基于改进支持向量机的MMC-MG系统桥臂故障诊断

模块化多电平换流器(modular multilevel converter, MMC)半桥串联结构微电网系统桥臂中各发电模块通过串联方式连接，其投入和切除由半桥变流器(half-bridge converter,?HC)中绝缘栅双极型晶体管(insulated gate bipolar transistor, IGBT)的开通与关断来实现。而该系统在并网双闭环控制下，若桥臂中HC及其连接线路发生故障，会对系统的输出特性造成一定影响。为此，分析了HC中IGBT与其反并联二极管发生开路或短路故障，以及HC之间的连接线路发生开路故障时，桥臂输出电压电流、相间环流、并网电流等参数的变化情况。选取异常变化明显的参数作为特征属性，并用其构造样本数据集。另外，在系统桥臂的故障诊断中，针对采用传统支持向量机(support vector machine, SVM)时其准确率较低的问题，建立基于鲸鱼改进SVM的故障诊断模型。结合不同数据集，通过仿真实验对所建模型的有效性进行验证。结果表明：与传统SVM和BP神经网络算法相比，基于鲸鱼改进SVM的故障桥臂诊断方法准确率更高。

Bridge arm fault diagnosis of the MMC-MG system based on an improved support vector machine

The power generation modules in the bridge arm of the modular multilevel converter (MMC) half-bridge series microgrid system are connected in series, and its input and removal are realized by turning on and off the insulated gate bipolar transistor (IGBT) in the half-bridge converter (HC). However, under the grid-connected double closed-loop control of the system, if a fault occurs in the HC in the bridge arm and its connecting lines, it will have a certain impact on the output characteristics of the system. For this reason, the changes of the bridge arm output voltage and current, inter-phase circulating current, grid-connected current and other parameters are analyzed when the IGBT and its anti-parallel diode in the HC has an open-circuit or short-circuit fault, and the connection line between the HC''s has an open-circuit fault. Parameters with obvious abnormal changes are selected as feature attributes and used to construct a sample data set. In addition, in the fault diagnosis of the system bridge arm, a fault diagnosis model based on a whale-optimized support vector machine (WO-SVM) is established to solve the problem of low accuracy when using traditional SVM. Combined with different data sets, the validity of the model is verified through simulation. The results show that compared with the traditional SVM and BP neural network algorithms, the fault bridge arm diagnosis method based on the improved SVM has higher accuracy.