一种低频振荡模式在线辨识的鲁棒递推ARMA算法  

Low frequency oscillation modes online estimation with robust recursive ARMA algorithm

作　　者：刘贵富[1] 卢继平[1] 陈刚 王予疆[1] 徐玉韬[1] 何潜[3] 雷雨[3] 梁伟[3] 李军[3] 

机构地区：[1]重庆大学输配电装备及系统安全与新技术国家重点实验室,重庆400044 [2]南方电网科学研究院,广东广州510080 [3]重庆市电力公司,重庆400015

出　　处：《电力系统保护与控制》2012年第14期74-80,共7页Power System Protection and Control

基　　金：国家973计划资助项目(2009CB724505-1);国家111计划资助项目(B08036);重庆电力公司科技项目(20100658)~~

摘　　要：低频振荡模式在线辨识具有重要的工程应用价值。采用ARMA模型的常规递推算法(RLS)以及正则化鲁棒递推算法(R3LS)可有效辨识稳态类噪声信号,但对动态信号的辨识效果不太理想。分析了常规RLS算法在出现动态信号时可能发散的原因,通过在算法中引入基于输入信号自相关矩阵和互相关向量的L1范数的自适应权重参数,保证算法在各种条件下的收敛性,实现了一种全新的鲁棒递推ARMA算法(NRRLS)。采用IEEE-39节点系统时域仿真和某电网的PMU实测数据进行了大量的分析测试,并通过与RLS算法和R3LS算法的辨识效果进行比较,验证了NRRLS算法具有更好的鲁棒性和辨识精度,扩展了ARMA模型的应用范围,具有较高的理论和工程实用意义。Low frequency oscillation modes online estimation has important engineering application. It is effective to estimate ambient signal using ARMA model of conventional recursive algorithm （RLS） and regularization robust recursive algorithm （R3LS）, but is not appropriate to identify the ringdown data. This paper analyzes the possible divergence reason of conventional RLS algorithm in ringdown data, and proposes a new robust recursive ARMA algorithm （NRRLS） to ensure the convergence of the algorithm under various conditions, by choosing the weights on the basis of the L1 norm of the input-signal autocorrelation matrix and cross-correlation vector. The algorithm analyzes a lot of the simulation from IEEE-39 bus power system and the measured data in some power system, and compares it with the identification result of conventional RLS and regularized robust RLS（R3LS）. The result validates robustness and accuracy of the proposed algorithm, which extends the application of ARMA model, and has high theoretical and practical engineering significance. This work is supported by National Program on Key Basic Research Project （973 Program） （No. 2009CB724505-1）.

关 键 词：ARMA 鲁棒递推 在线辨识 动态信号 低频振荡模式 

分 类 号：TM712[电气工程—电力系统及自动化]