State estimator and convolutional neural networks-based fault localization approach for modern grids

Area of Science:

• Electrical Engineering
• Power Systems
• Renewable Energy Integration

Background:

• Direct current (DC) distribution networks are increasingly important due to high renewable energy penetration.
• These networks face significant challenges in fault detection, classification, and localization.
• Existing protection schemes may not be sufficient for the dynamic nature of modern DC grids.

Purpose of the Study:

• To develop and validate a robust two-stage fault identification and isolation scheme for DC distribution networks.
• To improve the speed and accuracy of fault management in grids with distributed generation.

Main Methods:

• A two-stage approach combining a cubature Kalman filter (CKF) and convolutional neural networks (CNNs).
• CKF estimates voltage and current signals for fault scenario simulation and CNN training data generation.
• Total harmonic distortion (THD) index and reactive power direction are used for fault detection, classification, and localization.

Main Results:

• The proposed scheme achieves fault detection in under five milliseconds.
• Demonstrates a high accuracy rate of 98% in fault identification and classification.
• Requires reasonable computational resources, making it practical for real-world deployment.

Conclusions:

• The proposed CKF-CNN based method offers a fast, accurate, and computationally efficient solution for DC distribution network protection.
• This approach effectively addresses the challenges posed by renewable energy integration in DC grids.
• The scheme enables reliable fault isolation, enhancing the stability and safety of modern power systems.