A Combination of Partial Discharge Signal Envelope Analysis and Artificial Neural Networks for Identifying Faults in High-Voltage Cable Joints

High-voltage cable joints are one of the critical points in the electric power system that are prone to damage due to insulation degradation. The phenomenon of Partial Discharge (PD) activity often appears as an early indicator of declining insulation performance. Contamination or internal defects in cable insulation can trigger PD. The limited amount of PD data and the complexity of PD signals are the major challenges in identifying the type of damage. This study proposes an identification method based on the extraction of envelopeshaped PD signals using data augmentation techniques combined with an Artificial Neural Network (ANN) for the classification and identification of damage in high-voltage cable joints. PD data was obtained by dividing the PD signal in the phase domain into several segments and taking the maximum value in each segment to obtain an informative numerical value representing the envelope signal. To address data limitations, the Deep Auto-Encoder Generative Adversarial Network (DAE-GAN) method was used to generate augmented data. The augmented data were validated using combination of DTW ratio and Pearson Correlation to ensure similarity of characteristics with the original data. The classification process was performed using ANN with a two-stage training method: pre-training and fine-tuning. The results of the study indicate that the proposed method is capable of improving classification and identification performance with high level of accuracy, achieving an accuracy of 93.33% and macro F1-score of 91.11%.