Fault diagnosis and location of independent sub-module of three-phase MMC based on the optimal deep BiD-LSTM networks

Abstract

Modular multilevel converter (MMC), which is highly modular, easy to be expanded, and has good output voltage waveform, is especially suitable for medium and high voltage power system applications. The MMC bridge arm inductance fault and the independent sub-module fault have the greatest influence on the transmission quality of transmission systems. However, there is few research on fault diagnosis and location for multiple sub-modules(SM) in a single arm at present. A deep bidirectional Long Short Term Memory (Bid-LSTM) network is proposed in this paper, which is used to diagnose and locate the faults of different sub-modules. which can deal well with the prediction problem with sequence dependence. In addition, because the Bid-LSTM model contains many hyperparameters, which seriously affects the prediction performance of BiD-LSTM, an optimal fruit fly optimization (FOA) algorithm based on the composite chaotic mapping of 3-dimensional Logistic-Sine (3d-CCFOA), which searches the optimal value of related hyperparameters to ensure the accuracy of classification. 3d-CCFOA has good chaos characteristics, which enables FOA to avoid falling into local optimal solution effectively and obtains an effective global optimal solution. Then, the MMC simulation system of RT-TAB is used to simulate the open circuit faults of different independent sub-modules of one bridge arm and collect relevant fault data. The analysis process and experimental results adequately indicate that this method can effectively and accurately diagnose and locate the faults of independent sub-modules. Meanwhile, this method also provides a reference diagnosis strategy for multi-bridge arm fault diagnosis and multi-sub-module location.