Recognition of Typical Partial Discharge Defects Using 1D-CNN Feature Extraction and D-S Decision Fusion

This study establishes a comprehensive framework for partial discharge detection that combines 1D-CNN feature extraction with enhanced D-S evidence fusion. The methodology employs dual 1D-CNN analysis of HFCT and UHF temporal waveforms, adaptive fundamental probability assignments with uncertainty quantification, and sensor weighting based on accuracy. Validation achieves a fusion accuracy of 96.2%, markedly improving challenging identifications, including metal particle discharge to 92.7% and internal discharge to 98.5%. The method effectively mitigates HFCT’s susceptibility to interference and UHF’s signal obstruction issues through D-S evidence fusion. The approach demonstrates considerable robustness for field applications by dynamically adjusting sensor contributions according to reliability measures. This dual-sensor fusion method represents a significant improvement in accurate insulation diagnostics in noisy operational environments, leveraging the complementary capabilities of sensors for enhanced fault identification effectiveness.