Pattern Recognition of Partial Discharge Faults in Switchgear Using a Back Propagation Neural Network Optimized by an...

Area of Science:

• Electrical Engineering
• Power Systems
• Artificial Intelligence

Background:

• Partial discharge (PD) is a critical phenomenon that compromises switchgear insulation, potentially causing equipment failure.
• Accurate detection of PD types (metal particle, suspended, creeping discharge) is vital for assessing insulation severity and preventing accidents.
• Partial discharge pattern recognition (PDPR) is essential for early identification of insulation defects in power supply systems.

Purpose of the Study:

• To propose an optimized Back Propagation Neural Network (BPNN) for enhanced partial discharge pattern recognition (PDPR) in switchgear.
• To improve the robustness and generalization ability of BPNN by addressing sensitivity to initial parameters.
• To reduce computational complexity and improve recognition efficiency through dimensionality reduction.

Main Methods:

• An improved Mantis Search Algorithm (MSA) was developed to optimize BPNN parameters, incorporating boundary handling and adaptive strategies.
• Principal Component Analysis (PCA) was employed to reduce the dimensionality of PD feature data from 14 to 7 features.
• The optimized BPNN was evaluated against Decision Tree (DT), k-Nearest Neighbor (KNN), and Support Vector Machine (SVM) classifiers.

Main Results:

• The dimensionality reduction using PCA decreased the BPNN parameter count from 183 to 113, significantly saving computation time.
• The proposed MSA-optimized BPNN achieved the highest recognition accuracy for metal particle discharge and suspended discharge types.
• Comparable recognition accuracy was maintained despite the substantial reduction in feature dimensions.

Conclusions:

• The developed MSA-optimized BPNN offers a robust and efficient solution for PDPR in switchgear.
• This approach enhances the early detection of insulation defects, contributing to improved power system reliability and safety.
• The integration of PCA and MSA provides a computationally efficient and accurate method for switchgear condition monitoring.