Fault Diagnosis and Tolerance Control of Five-Level Nested NPP Converter Using Wavelet Packet and LSTM

Abstract

The five-level nested neutral-point-pilot (NPP) topology, as a new structure for converters, bears the advantages of a high power density, robustness, and flexibility and is therefore suitable for high-voltage and high-power applications. For a multilevel converter, as the number of power electronic switches increases, the risk of switch failure increases, together with the complexity of fault detection and tolerance control. The requirements for a higher operational stability and reliability continue to grow. However, studies on fault tolerance for multilevel converters are limited. In this paper, a fault diagnosis and tolerance solution for a five-level nested NPP converter is proposed. For the fault diagnosis, a deep learning method integrating the wavelet packet transform and long short-term memory is presented. Both open- and short-circuit switch failures can be precisely detected and located without the requirement of a large sample set. Two software-based control strategies for fault tolerance are adopted, and low-cost hardware reconfigurations are also implemented to prevent failure expansion and ensure continuous operation. Furthermore, the voltages of dc-link capacitors and flying capacitors are effectively balanced with the improved algorithm even when a failure occurs. Finally, the effectiveness of the fault-tolerant strategy are proven by simulations and experiments.