Reinforcement Learning from Human Feedback in Power System Risk Point Identification

Abstract

Abstracts：With the expansion of power system scale and complex structure, the difficulty of risk identification increases, and the traditional identification methods have limitations. This paper focuses on the application of Reinforcement Learning from Human Feedback (RLHF) in power system risk point identification. RLHF integrates human feedback into reinforcement learning, which is trained through multiple strategies to generate samples and collect feedback, train reward models, and train reinforcement learning strategies, among other steps. It is used in a wide range of application scenarios in power system risk point identification, such as line tripping risk judgement, construction site safety hazard investigation, etc. RLHF can integrate multi-source data, improve the accuracy and efficiency of risk point identification, and play an important role in safeguarding the stability of the power system. However, RLHF applications face challenges such as data quality, model complexity and regulatory ethics. In the future, technological innovation, algorithm optimisation and integration with other technologies will promote its development in power system risk point identification, and help power system intelligent and efficient development.