Transformer fault identification based on GWO-optimized Dual-channel M-A method

Area of Science:

• Electrical Engineering
• Artificial Intelligence
• Computational Intelligence

Background:

• Transformer fault identification is crucial for power system reliability.
• Existing methods often lack sufficient accuracy, necessitating advanced diagnostic approaches.
• Nature-inspired algorithms offer potential for optimizing complex identification models.

Purpose of the Study:

• To develop an accurate transformer fault identification method using a Grey Wolf Optimizer (GWO)-optimized Dual-channel Multilayer Perceptron (MLP)-Attention model.
• To enhance the performance of transformer fault diagnosis by leveraging hybrid AI techniques.
• To evaluate the effectiveness of the proposed GWO-optimized model against conventional methods.

Main Methods:

• Construction of a Dual-channel model integrating Attention Mechanism (AM) and MLP.
• Optimization of the Dual-channel MLP-Attention model's hidden layer structure using the GWO algorithm.
• Simulation of typical transformer faults using a Digital Dynamic Real-Time Simulator (DDRTS) system.

Main Results:

• The GWO-optimized method achieved transformer fault identification accuracy rates between 95.3% and 96.7%.
• Significant accuracy improvements were observed: 14.1% over BP, 9.6% over SVM, 9.3% over MLP, and 3.3% over single-channel M-A models.
• Experimental validation confirmed the superiority of the proposed approach.

Conclusions:

• The GWO-optimized Dual-channel MLP-Attention model is a rational and effective method for transformer fault identification.
• This approach offers a substantial improvement in diagnostic accuracy for power transformers.
• The study highlights the potential of combining nature-inspired optimization with deep learning for critical infrastructure monitoring.