Research On Fault Diagnosis And Fault Tolerant Methods For Cascaded Multilevel Converter

Cascade multilevel converter is widely used in the DC power transmission, energy conversion with different power and voltage for its modularization and easy extension. However, cascade multilevel converter has a great of power electronic devices and often operates in high voltage, large current and high temperature, and is prone to failure. The technology of fault diagnosis and tolerant for cascade multilevel converter have great significance to maintain system rapidly, improve reliability, and reduce maintenance cost.Based on the deep analysis of cascade multilevel topology and modulation strategies, failure mode and distinguishable ability of the inverter is analyzed by establishing the fault model and simulation with the example of cascade seven-level inverter in this paper. In order to effectively extract the fault features from output terminal, the output voltage signal of cascade multilevel inverter is decomposed by multi-scale wavelet, and the energy in the low and high frequency are extracted as the features. Fault samples database is constructed by using simulation.In order to improve the speed, accuracy and intelligentialization of fault diagnosis, a novel approach based on wavelet transform and extreme neural network is proposed. The single hidden layer neural network classifier is designed and is trained by the extreme learning algorithm to obtain the network parameters and the classifier. The test results show that the proposed method based on extreme learning machine and wavelet is feasible. In order to illustrate the performace, the proposed method is compared with conventional BP network method and shows that the proposed method has less training time, more generalization and robust performances than conventional BP network method.In order to improve the reliability of cascade multilevel inverter, considering the modulation strategy and structure selection for converters, the fault tolerance issues of cascade multilevel converters(CMLC) is studied in the paper. A new fault-tolerant method is proposed based on carried modulation by varying the amplitude and phase of modulation signal. The reliability model of the modules is built as the k/n(G) mode, and the k/n(G) model is built for the fault-tolerant converter after. According to the k/n(G) model, the reliability of the fault-tolerant converters is calculated and a structural optimization method is proposed. At last, the feasibility of the proposed converters is validated by examples.