Machine learning for anomaly states recognition in cable joints: a multidimensional dataset-driven approach for industrial intelligent monitoring

Abstract Accurate and rapid recognition of anomaly states in cable joints is critical for maintaining the operational safety and stability of power transmission systems. However, most existing studies focus on single fault or parameter discrimination. In this work, a cost-effective and efficient strategy is proposed to address the monitoring of multiple anomaly states in cable joints. A multi-dimensional dataset is constructed using real-world collected data. Six machine learning algorithms are employed to perform anomaly recognition. Results demonstrate that XGBoost achieves the highest accuracy of 97.14% as the most effective model. Additionally, cross-validation confirms its strong generalization capability. This work demonstrates practical applicability and lays the foundation for intelligent industrial monitoring.