基于稀疏深度森林的调相机轻微定子匝间短路故障诊断  

Fault Diagnosis of Slight Stator Inter-turn Short Circuit of Synchronous Compensator Based on Sparse Deep Forest

作　　者：张玉良 马宏忠[1] 朱昊 颜锦 蒋梦瑶 林元棣 ZHANG Yuliang;MA Hongzhong;ZHU Hao;YAN Jin;JIANG Mengyao;LIN Yuandi(College of Energy and Electrical Engineering,Hohai University,Nanjing,211100,China;State Grid Jiangsu Electric Power Co.,Ltd.,Nanjing 211103,China)

机构地区：[1]河海大学能源与电气学院,南京211100 [2]国网江苏省电力有限公司电力科学研究院,南京211103

出　　处：《高电压技术》2022年第5期1875-1883,共9页High Voltage Engineering

基　　金：国家自然科学基金(51577050);国网江苏省电力公司科技项目(J2019114);111引智计划(B14022)。

摘　　要：针对调相机轻微定子匝间短路故障难以辨识的问题,提出了一种基于稀疏深度森林的调相机轻微定子匝间短路故障诊断新方法。该方法首先提取将定子电流的1、3、5、7次谐波幅值作为预特征,再利用稀疏滤波网络将特征信息增强,然后在深度森林中引入XGBoost和逻辑回归的基学习器,在提高单一学习模型分类能力的同时,缩短了运行时间,并在级联森林部分加入决定系数以平衡运算效率与模型精度,选取级联层最后一层的平均值作为输出,最后,采用改进后的深度森林和稀疏滤波构建故障诊断模型。通过定制同步电机模拟调相机轻微定子匝间短路故障实验来验证所提方法,实验结果表明,该方法提取的特征能有效反映定子匝间短路故障,且具有较高的诊断精度与效率。Aiming at the difficulty in identifying the inter-turn short circuit fault of the slight stator of the synchronous compensator,we proposed a new method for fault diagnosis of the slight stator inter-turn short circuit of the synchronous compensator based on sparse filtering deep forest.The harmonic amplitude characteristics at the 1,3,5,and 7 times of the stator current were firstly extracted.The characteristic information of sparse filtering network was adopted to strengthen characteristic information,and the XGBoost and logistic regression base learners were introduced in the deep forest.Moreover,while enhancing the capacity of a single learning model classification,the running time was reduced,and the cascade forest part decision coefficient was joined in to balance the operation efficiency and the precision of model.Then,the average of the last layer of cascading layer was selected as the output.Finally,the improved deep forest and sparse filtering was used to construct fault diagnosis model.The proposed method is verified by a custom synchronous motor to simulate a slight inter-turn short circuit fault of the stator.Experimental results show that the features extracted by the proposed method can effectively reflect the stator interturn short circuit fault,and it has high diagnostic accuracy and efficiency.

关 键 词：调相机 定子匝间短路故障 定子电流 稀疏滤波 稀疏深度森林 特征提取 故障诊断 

分 类 号：TM721.1[电气工程—电力系统及自动化] TP181[自动化与计算机技术—控制理论与控制工程]