Domain-Adaptive Transformer Partial Discharge Recognition Method Combining AlexNet-KAN with DANN.

Area of Science:

• Electrical Engineering
• Artificial Intelligence
• Machine Learning

Background:

• Power transformer operating conditions can alter partial discharge (PD) data distributions, leading to unlabeled new data.
• This data shift degrades existing PD detection models, reducing classification performance.
• Accurate PD recognition is crucial for transformer health monitoring and preventing failures.

Purpose of the Study:

• To develop a domain-adaptive method for recognizing transformer partial discharge data under changing operating conditions.
• To improve the accuracy and robustness of PD detection models when faced with unlabeled or sparsely labeled new data.
• To address the challenge of distribution shift in PD data.

Main Methods:

• Proposed a novel AlexNet-KAN model, integrating the Kolmogorov-Arnold Network (KAN) to enhance the AlexNet architecture for improved PD recognition accuracy.
• Applied domain adversarial neural networks (DANNs) from domain adaptation theory to the PD recognition domain.
• Developed a domain-adaptive transformer PD recognition model combining AlexNet-KAN with DANNs.

Main Results:

• The proposed AlexNet-KAN model demonstrated improved accuracy in transformer partial discharge recognition.
• The domain-adaptive model effectively adapted PD data from source to target domains.
• Successfully addressed the distribution shift issue in transformer PD data, even with limited or no new labels.

Conclusions:

• The proposed domain-adaptive transformer partial discharge recognition method effectively handles distribution shifts in PD data.
• The integration of AlexNet-KAN and DANNs provides a robust solution for recognizing PD with unlabeled or few-labeled new data.
• This approach significantly enhances the reliability of PD detection models in dynamic power transformer environments.