Underground Power Cable Condition Monitoring and Risk Assessment Based on Distributed Perception and Optical Fiber Sensing

This paper proposes a condition monitoring and fault diagnosis method for underground power cables based on distributed optical fiber sensing (DOFS) and deep learning. A Raman-scattering-based DOFS system is deployed to collect real-time temperature, strain, and vibration data along key cable sections. To improve data quality, a wavelet-based multi-scale denoising technique is applied, followed by feature extraction using Short-Time Fourier Transform (STFT) and Hilbert-Huang Transform (HHT) for enhanced time-frequency analysis. A multi-channel Convolutional Neural Network (CNN) is developed to classify fault types by separately processing and then fusing features from temperature, strain, and vibration signals. This model accurately identifies common fault modes such as overheating, mechanical deformation, and external disturbances. Transfer learning is employed to improve generalization across different cable types. For risk assessment, a fuzzy comprehensive evaluation model is constructed, combining real-time sensing inputs with Analytic Hierarchy Process (AHP) for weight determination and Grey Relational Analysis (GRA) for fault correlation. This enables dynamic evaluation of both internal and external risks, such as construction activities or ground shifts. Experimental results verify that the proposed system achieves high fault recognition accuracy and robust early-warning capability, offering effective support for predictive maintenance in power systems.