Classification of Partial Discharge Measured under Different Levels of Noise Contamination

Area of Science:

• Electrical Engineering
• Materials Science

Background:

• Cable joint insulation breakdown leads to significant power company losses.
• Early detection of insulation failure is vital for preventing such losses.
• Partial discharge (PD) patterns correlate with insulation quality, but recognition is often hindered by noise and lack of real-world data.

Purpose of the Study:

• To classify actual cable joint defect types from partial discharge (PD) data contaminated by noise.
• To evaluate the effectiveness of different feature extraction methods and artificial intelligence classifiers under noisy conditions.

Main Methods:

• Five cross-linked polyethylene (XLPE) cable joints with artificial defects were prepared.
• Three feature types (statistical, fractal, PCA) were extracted from PD patterns in a noisy environment.
• Classifications were performed using Artificial Neural Networks (ANN), Adaptive Neuro-Fuzzy Inference System (ANFIS), and Support Vector Machine (SVM).

Main Results:

• Classification accuracy decreased with increasing noise levels.
• Principal Component Analysis (PCA) features, when used with Support Vector Machine (SVM) and Artificial Neural Networks (ANN), demonstrated the highest tolerance to noise contamination.
• The study successfully classified PD defect types in a realistic, noisy environment.

Conclusions:

• The developed method shows promise for diagnosing cable joint insulation quality in the presence of noise.
• PCA features combined with SVM and ANN are effective for robust PD pattern recognition in noisy conditions.
• This research contributes to improved reliability and reduced losses in power distribution systems.