Abstract: The multi-sensor array demonstrates substantial potential in the intelligent diagnostic analysis of electrical equipment. Addressing the recurrent issue of ablation faults in the insulating layers of high-voltage cross-linked polyethylene cables, this study introduces a novel solution based on a multi-sensor array, leveraging the detection of characteristic gases for early fault warning and diagnosis. To optimize diagnostic accuracy and computational efficiency, a deep learning model is developed, denoted as PCA-KAN, which integrates principal component analysis(PCA) with Kolmogorov-Arnold networks(KAN). PCA facilitates data augmentation, while the pruning technique within the KAN network effectively reduces both the model's parameter count and computational complexity. Experimental results reveal that the PCA-KAN model achieves a recognition accuracy of 98.26% on the test set, surpassing traditional deep learning approaches. Additionally, the model demonstrates remarkable computational efficiency, with only 240 parameters, rendering it well-suited for deployment in edge computing environments. This system not only excels in high-voltage cable fault diagnosis, but also offers broad applicability in industrial IoT applications, including partial discharge detection in switchgear and fire prevention.