文章摘要
高昆仑,杨帅,刘思言,等.基于一维卷积神经网络的电力系统暂态稳定评估[J].电力系统自动化,2019,43(12):18-26. DOI: 10.7500/AEPS20180911006.
GAO Kunlun,YANG Shuai,LIU Siyan, et al.Transient Stability Assessment for Power System Based on One-dimensional Convolutional Neural Network[J].Automation of Electric Power Systems,2019,43(12):18-26. DOI: 10.7500/AEPS20180911006.
基于一维卷积神经网络的电力系统暂态稳定评估
Transient Stability Assessment for Power System Based on One-dimensional Convolutional Neural Network
DOI:10.7500/AEPS20180911006
关键词: 电力系统  暂态稳定评估  一维卷积神经网络  深度学习  时间序列
KeyWords: power system  transient stability assessment  one-dimensional convolutional neural network  deep learning  time series
上网日期:2019-04-03
基金项目:国家电网公司科技项目(SGGR0000JSJS1800569)
作者单位E-mail
高昆仑 全球能源互联网研究院有限公司, 北京市 102209  
杨帅 华北电力大学电气与电子工程学院, 北京市 102206 monkey_d_s@163.com 
刘思言 全球能源互联网研究院有限公司, 北京市 102209  
李向伟 华北电力大学电气与电子工程学院, 北京市 102206  
摘要:
      系统遭遇暂态故障的过程是随时间发展的过程,基于传统机器学习的暂态稳定评估方法通常难以捕捉其时间维度信息,限制了评估性能的提高。针对该问题,提出了一种基于一维卷积神经网络(1D-CNN)的暂态稳定评估方法。该方法直接面向底层量测数据,凭借其特有的一维卷积和池化运算特性,能自动提取出暂态过程所蕴含的时序特征,从而达到对系统暂态稳定状态准确刻画的目的。设计了一种适用于暂态稳定评估的四卷积层1D-CNN模型,实现了端到端的“时序特征提取+暂态稳定性分类”,并通过调整模型关键参数以提高失稳样本查全率,增强了评估结果的可靠性。新英格兰10机39节点测试系统的仿真实验表明,相较于传统机器学习暂态稳定评估方法,所提方法能以更短的响应时间做出更准确的暂态稳定性判断,满足在线暂态稳定评估准确性与快速性的要求。
Abstract:
      The process of a system encountering transient faults is evolving over time. And its time dimension information is difficult to be captured using a transient stability assessment method based on traditional machine learning, which limits the improvement of assessment performance. Focusing on the above problem, a transient stability assessment method based on one-dimensional convolutional neural network(1D-CNN)is proposed, which can specifically be used for automatically extracting sequential features in the transient process based on measurement data at the bottom, so as to achieve the target of accurately specifying the system transient stability process with the help of its unique one-dimensional convolution and pooling operation performance. A 1D-CNN model with four convolutional layers that is applicable for transient stability assessment is designed to implement the end-to-end ‘sequential feature extraction + transient stability classification'. Moreover, the reliability of the assessment result is strengthened by adjusting key parameters of the model to improve the recall rate of the samples losing stability. According to the simulation experiment of New England 10-machine 39-bus test system, it shows that the proposed method can judge transient stability more accurately in a shorter responding time in comparison to the transient stability assessment method based on traditional machine learning, well satisfying the accuracy and rapidity required in the online transient stability assessment.
查看全文(Free!)   查看/发表评论  下载PDF阅读器