中文摘要：

针对NPC三电平逆变器主电路故障难以诊断问题，提出一种基于极限学习机、小波技术和规则推理的二级故障诊断方法；方法通过对比分析逆变器输出电压信号和电流信号在不同故障时差异性，确定了逆变器故障模式和可区分性；根据故障模式和可区分性，提出采用小波技术从输出电流中提取故障低频小波能量作为故障诊断特征和采用极限学习机进行故障电流信号的初级分类；对于根据电流信号的初级诊断不可区分的故障，提出采用逆变器桥臂输出电压信号特征和规则推理法进行故障二级精确诊断；该方法充分利用了输出电压和电流信号区分故障的特点，不仅能区分NPC三电平逆变器主电路单故障，还能同时区分多故障；诊断实验表明，所提方法故障诊断速度快，准确率高，鲁棒性强。

英文摘要：

Aiming to the difficulty of faults diagnosis of power devices in NPC converters, a new fault diagnosis method based on extreme learning machine （ELM）, wavelet techniques and rules inference （RI） theories for NPC inverter is proposed in this paper. By comparing the waveforms of output current and output voltage of the NPC inverter on different faults, the fault modes and its distinguishability with current output and output voltage of the NPC inverter are obtained in this paper. According to the distinguishability and fault modes, the primary di- agnosis method is proposed by using the low frequency features extracted from output current with wavelet technique and by using the ex- treme learning machine for primary classification. If the primary result is the current undistinguishable fault, the rule inference based on in- formation of bridge voltage is proposed to accurately locate the fault components, which is the second level diagnosis. The proposed method takes full distinguishable advantages of output current and output voltage of the NPC inverter, and can not only diagnosis and detect the sin gle fault, but also the multi--fault patterns. The examples shows that the method is effective, fast and robust.