Fault Detection and Classification in MMC-HVDC Systems Using Learning Methods

Area of Science:

• Electrical Engineering
• Power Systems
• Artificial Intelligence

Background:

• Modular Multilevel Converters (MMCs) are crucial for High Voltage Direct Current (HVDC) transmission.
• Reliable open-circuit fault diagnosis is essential for MMC operational integrity.
• Existing fault diagnosis methods may require extensive feature engineering.

Purpose of the Study:

• To evaluate and compare deep learning methods for open-circuit fault diagnosis in MMCs.
• To assess the performance of Convolutional Neural Networks (CNN), Autoencoder-based Deep Neural Networks (AE-based DNN), and SoftMax classifiers.
• To determine the effectiveness of these methods using only AC-side and bridge currents without explicit feature extraction.

Main Methods:

• Implemented a two-terminal MMC-HVDC system in PSCAD/EMTDC for simulations.
• Applied CNN and AE-based DNN for fault detection and classification.
• Utilized a stand-alone SoftMax classifier for comparison.
• Input data comprised AC-side three-phase currents and MMC upper/lower bridge currents.

Main Results:

• All three methods (CNN, AE-based DNN, SoftMax) demonstrated high detection and classification accuracy.
• SoftMax classifier outperformed CNN and AE-based DNN in accuracy and testing speed.
• AE-based DNN showed slightly better detection accuracy than CNN.
• CNN required less training time compared to AE-based DNN and SoftMax.

Conclusions:

• Deep learning approaches, including CNN, AE-based DNN, and SoftMax, are effective for MMC open-circuit fault diagnosis.
• The SoftMax classifier offers a compelling balance of high accuracy and rapid testing for this application.
• The study highlights the potential of data-driven methods using readily available current measurements for MMC fault monitoring.