Fault diagnosis of HVCB via the subtraction average based optimizer algorithm optimized multi channel CNN-SABO-SVM...

Area of Science:

• Electrical Engineering
• Artificial Intelligence
• Mechanical Engineering

Background:

• Ensuring electric power system stability relies on accurate High Voltage Circuit Breaker (HVCB) mechanical fault diagnosis.
• Deep learning methods often exhibit poor performance with limited sample data, posing a challenge for HVCB fault detection.

Purpose of the Study:

• To develop an advanced HVCB operating mechanism fault diagnosis model that overcomes the limitations of deep learning with small datasets.
• To enhance the accuracy and generalization ability of fault diagnosis for HVCB by leveraging multimodal data fusion and optimized classification.

Main Methods:

• A multi-channel Convolutional Neural Network (CNN) was employed for feature extraction and fusion from multimodal HVCB data (vibration and sound).
• Support Vector Machine (SVM) was utilized for classifying fused features, offering improved performance over Softmax in limited data scenarios.
• The Subtraction-Average-Based Optimizer (SABO) was introduced for hyperparameter optimization of the SVM classifier, further boosting diagnostic accuracy.

Main Results:

• The proposed multi-channel CNN-SABO-SVM (MCCSS) model demonstrated superior performance compared to unimodal CNN and multi-channel CNN-SVM models.
• The MCCSS model achieved accuracy improvements of 2.66% and 10.66% over the comparative models.
• Experimental validation on an HVCB fault test platform confirmed the model's effectiveness in diagnosing faults with limited sample data.

Conclusions:

• The developed MCCSS model effectively addresses the challenge of HVCB fault diagnosis under limited sample conditions.
• Multimodal data fusion combined with an optimized SVM classifier provides a robust solution for improving the reliability of power systems.
• The study highlights the potential of advanced deep learning and optimization techniques for critical infrastructure monitoring.