基于WPD-PCA与GA-BP的MMC子模块故障定位方法研究

Fault location of MMC sub-module based on WPD-PCA and GA-BP

针对模块化多电平换流器(MMC)结构复杂、IGBT开路故障定位困难问题,提出一种基于小波包分解(WPD)与主成分分析(PCA)的数据压缩与降噪算法以及一种基于遗传算法(GA)和BP神经网络的故障定位算法;对比分析IGBT开路故障状态下的3种典型参量,选取子模块电容电压作为故障参量,并采用所提算法进行数据处理;提出双标签编码方式,进行故障定位;采用CPS-SPWM调制策略,构建模块化五电平换流器仿真模型,仿真结果表明,发生故障时,故障子模块节点能量明显高于其他子模块;故障特征相量经PCA降维后的前7个主成分累计贡献率可达99%;提出的WPD-PCA联合GA-BP定位算法与BP、SSAE-SOFTMAX、WPD-PNN、WPD-BP、FFT-PCA-BP 5种算法进行对比,测试准确率可达到100%,且抗噪声性能优良;提出的多标签编码方式减小网络训练难度,提高数据利用率;该算法能对故障桥臂、故障子模块以及IGBT进行精准定位。

Aiming at the complex structure of modular multilevel converter(MMC)and the difficulty open-circuit fault location of IGBT,a data compression and noise reduction algorithm based on wavelet packet decomposition(WPD)and principal component analysis(PCA)and a fault location algorithm based on genetic algorithm(GA)and BP neural network were proposed.Three typical parameters under the open-circuit fault of IGBT were analyzed,and the capacitor voltage of the sub-module was selected as the fault parameter to be used to deal with by the proposed algorithm.Constructed a simulation model of modularized five-level converter by the CPS-SPWM modulation strategy,and the results show that the node energy of the fault sub-module is significantly higher than that of other sub-modules when a open-circuit fault occurs.After PCA dimension reduction,the cumulative contribution rate of the first 7 principal components can reach 99%.Compared WPD-PCA combined GA-BP algorithm with BP,SSAE-SOFTMAX,WPD-PNN,WPD-BP,FFT-PCA-BP and WPD-BP,the result shows that the test accuracy can reach 100%,and the anti-noise performance is excellent.Multi-label coding can reduce the difficulty of network training and improve the utilization of data.The proposed method can accurately locate the fault bridge arm,fault submodule and fault IGBT.