Adaptive Disconnector States Diagnosis Method Based on Adjusted Relative Position Matrix and Convolutional Neural...

Area of Science:

• Electrical Engineering
• Power Systems
• Machine Learning

Background:

• High-Voltage Disconnectors (HVDs) are susceptible to faults due to outdoor operation.
• Existing HVD diagnostic methods often lack generalizability across different HVD models.
• Data-driven fault diagnosis for HVDs faces challenges in acquiring sufficient fault samples.

Purpose of the Study:

• To propose an adaptive HVD state diagnosis method with enhanced generalization capabilities.
• To address limitations in traditional methods, particularly information loss during signal processing.
• To develop a robust approach for identifying HVD faults despite limited fault data.

Main Methods:

• Generating Fault Difference Signals (FDS) by subtracting normal and operational power signals.
• Improving the Relative Position Matrix (RPM) calculation to preserve amplitude information, converting 1D FDS to 2D images.
• Utilizing a Convolutional Neural Network (CNN) with Batch Normalization (BN) and GELU activation for classification.

Main Results:

• The proposed FDS-ARPM-CNN method achieved high-accuracy diagnosis and classification of HVD states.
• Experimental validation confirmed strong generalization capabilities of the CNN model across different HVD models.
• The method effectively mitigates the difficulty of acquiring fault samples for data-driven HVD diagnosis.

Conclusions:

• The FDS-ARPM-CNN method offers a practical and valuable solution for HVD state diagnosis.
• The adaptive recognition and generalization capabilities are key advantages for real-world applications.
• This research contributes to improving the reliability and maintenance of High-Voltage Disconnectors.