This paper presents a fault diagnosis method and system reconfiguration strategy of three level neutral-pointclamped cascaded inverter (3L-NPC-CI) to enhance reliability of power supply in electrified railway. First, neural network is employed to perform open-circuit fault diagnosis and identify fault switches. Then, system reconfiguration strategy is proposed to perform single phase 3L-NPC-CI reconfiguration. Through the analysis of the operation status of 3L-NPC, there are eight fault modes in single bridge arm, including single-switch fault and double-switches fault. By the feature analysis for the output arm voltage harmonic of normal and eight fault modes, seven harmonic parameters are selected as fault feature vectors. Meanwhile, a three-layer neural network is constructed, and the seven feature vectors are the input layer of the neural network. By training algorithm, the fault switches location can be identified accurately about one modulation period. Then, the fault module is bypassed and the 3L-NPC-CI system is stable by reconfiguring the modulation strategy of other normal modular. Simulation and experimental results are given to validate the proposed fault diagnosis and system reconfiguration strategy. The proposed method is able to improve robustness and reliability of 3L-NPCCI in traction power supply system.

An advanced traction power supply system based on a single phase neutral-point-clamped (NPC) cascaded inverter is studied. The big triangular carrier equivalence method in double coordinate system is proposed, which can reduce one coordinate system, thus simplifying the calculation. Based on the big triangular carrier equivalence method, the harmonic characteristics of a single phase NPC cascaded inverter are calculated by double Fourier transform and voltage harmonics expressions of 5-level, 9-level and 13-level output waveforms are derived. Finally, the performance and calculated results of the proposed method were verified by simulations and experiments. The result provides a theoretical basis for further studies on traction network resonance.