基于特征分离型神经网络的电力系统暂态稳定性评估与优化方法  

Assessment and Optimization of Power System Transient Stability Based on Feature-separated Neural Networks

作　　者：周子涵 卜广全 马士聪 王国政 邵德军 徐友平 党杰 ZHOU Zihan;BU Guangquan;MA Shicong;WANG Guozheng;SHAO Dejun;XU Youping;DANG Jie(China Electric Power Research Institute,Haidian District,Beijing 100192,China;Central China Branch of State Grid Corporation of China,Wuhan 430077,Hubei Province,China;Department of Electrical Engineering,Tsinghua University,Haidian District,Beijing 100084,China)

机构地区：[1]中国电力科学研究院有限公司,北京市海淀区100192 [2]国家电网公司华中分部,湖北省武汉市430077 [3]清华大学电机工程与应用电子技术系,北京市海淀区100084

出　　处：《电网技术》2021年第9期3658-3666,共9页Power System Technology

基　　金：国家电网公司华中分部科技项目:基于人工智能的特高压交直流混联电网暂态稳定评估技术研究。

摘　　要：该文基于神经网络(artificial neural networks,ANN),提出一种特征分离型暂态稳定智能评估模型,并针对迁移学习样本生成过程提出样本生成方法。根据不同电气特征对电力系统暂态稳定性的关联程度不同,利用神经网络构建了特征分离型暂态稳定智能评估模型;针对潮流变化或拓扑变化的影响,引入迁移学习方法对评估模型进行再训练,提出关键故障位置原则和关键故障持续时间原则指导迁移学习样本生成过程;进而提出通过调节机组出力提升暂态稳定性的优化算法。算例结果验证了分离特征对评估性能提升的有效性;采用迁移学习样本生成原则在减少样本生成数量、提升评估性能方面效果显著;所提优化模型能够有效提升电力系统暂态稳定性,为电力系统暂态稳定性智能评估与优化提供了新的思路。Based on artificial neural networks(ANN), a feature-separated intelligent transient stability assessment model and a sample generation method for transfer learning are proposed in this paper. The intelligent model is first designed according to the different degrees of correlation between different electrical characteristics and power system transient stability. In view of the influence operating mode or topological structure change, the pre-trained model is then retrained with transfer learning, and the key fault location principle and the key fault duration principle are proposed to guide the generation of transfer learning samples. And further an optimization algorithm is proposed to improve the transient stability by adjusting generator units’ output. The test result shows that the separation of features is effective for improving the evaluation performance. The transfer learning samples generation principle has a significant effect in reducing the number of new samples and improving the evaluation performance. The proposed optimization model can effectively improve the power system transient stability, and provides a new idea for the power system transient stability intelligent assessment and optimization.

关 键 词：电力系统 暂态稳定 机器学习 迁移学习 神经网络 

分 类 号：TM721[电气工程—电力系统及自动化]