文章摘要
李洋麟,江全元,颜融,等.基于卷积神经网络的电力系统小干扰稳定评估[J].电力系统自动化,2019,43(2):50-57. DOI: 10.7500/AEPS20171203004.
LI Yanglin,JIANG Quanyuan,YAN Rong, et al.Small-signal Stability Assessment of Power System Based on Convolutional Neural Network[J].Automation of Electric Power Systems,2019,43(2):50-57. DOI: 10.7500/AEPS20171203004.
基于卷积神经网络的电力系统小干扰稳定评估
Small-signal Stability Assessment of Power System Based on Convolutional Neural Network
DOI:10.7500/AEPS20171203004
关键词: 卷积神经网络  深度学习  小干扰稳定评估  关键特征值  广域测量系统
KeyWords: convolutional neural network  deep learning  small-signal stability assessment  key eigenvalues  wide-area measurement system
上网日期:2018-11-16
基金项目:国家自然科学基金资助项目(51677164)
作者单位E-mail
李洋麟 浙江大学电气工程学院, 浙江省杭州市 310027  
江全元 浙江大学电气工程学院, 浙江省杭州市 310027  
颜融 浙江大学电气工程学院, 浙江省杭州市 310027  
耿光超 浙江大学电气工程学院, 浙江省杭州市 310027 ggc@zju.edu.cn 
摘要:
      随着电力系统规模的增大,通过传统数值方法计算系统特征值来进行小干扰稳定评估已法满足实时分析的要求。因此,提出了一种基于深度学习(卷积神经网络)的电力系统小干扰稳定评估方法。该方法以广域测量系统可监测变量作为模型的输入,关键特征值作为输出,对输入数据和输出数据进行相应处理后,利用深层架构对其映射关系进行分析;并针对大系统维数较高、训练速度较慢的问题,采用了离散余弦变换和图形处理器并行技术。算例结果表明,在不考虑控制参数变化的情况下,经过历史数据的离线训练后,该方法能够较准确地计算出系统的关键特征值。
Abstract:
      As the scale of power system increases, traditional numerical methods in eigenvalue computation for small-signal stability assessment are unable to meet the requirement of real-time analysis. Therefore, this paper proposes a small-signal stability assessment method based on deep-learning(convolutional neural network). This method takes the signals of wide-area measurement system as inputs of the model and generates critical eigenvalues as outputs. After the necessary preprocessing of the inputs and outputs, the mapping relationship between inputs and outputs can be established by deep neural network. Discrete cosine transformation and graphics processing unit parallelization techniques are employed to overcome the challenges from high dimension and slow training rate of large-scale system. Case study results indicate that the proposed method is able to accurately obtain the critical eigenvalues of the studied system after offline training using historic data, given no significant change in control parameters.
查看全文(Free!)   查看附录   查看/发表评论  下载PDF阅读器