Partial Discharge Signal Analysis Method for Transmission Cables Supported by Deep Learning

This work aims to construct an efficient model for analyzing partial discharge signals in cables using deep learning technology. By employing a convolutional neural network (CNN) as the core architecture and combining it with the publicly available dataset MCSG-PD-6016 for model training and validation, accurate identification and classification of cable partial discharge signals have been achieved. First, this work preprocesses the cable partial discharge signals and extracts key features. Then, a deep learning model based on CNN is constructed. Finally, the performance of the model is optimized by continuously adjusting the model parameters and structure, leading to improved accuracy in detecting cable partial discharge signals. Experimental results show that the accuracy of the model in actual detection ranges from 81% to 94%, with a recall rate of approximately 83% to 96%. The research results demonstrate significant achievements in the detection of cable partial discharge signals with the model designed. The model’s accuracy remains stable at a high level, and the recall rate performs excellently, fully demonstrating the effectiveness of deep learning in analyzing cable partial discharge signals. This work not only provides new technical means for cable fault detection and diagnosis but also offers valuable references for the application of deep learning in power systems.