基于最小二乘加权融合集成神经网络的电力变压器故障识别  

POWER TRANSFORMER FAULT DIAGNOSIS BASED ON COMBINING NEURAL NETWORK WITH LEAST SQUARE WEIGHTED FUSION ALGORITHM

作　　者：吕干云[1] 董立新[1] 程浩忠[1] 

机构地区：[1]上海交通大学电气工程系,上海200030

出　　处：《电网技术》2004年第16期52-55,共4页Power System Technology

摘　　要：摘要:提出了一种基于最小二乘加权融合集成神经网络的变压器故障识别新方法。首先对色谱分析法检测到的特征气体含量进行数值预处理,提取出故障识别所需的6个特征量,再应用5个不同结构的BP子网络分别进行识别,接着运用最小二乘加权融合算法对各个子网络的识别结果进行信息融合,最后根据融合结果来识别故障。与单个神经网络识别方法相比,该最小二乘加权融合集成神经网络可在故障特征比较类似的情况下,正确识别故障类型,且该方法的识别结果具有更大的安全间隔空间、可靠性更高。测试结果也表明了这些特征。A new method to diagnose power transformer faults based on the combination of neural network with least square weighted fusion algorithm is presented. Firstly, the numerical preprocessing to the contents of five characteristic gases obtained by chromatography is performed and the six characteristic quantities which are necessary to fault diagnosis are abstracted, and five back-propagation(BP) artificial neural networks(ANN) with different structures are applied to identify respectively, then the information fusion of the identified results from the subnetworks is carried out by use of least square weighted fusion algorithm. Finally, according to the situation of the fusion the fault is diagnosed. Compared with the diagnosis method based on single neural network, the presented method can correctly determine the type of the fault under the similar conditions, the obtained diagnosis result possesses wider safe interval and is more reliable. The testing results of the presented method prove above mentioned advantages of this method.

关 键 词：电力变压器 故障识别 最小二乘加权融合 集成 神经网络 色谱分析 

分 类 号：TM41[电气工程—电器]