基于带宽自适应滤波的低频振荡Prony分析

Prony Analysis of Low Frequency Oscillations Based on Adaptive Filtering of Bandwidth

针对传统Prony法在分析低频振荡时对噪声非常敏感的缺点,提出了一种基于带宽分析的余弦基神经网络滤波方法。首先,利用余弦基神经网络逼近低频振荡信号,通过对权值的分析,确定信号有效带宽;然后根据信号带宽进行带通滤波,再将输出信号导入Prony模块分析。其中,针对有效带宽范围的确定,提出了固定带宽与动态带宽的分析方法。分别在脉冲噪声、高频谐波噪声、随机白噪声背景下进行了算例分析,表明了该方法对含噪低频振荡信号具有较好的滤波效果,有效地提高Prony算法的振荡主导模式识别精度。能满足电力系统低频振荡特征分析的需要。

As the traditional Prony method is very sensitive to noise in the analysis of the low frequency oscillation, proposed a cosine basis neural network filtering method based on bandwidth analysis. First, use the cosine-based neural network to approach the low-frequency oscillation signal and determine the effective bandwidth through the analysis of the weights. Band-pass filtering is then performed according to the signal bandwidth, and then introduces the output signal into Prony module. Methods to analyze the fixed and dynamic bandwidth are proposed in this paper to determine the scope of the effective bandwidth. The analysis of examples based on impulse noise, high-frequency harmonic noise and random white noise show that this method has a better filtering effect on noisy signal of low frequency oscillation, which can effectively improve the oscillation dominant mode identification accuracy of the Prony algorithm. It can meet the needs of low frequency oscillation characteristic analysis.