基于小波与隐式马尔科夫模型的发电机局部放电信号去噪

WAVELET-BASED PARTIAL DISCHARGE DENOISING USING HIDDEN MARKOV MODEL

目前基于小波变换的马尔科夫模型(HMMs)被用于图像信号处理。其方法的优势在于考虑了小波系数之间的相关性,而且在去噪时不存在待定的自由参数,具有更强的自适应性。该文使用小波域HMMs方法去除发电机局部放电信号中的白噪声。为了验证方法的有效性,采用了实验室的线棒放电信号和电厂的发电机中性点电流局放信号进行实验验证。结果证明,与传统的SURE、Fixed form门限、罚函数和Minimax等方法对比,HMMs方法能获得更高的信噪比,正确区分和保留更多的脉冲数,对进一步的绝缘状态分析有重要意义。

Wavelet-domain hidden markov models (HMMs) have recently been introduced and applied to signal and image processing. The advantage of the method is that the HMMs measure the dependency between the wavelet coefficients and has no free parameters in denoising. In this paper, the HMMs method is applied in reducing partial discharge (PD) white noise. The EM algorism is used to estimate model parameters and probabilities and the empirical Bayesian wavelet-based denoising method is applied to compute the denoised signal. The effectiveness of the method is demonstrated by using numerical simulations and real-world data of neutral point current of generator. Compared with SURE Shrinkage, Fixed form threshold, Minimax and Penalty method, the result shows that the Wavelet-based HMMs method is better in enhancing signal-to-noise ratio and reserves more PD impulses. The method can identify Partial Discharge correctly and has further use in the insulation online monitoring.