Classification of the Type of Harmonic Source Based on Image-Matrix Transformation and Deep Convolutional Neural Network

Abstract

The classification of harmonic source types is a necessary step to alleviate harmonic pollution. This study proposes a method for the harmonic source classification based on 2-D image-matrix transformation (IMT) and deep convolutional neural network (CNN). The method firstly converts the V-I waveforms of different harmonic source types into a matrix by IMT, which was then applied to express the V-I waveform of each harmonic source as an input CNN model. Thereafter, features were extracted automatically from the harmonic source IMT matrix by the CNN model. Finally, favorable results are achieved after identifying the type of harmonic source via a classifier. In case study, a number of verifications were performed on the proposed method. Moreover, compared with existing harmonic source classification methods, the proposed strategy in this study not only avoids extracting features through expert knowledge, but also reduces the amount of data required for the harmonic source classification, and achieves higher accuracy than other methods. INDEX TERMS IMT, CNN, type of harmonic source, classification.

Multiple harmonic sources identification including inverter‐based distributed generations using empirical Fourier decomposition

“…Besides, the proposed method has better detection accuracy for the IEEE‐69 bus test system. In this paper, the best classification accuracy for nineteen classes is 99.89%, and its detection time is 0.028432 s. Also, in [12, 17, 19, 21], the detection accuracy and the detection time of the signals contaminated with different noise levels were not studied. The papers [10–13] use multiple‐point measurement methods which have drawbacks such as high computational burden and high memory storage.…”

“…Nevertheless, gathering information from both voltage and current is time-consuming and different harmonic sources, such as inverter-based DGs were not considered. Two-dimension image-matrix transformation and deep convolutional neural networks were used in [17] to classify the harmonic sources. The authors in [18] presented a method based on the artificial neural network (ANN) for harmonic source identification.…”

“…The authors in [18] presented a method based on the artificial neural network (ANN) for harmonic source identification. The papers [17,18] did not consider sources like the doubly-fed induction generator (DFIG) wind turbine and the EAF as harmonic sources. The advantages and drawbacks of the four types of neural networks cascade, namely, cascade-forward back propagation neural network (CFBPN), feed-forward back propagation neural network (FFBPN), radial basis function neural network (RBFN), and exact radial basis function neural network (ERBFN) were investigated in [19] for harmonic source detection.…”

“…The classification accuracy, detection time, and methods were compared in this table. As can be seen in this table, references [10,11,12,13,15,16,17,19,21] did not talk about the time used for detecting the harmonic sources, which is one of the essential issues in this field. Also, the inverter-based DGs were not considered in those articles.…”

“…In [12], the classification performance for ten classes was 93.10%, which is 6.48% less than our proposed approach. In [17], only five classes were investigated; thus, 15 classes were fewer than our proposed method. The classification accuracy in [21] is 99.89%, and its detection time is 0.0194 s; however, the number of classes is nine, which is ten numbers fewer than our article.…”