Intelligent islanding detection framework for smart grids using wavelet scalograms and HOG feature fusion.

Area of Science:

• Electrical Engineering
• Power Systems
• Machine Learning

Background:

• Unintended islanding in electrical distribution networks poses significant risks due to increasing distributed generation (DG).
• Conventional islanding detection schemes (IDS) struggle with balanced load-generation conditions, leading to the non-detection zone (NDZ).

Purpose of the Study:

• To develop a novel, reliable, and robust machine learning-based islanding detection scheme.
• To address the limitations of existing IDS, particularly within the NDZ.

Main Methods:

• Utilizing Continuous Wavelet Transform (CWT) to generate scalogram images from total harmonic distortion (THD) signals of voltages and currents.
• Extracting Histogram of Oriented Gradient (HOG) features from scalogram images to capture islanding signatures.
• Employing a Random Forest classifier for robust detection and minimal parameter tuning.

Main Results:

• The proposed HOG feature-based machine learning approach significantly outperforms state-of-the-art methods in accuracy, precision, recall, and F1-score.
• Demonstrated high reliability and robustness across various noise conditions and challenging scenarios.
• Real-time testing on the OPAL-RT platform confirmed practical applicability and system robustness.

Conclusions:

• This research introduces a highly effective solution for unintended islanding detection, enhancing power grid stability and safety.
• The novel machine learning methodology provides a significant advancement over conventional islanding detection schemes.
• The system offers practical reliability for contemporary electrical distribution networks with high DG penetration.