An Open-Circuit Faults Diagnosis Method for MMC Based on Extreme Gradient Boosting

Abstract

The diagnosis of open circuit faults is required for the reliability of modular multilevel converters (MMC) and the complex operating environment of MMC may cause missing data and external noise in signal detections, where some of the present open circuit fault detection methods can be disabled. This paper proposed a machine learning (ML) diagnosis strategy for open circuit faults in MMC based on extreme gradient boosting (XG-Boost). In this method, after data processing, the data segments composed of real-time capacitance voltage and current data are input into the trained XG-Boost multi-classification model for fault diagnosis without manually setting the empirical threshold, as is required in traditional change-of-rate-of-voltage-based diagnosis methods. This method has excellent robustness against external noise and missing data while better accuracy, higher speed and lower real-time calculation cost than existing ML-based methods are achieved. The effectiveness of the proposed method is verified by experiment results. It can diagnose the fault within 20ms with accuracy as high as 99.6%, even under the interference of Gaussian (GS) noise and the capacitor voltage data missing.