Knowledge-GPT Guided Generalizable Reinforcement Learning for Intelligent Emergency Generator Tripping in Power...

Area of Science:

• Electrical Engineering
• Artificial Intelligence
• Power Systems Stability

Background:

• Emergency control is critical for maintaining transient stability in power systems following faults.
• Existing emergency control methods face limitations in adaptability and effectiveness.
• Intelligent control strategies are needed to address the complexities of modern power grids.

Purpose of the Study:

• To propose a knowledge-generative pretrained transformer (GPT)-guided generalizable reinforcement learning (RL) approach for intelligent emergency generator tripping.
• To enhance the efficiency and electrical consistency of deep reinforcement learning (DRL) training by incorporating electrical principles and expert knowledge.
• To improve the generalization capability of control strategies under topological changes in power systems.

Main Methods:

• Integration of general electrical principles for identifying disturbed generators and selecting control actions via dynamic probability.
• Utilization of a knowledge-GPT model to extract insights from an expert strategy knowledge base, refining the DRL reward structure.
• Incorporation of message passing neural networks (NNs) into the DRL architecture to simulate power flow dynamics and enhance generalization.

Main Results:

• The proposed GPT-guided RL approach demonstrated superior control effectiveness in simulations on the IEEE 39-bus system and the Northeast China power grid.
• The method showed enhanced adaptability compared to existing approaches, particularly under topological changes.
• Validation confirmed improved training efficiency and electrical consistency in the emergency control strategy.

Conclusions:

• The developed intelligent emergency generator tripping method offers a more robust solution for ensuring transient stability in complex power systems.
• The synergistic use of GPT, RL, and electrical principles provides a powerful framework for advanced power system control.
• This approach significantly advances the state-of-the-art in intelligent emergency control for power grids.