Fault Diagnosis of Distribution Terminal Units' Measurement System Based on Generative Adversarial Network Combined with Convolutional Neural Network. Issue 1 (January 2020)

• Record Type: Journal Article
• Authors: Zhou, Q F; Ge, Y; Song, X D; Lai, K; Guo, J C
• Is Part Of: IOP conference series. Volume 752:Issue 1(2020)
• Publication Date: 2020-01
• Subjects: Materials science -- Periodicals; 620.1105
• DOI: 10.1088/1757-899X/752/1/012016
• Languages: English
• ISSNs: 1757-8981

Abstract

Abstract: As the condition monitoring and control device in the distribution automation system, the abnormal or fault state of distribution terminal units' measurement system will negatively affect the quality of measured electrical quantities, and therefore, the fast and accurate discrimination of the abnormal state's data will improve the reliability of distribution automation system. This paper proposes a method, which is based on generative adversarial network (GAN) combined with convolutional neural network (CNN), to discriminate the specific fault category of distribution terminals' measuring electrical data. Firstly, four fault state characteristic time-frequency domain graph models of terminals' AC voltage sampling data are established, which based on Fourier transform (STFT). Then, take advantage of GAN's reconstruction of input graph data, to generate additional time-frequency sample graphs expanding the sample size of training set, which will be used to train a CNN to diagnosis and classify the fault state of terminals from measuring data. Finally, three training sets with different capacity expansion modes are set up to compare and verify that the method of GAN combined with CNN proposed in this paper improves the discrimination accuracy on the fault data, and the validation of diagnosis terminals' measurement system.