Partial Discharge Detection and Identification in High-Voltage Cable Systems Based on Heterogeneous Sensor Fusion and Enhanced Deep Learning

High-voltage cable systems are critical components of power transmission networks, where partial discharge (PD) can cause insulation degradation and severe operational failures. Accurate detection and reliable identification of PD are therefore essential, yet conventional methods are often vulnerable to electromagnetic interference and exhibit limited recognition performance. This paper proposes a PD detection and identification framework for high-voltage cable systems based on heterogeneous sensor fusion and enhanced deep learning. A high-speed optical electric-field sensor is employed to localize potential insulation defects, while an acoustic pressure wave sensor is used to confirm PD occurrence and intensity. An improved adaptive-threshold discrete wavelet transform is applied for signal denoising, and an optimized Gramian Angular Field transformation converts one-dimensional signals into two-dimensional feature representations. A residual convolutional neural network incorporating an efficient channel attention mechanism is then developed for PD pattern identification. Experiments involving corona, void, and surface discharges demonstrate that the proposed system achieves a 100% PD detection rate and a recognition accuracy of 96.0% under laboratory conditions. Field tests on 220 kV tunnel-laid cables further verify that both detection and identification accuracies reach 100%, with superior robustness and environmental adaptability compared with conventional approaches.