[Objective] To address issues in traditional low-voltage power cable fault diagnosis, such as reliance on a single signal, insufficient feature extraction, and weak anti-interference capability, an intelligent diagnostic strategy is proposed that can achieve high robustness and high-precision identification under complex operating conditions. [Methods] An intelligent diagnostic method integrating variational mode decomposition-Hilbert transform (VMD-HT) and multi-source one-dimensional convolutional neural network (MS-1DCNN) was proposed. A time-frequency analysis framework was constructed using VMD and HT to adaptively decompose signals of different modes and quantify feature parameters. Meanwhile, the MS-1DCNN structure was designed to achieve unified modeling and diagnosis of multiple types of cable faults. [Results] The experimental results demonstrated that the proposed MS-1DCNN diagnostic model outperformed conventional methods in terms of fault feature separability, classification accuracy, and stability under complex noise conditions. Superior robustness to hyperparameter variations was also verified. [Conclusion] The proposed MS-1DCNN model significantly enhances the reliability of fault identification in low-voltage cables, making it suitable for online monitoring and early warning scenarios in actual power grids. It provides a scalable technical solution for ensuring the operational safety of low-voltage distribution systems. Key words: low-voltage power cable; fault diagnosis; variational mode decomposition; Hilbert transform; multi-source one-dimensional convolutional neural network