Sensor-Array Optimization Based on Multicorrelation Analysis for Cable Thermal Fault Diagnosis

Cables are critical components in power systems, and diagnosing their thermal faults holds significant engineering value. Electronic nose systems have been widely adopted for this purpose, enabling cost-effective and efficient spatial temperature monitoring. However, existing methods for optimizing sensor array data remain insufficient, requiring further refinement. This study introduces, for the first time, the HPMI method—a multi-correlation evaluation framework for array features. HPMI quantifies linear (via Pearson correlation) and nonlinear (via Hilbert-Schmidt Independence Criterion, HSIC) relationships between features and temperature labels while assessing feature redundancy using mutual information and conditional mutual information. A dedicated sensor array system and experimental platform for cable thermal fault diagnosis were also developed. Experimental results demonstrate that the proposed method outperforms existing approaches in selecting optimal feature subsets for temperature identification. On balanced datasets, it reduced features from 90 (9 sensors) to 24 (6 sensors) while improving accuracy by 11.81%. For imbalanced datasets, it achieved a 6.82% accuracy gain using only 26 features, surpassing existing methods.