Random Forest Regressor-Based Approach for Detecting Fault Location and Duration in Power Systems

Area of Science:

• Electrical Engineering
• Data Science
• Power Systems Analysis

Background:

• Power system faults cause significant socio-economic disruption due to lengthy service interruptions.
• Current fault detection systems (relays, digital fault recorders) lack the speed for rapid fault characteristic communication.
• High-resolution Phasor Measurement Units (PMUs) enable real-time, event-driven fault analysis.

Purpose of the Study:

• To develop a data-driven method for real-time determination of power system fault characteristics.
• To accurately and simultaneously detect fault location and predict fault duration.
• To evaluate the proposed model's performance against established algorithms.

Main Methods:

• Utilized a Random Forest Regressor (RFR) model trained on fault trajectory data.
• Implemented boosting and aggregation techniques for robust generalization and accuracy.
• Evaluated RFR across four scenarios: fault location detection, duration prediction, missing data handling, and real-time streaming analysis.

Main Results:

• RFR demonstrated superior performance in fault detection accuracy and prediction error.
• The model achieved faster processing times compared to deep neural networks, Hoeffding trees, and other benchmark algorithms.
• RFR effectively handled missing data and identified fault location and length in real-time.

Conclusions:

• The proposed RFR-based approach offers a highly accurate and efficient solution for real-time power system fault analysis.
• This method significantly reduces service interruption times by enabling faster fault characteristic identification.
• RFR provides a robust and scalable alternative to existing fault detection mechanisms.