Machine Learning-Based Fault Location for Smart Distribution Networks Equipped with Micro-PMU

Area of Science:

• Electrical Engineering
• Power Systems
• Machine Learning Applications

Background:

• Unpredictable faults in distribution networks threaten public safety and cause power outages.
• Distributed generations (DGs) in smart grids introduce variable current levels and impedances, complicating fault detection.
• Accurate fault location is crucial for rapid network restoration and minimizing financial losses.

Purpose of the Study:

• To develop a novel machine learning-based fault location method for smart distribution networks.
• To ensure the method's effectiveness regardless of fault characteristics and DG performance.
• To utilize micro-phasor measurement units (micro-PMUs) for enhanced fault detection accuracy.

Main Methods:

• Employed machine learning algorithms using voltage data recorded by micro-PMUs at substations and DGs.
• Utilized frequency components of voltage signals as feature vectors.
• Applied Neighborhood Component Feature Selection (NCFS) for feature extraction and dimensionality reduction, followed by a Support Vector Machine (SVM) classifier.

Main Results:

• The proposed method demonstrated notable accuracy in identifying faulty sections across various fault types.
• Simulations on an 11-node IEEE standard feeder with three DGs validated the algorithm's performance.
• The approach effectively reduced uncertainties in protection systems, even with complex DG interactions.

Conclusions:

• The developed machine learning approach offers a robust solution for fault location in smart distribution networks.
• Micro-PMU data, processed via NCFS and SVM, provides a reliable basis for accurate fault identification.
• This method enhances the resilience and efficiency of power distribution systems.