Abstract:  In the capacitor online monitoring system,the disturbance of lines on different position monitoring device is different. Therefor, the use of harmonic analysis in the calculation of dielectric loss angle has instability problem in engineering. This paper proposes a capacitor dielectric loss angle identification algorithm based on the synchronous monitoring and deep learning. Firstly, we present the wireless synchronous monitoring method of capacitor current and voltage signal, and the computation process of dielectric loss angle identification signal Dδ(t) for deep learning. Then, we verify the effectiveness of the proposed method through simulation, and compare the results with the Hanning windowed harmonic analysis method. Finally, we analyze the visualization of deep neural networks hidden layer. The results show that the algorithm accuracy is affected by white-noise level, harmonic amplitude ratio and the variation level of dielectric loss angle. In situations when harmonic amplitude ratio less then 10%, the algorithm accuracy has been fewer affected by frequency deviation, phase difference of harmonics.

Key words:  deep learning, dielectric loss angle, synchronous monitoring, capacitor