Deep learning-based fault detection and location in underground power cables using resonance frequency analysis

Despite the numerous benefits of underground cabling in modern power distribution systems, the persistent challenge of fault detection and localization remains a critical concern. The reliability of underground power systems hinges upon swift and accurate fault detection to facilitate rapid grid restoration. Addressing this challenge, this study introduces a pioneering approach that integrates deep learning with conventional power signal analysis techniques, with a specific emphasis on frequency domain analysis. Key to the proposed method is the exploitation of the resonance frequency behavior exhibited by underground power cables in response to signal fluctuations. These resonance frequencies, inherently linked to cable parameters, offer a unique insight into fault detection and location. By leveraging this feature, the proposed methodology enables precise identification and location of faults. This study introduces a pioneering approach that integrates long short-term memory networks with conventional power signal analysis techniques with a specific emphasis on frequency domain analysis. Central to the approach is the utilization of the system’s normal resonance frequency as a benchmark for training the deep learning network. Through the induction of simulated faults at various positions along the cable network and subsequent analysis of resulting resonance frequency variations, the network is effectively trained to discern faults accurately. Validation of the proposed methodology is conducted through extensive MATLAB simulations on a representative network. The outcomes validate the method’s effectiveness and precision, highlighting its capacity to improve the dependability and efficiency of subterranean power distribution networks.