电网技术2025，Vol.49Issue(7)：2649-2658,中插1,11.DOI:10.13335/j.1000-3673.pst.2024.0767
面向电力系统暂态稳定性评估的深度学习模型智能增强方法
Model Enhancement for Deep Learning Based Transient Stability Assessment Models
摘要
Abstract
At present,deep learning is increasingly applied to power system transient stability assessment,but the accuracy of deep networks is still insufficient after the initial training.Therefore,this paper studies the intelligent enhancement method of the deep learning model for transient stability evaluation of power systems to improve the accuracy and reliability of the deep stability judgment model.Firstly,the problem description and overall framework of the deep learning model in transient stability evaluation of power systems are briefly introduced,the integration method of model intelligence enhancement in the framework is analyzed,and the significance of improving the accuracy of the evaluation model is analyzed.Then,the principles and application methods of retraining,error-free positioning fine-tuning,and error-free positioning fine-tuning are introduced.The error-free positioning fine-tuning takes the FAMR algorithm as an example.The algorithm first clusters the right and wrong samples and then fixes the original model based on the clustering results.Taking the Arachne algorithm as an example,the algorithm first locates the weight parameters of the fault according to the forward influence and loss gradient.Then it uses differential evolution to optimize these parameters.Then,an example is designed on the IEEE-39 node system for testing and verification,and the principle and test results of the algorithm are combined to compare and summarize the three methods.关键词
暂态稳定性评估/深度学习/模型修复/智能增强/参数调整Key words
transient stability assessment/deep learning/model repair/intelligent enhancement/parameter fine-tuning分类
信息技术与安全科学引用本文复制引用
郑乐,刘思远,周小添,徐式蕴,李宗翰,李庚银..面向电力系统暂态稳定性评估的深度学习模型智能增强方法[J].电网技术,2025,49(7):2649-2658,中插1,11.基金项目
国家重点研发计划项目:"响应驱动的大电网稳定性智能增强分析与控制技术"(2021YFB2400800).Project Supported by the National Key Research & Development Program of China:"Response-driven intelligent enhanced analysis and control for bulk power system stability"(2021YFB2400800). （2021YFB2400800）
