Location Method of Multiple Partial Discharges Sources in Substation Space by UHF Method Based on Intelligent Clustering Algorithm

The partial discharge (PD) location technology in substations based on ultra high frequency (UHF) method has been ex-tensively applied, however, the current research mainly focuses on single source location, and there is a lack of exploration of the problem of multi-source PDs location. Consequently, we investigated the application of clustering algorithm in multi-source PDs location. First, the time difference of arrival (TDOA) of two and three sources UHF signals was calculated by combining the threshold method and the energy accumulation function method, and the data were collected for clustering analysis. Compared the commonly used K-mean clustering algorithm with the Two-step clustering algorithm, the average error of the latter was 18.7% lower than that of the former. Therefore, based on the Two-step algorithm, a 3σ-Two step algorithm suitable for single and multiple sources PD was proposed according to the idea of two-dimensional Gaussian normal distribution. Compared with the Two step algorithm, the average error is 22.5% lower. Secondly, the improved algorithm was applied to the positioning test of multi-source PD, and the Stable Point Iteration method was used to solve the coordinates of the PD sources. The laboratory location results show that, the average error is obtained after multiple positioning tests, the error of each point of the two-source of two group tests are 0.34 m and 0.49 m, the error of each point of the three-source is 0.72 m, the distance between the two sources increases by 2.62 m, and location error is reduced by 30.6%. When the distance between the dual source and the sensor is doubled, the average positioning error of each point is 0.46 m, and increases by 35.3% compared with the case of not increasing the distance between the double source and the sensor. Finally, the algorithm and equipment were applied to a substation in Fuzhou. After tests for many times, a corona discharge on a pillar insulator conld be located, an occasional discharge could be located at an isolating switch, and the location errors of each point was within the range of 0.6~2.2 m. Compared with the results in the laboratory, the location accuracy is lower but still meets the actual requirements, which verifies the effectiveness of the algorithm. © 2020, High Voltage Engineering Editorial Department of CEPRI. All right reserved.