Power line fault diagnosis based on convolutional neural networks

Area of Science:

• Electrical Engineering
• Computer Science
• Artificial Intelligence

Background:

• Power security is crucial for national and economic stability.
• Power line faults significantly disrupt social production and daily life.
• Accurate and timely fault diagnosis is essential for maintaining grid reliability.

Purpose of the Study:

• To investigate the effectiveness of Convolutional Neural Networks (CNNs) for power line fault diagnosis.
• To compare the performance of CNNs against other deep learning models like recursive neural networks and deep belief networks.
• To analyze the impact of CNN structural parameters on fault diagnosis accuracy and stability.

Main Methods:

• Analysis of the structure and working principles of Convolutional Neural Networks (CNNs).
• Application of CNN models for diagnosing faults in power lines.
• Simulation and analysis of results, comparing CNNs with recursive neural networks and deep belief networks.

Main Results:

• CNNs achieved the highest stable fault diagnosis accuracy (100%), outperforming recursive neural networks (93.4%) and deep belief networks (91.5%).
• CNNs demonstrated superior stability, with an accuracy standard deviation close to 0.
• Optimizing CNNs by reducing batch numbers and increasing training sessions positively impacts fault diagnosis accuracy.

Conclusions:

• CNNs are highly effective and stable for power line fault diagnosis.
• The architecture and training parameters of CNNs significantly influence their diagnostic performance.
• CNNs represent a promising deep learning approach for enhancing power system reliability.