arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 1904.02836eess.SP
收藏

基于机器学习的EMD域高阶统计电能质量扰动分类方法

A Machine Learning Based Classification Approach for Power Quality Disturbances Exploiting Higher Order Statistics in the EMD Domain

Faeza Hafiz, Celia Shahnaz

摘要:

提出了一种基于经验模式分解(EMD)和$k$最近邻($k$ -NN)分类器的电能质量扰动模式识别新方法。由于EMD将信号分解为与原信号长度相同的时域内的本征模函数(IMF),它保留了隐藏在傅立叶域或小波系数中的信息。在该方法中,电力信号被分解成EMD域的IMF。由于原始信号中存在非线性和噪声,很难用二阶统计量进行分析。因此,一个有效的特征集开发考虑高阶统计量(HOS),如方差,偏度,峰度从分解的前三个IMF。该特征向量被馈送到不同的分类器中,如$k$ -NN、概率神经网络(PNN)和径向基函数(RBF)。在所有分类器中,$k$ -NN在训练和测试中均表现出较高的分类精度和鲁棒性,以检测PQ扰动事件。仿真结果表明,与其他最先进的方法相比,所提出的基于HOS-EMD的方法和$k$ -NN分类器在分类精度和计算效率方面都优于其他方法。

英文摘要:

The aim of this paper is to propose a new approach for the pattern recognition of power quality (PQ) disturbances based on Empirical mode decomposition (EMD) and $k$ Nearest Neighbor ($k$-NN) classifier. Since EMD decomposes a signal into intrinsic mode functions (IMF) in time-domain with same length of the original signal, it preserves the information that is hidden in Fourier domain or in wavelet coefficients. In this proposed method, power signals are decomposed into IMFs in EMD domain. Due to the presence of non-linearity and noise on the original signal, it is hard to analyze them by second order statistics. Thus, an effective feature set is developed considering higher order statistics (HOS) like variance, skewness, and kurtosis from the decomposed first three IMFs. This feature vector is fed into different classifiers like $k$-NN, probabilistic neural network (PNN), and radial basis function (RBF). Among all the classifiers, $k$-NN showed higher classification accuracy and robustness both in training and testing to detect the PQ disturbance events. Simulation results evaluated that the proposed HOS-EMD based method along with $k$-NN classifier outperformed in terms of classification accuracy and computational efficiency in comparison to the other state-of-art methods both in clean and noisy environment.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑