Intelligent System of Partial Discharge Diagnostics in Power Mains

The paper is concerned with the system developed with the aid of artificial intelligence, which is capable of detecting partial discharges in overhead power line insulation. To improve fault detection accuracy, several stages of preprocessing are applied including division into frequency components, balancing, segmentation and normalization of data, which represent measuring of high-frequency parts of voltage band for each transmission line phase. It should be noted that electromagnetic waves within the frequency range of 10 and 20 MHz, which are formed at partial insulation failure, play the key role in partial discharge detection. For this purpose, the initial data are refined to separate high frequency and low frequency components. The accuracy of neural networks trained on the composite signal and its high frequency part differ by less than 1 %, while the accuracy of the network trained on the low frequency part is 17% lower. Real data were used to test the proposed system, thus confirming its applicability in practice. The developed intelligent system provides relatively high accuracy in diagnostics achieving about 88%. Thus, data mining shows that high frequency components prevail in partial discharge detection, that is, electromagnetic waves in the certain frequency band, which arise during partial insulation failure. This opens up new fields of use in developing new methods of partial discharge diagnostics on the basis of this band.