基于EMD盲源分离算法的电力系统低频振荡模式识别

针对广预测量系统低频振荡过程中的高斯噪声干扰和定阶问题,提出了基于EMD(empirical mode decomposition)盲源分离(blind source separation,BSS)算法的单通道低频振荡信号的模式分析方法。首先将信号利用经验模态分解得到一系列本征模函数分量组合的新信号;其次针对存在模态混叠的本征模函数分量,提出利用信号周期性构造其多路信号,并利用独立分量分析消除模态混叠的有效方法;然后利用盲源分离技术--二阶盲辨识算法(second order blind identification,SOBI),处理多通道观测信号矩阵,从中提取出不同的单模式信号;最后将去噪、定阶后的信号运用最小二乘-旋转不变技术(TLS-ESPRIT)算法辨识,得到低频振荡模态参数。数值算例仿真、IEEE四机两区域仿真实验表明该算法能够有效分离源信号,相比于其他方法具有抗噪性能好、拟合精度高等优点。

Low frequency oscillation pattern recognition of power system based on EMD blind source separation algorithm

Aiming at Gaussian noise interference and ordering in the low-frequency oscillation process of wide predictive systems, a mode analysis method based on EMD(empirical mode decomposition) blind source separation(BSS) algorithm for single-channel low-frequency oscillation signals is proposed. Firstly, the signal is decomposed by empirical mode to obtain a series of new signals combining eigenmode function components. Secondly, for the eigenmode function component with modal aliasing, it is proposed to construct its multipath signal periodically by using signals and use independent components. Analyze an effective method to eliminate modal aliasing;then use the blind source separation technique-Second Order Blind Identification(SOBI) to process the multi-channel observation signal matrix, and extract different single-mode signals from it;The denoised and fixed signals are identified by the least squares-rotation invariant technique(TLS-ESPRIT) algorithm to obtain the low frequency oscillation modal parameters. The numerical example simulation and IEEE four-machine two-region simulation experiments show that the algorithm can effectively separate the source signals, and has the advantages of good anti-noise performance and high fitting precision compared with other methods.