Researchers use LLMs to revolutionize power grid modeling
LLM-powered agents cut power grid modeling time by 25.7% with 0.19% error
A new LLM-based multi-agent collaborative framework discovers differential-algebraic dynamic models in power systems. It integrates exploratory agents, candidate model memories, parameter fitting, and a coordinator to jointly recover state dynamics, algebraic constraints, and key intermediate variables under weak prior information. Case studies on synchronous generators and grid-forming inverters show it beats single-agent LLM discovery and conventional symbolic regression in accuracy, generalization, search efficiency, and noise robustness. In the generator case, out-of-distribution MAPE reaches 0.19%; in the inverter case, discovery time drops 25.7% versus a single-agent LLM baseline.
- Developed by Xi'an Jiaotong University researchers using LLMs as multi-agent systems for power system modeling
- Achieved 0.19% OOD MAPE on synchronous generators and 25.7% faster discovery on grid-forming inverters
- Combines differential equation discovery with algebraic closure recovery using coordinated agent teams
Why It Matters
Enables faster, more accurate power grid modeling for modern renewable-heavy systems with black-box components