The method to reduce identification feature of different voltage sag disturbance source based on principal component analysis

Abstract

Voltage sag is a typical power quality disturbance. Identify the type of disturbance source causing voltage sag accurately is one of the important matters in power quality monitoring and management. Due to the correlativity and redundancy of the features, the identification method for voltage sag disturbance source is low accuracy. To resolve the problem, this paper proposes a method of feature reduction of voltage sag disturbance based on principal component analysis (PCA). Through the analysis of single disturbance source of voltage sag and composite disturbance source of voltage sag, multiple feature indices of voltage sag are obtained using wavelet coefficients in terms of statistics , wave morphology , entropy , energy , etc. These original feature indices are correlative and redundant. Based on PCA, the original feature indices are normalized, and then the correlation coefficient matrix is calculated, a couple of comprehensive feature indices after reduction can be obtained lastly. The correlativity and redundancy of the comprehensive feature indices are eliminated effectively. The support vector machines (SVM) is used to verify the method. The simulation results show that comprehensive feature indices after reduction can effectively reduce the number of feature vectors which are input to SVM and the identification accuracy which is obtained using comprehensive feature indices is higher than the original features indices in the classification and identification of single and composite disturbance sources of voltage sag under different noisy conditions.

Classification of voltage sag disturbance sources using fuzzy comprehensive evaluation method

“…By comparing the usage frequency of U 、 R 、 H and considering the effect of classification, according to Table 1, the judgment matrix A can be got as (4) shows. 1 2 2 1 / 2 1 2 1 / 2 1 / 2 2      A (4) Where, the rank of the indexes is U, R, and H. Then eigenvalues of A were calculated and the weight vector was got as w={0.4934 0.3108 0.1958}. The consistency of it satisfied the requirement: CR=CI/0.52=0.0515<0.1.…”

“…Recently, studies on classification or recognition of voltage sag disturbance sources tend to use frequency domain transform methods (such as S transform, Hilbert-Huang transform, wavelet transform and so on) in order to extract features of voltage sags which followed by adopting different classification strategies to identify the disturbance sources. For instance, as for extraction of features for sags, S transform was applied in [2], wavelet transform in [3][4], and Hilbert-Huang transform in [5].…”

“…As for classification strategies, neural network was adopted in [2] and support vector machine (SVM) was used in [3][4]. Similarity analysis between measured signal template and standard template is an alternate.…”

Recognition method of voltage sag causes based on two‐dimensional transform and deep learning hybrid model

“…Pattern recognition uses a classification algorithm to design a classifier for determining the voltage sag causes of the disturbance signals. Common methods include neural network [11][12][13], support vector machine [14], principal component analysis reduction [15], fuzzy comprehensive evaluation [16], label propagation [17], etc. Although these methods have obtained good performance, they require the setting of several threshold levels, human expert knowledge or extracting the fundamental frequency component, which will result in an incomplete description of the data.…”

Research on classification of voltage sag sources based on recorded events

“…[3]- [4]), detection and identification (e.g. [5]-[8]) of voltage sags. In [3], the voltage sag events obtained from surveys in medium and low voltage networks were classified.…”