Local outlier factor based condition monitoring for sub-module capacitors in modular multilevel converters

Abstract

Modular multilevel converters (MMC) as a key technology for future power grids is rapidly developing. The failure rate of sub-module (SM) capacitors is only lower than that of power devices in MMC, so accurate evaluation of the reliability of capacitors is crucial for the safe operation of MMC. Both the increase of equivalent series resistance (ESR) and decrease of equivalent series capacitance (ESC) are features for capacitor deteriorating. ESC or ESR of the capacitor reaching to limited values indicates that the end of its lifespan and it must be replaced. Most of contemporary researches only focus on ESC or ESR. This article considering the characteristics of both ESC and ESR, proposes a method for monitoring the status of SM capacitors based on machine learning algorithm: local outlier factor (LOF). By utilizing the existing SM switching function and capacitor voltage signals within a fundamental frequency cycle in the control system of MMC, the proposed strategy can not only monitor the faulty capacitors, but also can sort the degrees of abnormal states of capacitors, and the strategy is applicable to MMC with single and multiple types of SM topologies. The proposed method is verified by the hybrid MMC model on Matlab/Simulink.