Data Description Technique-Based Islanding Classification for Single-Phase Grid-Connected Photovoltaic System.

Area of Science:

• Electrical Engineering
• Renewable Energy Systems
• Machine Learning Applications

Background:

• Conventional islanding detection methods in grid-connected photovoltaic (PV) systems suffer from non-detection zones.
• These zones pose risks to power system safety and stability during grid disturbances.

Purpose of the Study:

• To develop an advanced islanding classification mechanism for single-phase grid-connected PV systems.
• To address and overcome the limitations of non-detection zones in existing islanding detection techniques.

Main Methods:

• Adapted a support vector-based data description technique with Gaussian radial basis function kernels.
• Incorporated excess and deficit power imbalance conditions under various loading scenarios.
• Utilized voltage dip scenarios for feature extraction and machine learning classification.

Main Results:

• Experimentally validated the mechanism on a 5 kW grid-connected system.
• Achieved a high detection rate of 99.2% for islanding events.
• Demonstrated a low false alarm rate of 0.2% across multiple fault and power imbalance conditions.

Conclusions:

• The developed islanding classification mechanism effectively overcomes non-detection zones.
• The machine learning-based approach provides reliable and accurate islanding detection for PV systems.
• The proposed method enhances the safety and stability of grid-connected PV systems.