Intelligent learning approach for UHF partial discharge localisation in air-insulated substations

• Main Authors: Quanfu Zheng, Lingen Luo, Hui Song, Gehao Sheng, Xiuchen Jiang
• Format: Article
• Language: English
• Published: Wiley 2020-05-01
• Series: High Voltage
• Subjects: kalman filters; regression analysis; partial discharge measurement; particle filtering (numerical methods); wireless sensor networks; learning (artificial intelligence); gas insulated substations; air insulation; uhf measurement; uhf detectors; computerised instrumentation; intelligent learning approach; uhf partial discharge localisation; air-insulated substations; power equipment; early fault warning; data-driven partial discharge source localisation method; particle filter; wireless sensor arrays; rssi-based methods; economical adaptability solution; shadowing effects; uhf signal attenuation; kalman filter; rssi signal; semiparametric regression model; rssi ranging model; mean pd source localisation error; insulation deterioration motoring; ultrahigh frequency received signal strength indicator; uhf time-difference-based techniques; uhf received signal strength indicator; size 1.16 m
• Online Access: https://digital-library.theiet.org/content/journals/10.1049/hve.2019.0342

To achieve comprehensive insulation deterioration motoring of power equipment and early fault warning in air-insulated substations, a data-driven partial discharge (PD) source localisation method employing noisy ultra-high frequency (UHF) received signal strength indicator (RSSI) and particle filter is proposed in this study. Compared with the existing UHF time-difference-based techniques, UHF wireless sensor arrays and RSSI-based methods provide an economical and high-adaptability solution. However, owing to the multi-pathing and shadowing effects, UHF signal attenuation cannot be modelled. Therefore, a Kalman filter was employed to smoothen the RSSI signal. Furthermore, a semi-parametric regression model is proposed to achieve a more accurate relationship between the RSSI and the transmission distance. Finally, in contrast to traditional localisation algorithms directly based on the RSSI ranging model, a particle filter was used to achieve higher accuracy. It predicted the best distribution of the position of PD by learning and considering all the system states of the previous moment. The laboratory test was performed within an area of 6 m × 6 m, and the results demonstrate that the mean PD source localisation error was 1.16 m, which gives a potential application for the identification of power equipment with insulation deterioration in a substation, while the accuracy is still needed to be verified further by field tests.