distributed efficiently and safely.
- Another agent pulls in weather forecasts to simulate hurricanes and then checks the power system to spot where equipment could fail and how outages might spread.
These agents are coordinated by large language models (LLMs), which understand the task, analyze the situation, reason across different analyses and suggest explainable strategies.
Researchers ran experiments on various standard power grid models, using multiple state-of-the-art LLMs. The system was evaluated for accuracy, speed and reliability. The results were promising — GridMind consistently produced correct results and clear reasoning, even across different AI models.
View source version on businesswire.com: https://www.businesswire.com/news/home/20260326916280/en/
