A method for partial discharge localization and type recognition of transmission lines using UAV-mounted UHF sensors

Abstract Partial discharge (PD) is an essential clue to insulating material ageing of HV power equipment, which needs to be identified at the source to avoid equipment failure. Traditional PD monitoring techniques sometimes have limited localization accuracy, sensitivity to noise, and cannot perform both PD classification and source localization. The present work introduces an intelligent UAV-assisted scheme for PD-type recognition and discharge source estimation in the transmission-line scenario based on RF/UHF sensing signals. Time difference of arrival (TDOA)-based localization is introduced to infer spatial coordinates that serve as supervisory labels, and a Hybrid 1D CNN–LSTM model is used to learn the spatial-temporal characteristics of discharge data. In addition, particle swarm optimization (PSO) is applied to optimize the model hyperparameters. The experimental evaluation uses 3,500 RF/UHF waveform samples, including 2,800 training and 700 testing samples, together with a balanced PD classification dataset containing 1,827 image samples, including 1,461 training and 366 testing samples. According to experimental data, the suggested framework achieves classification accuracy of 0.9839, precision of 0.9839, recall of 0.9838, and F 1-score of 0.9838 for PD type recognition. For localization estimation, the framework obtains an RMSE of 0.0116, MAE of 0.0082, and R 2 of 0.9961 with respect to the TDOA-derived supervisory coordinates. These results demonstrate the capability of the proposed Hybrid 1D CNN-LSTM framework to learn discriminative spatio-temporal characteristics from RF/UHF signals for simultaneous PD type recognition and localization estimation. However, since experimentally measured PD source coordinates were not available in the current dataset, the localization results represent agreement with TDOA-derived reference coordinates rather than independently measured physical source locations. Further arena validation using UAV-mounted UHF measurements and independently established PD source locations is required.