基于新径向基函数网络的变压器故障诊断法

Fault Diagnosis of Power Transformer by Novel Radial Basis Function Neural Network Approach

油中溶解气体分析(DGA)是判别变压器内部绝缘状况及发现内部潜伏性故障的重要手段,而多层前馈网络(MLFNN)是应用广泛的故障诊断模型。为此,提出了以DGA数据为特征参数的新型径向基函数神经网络(RBFNN)诊断变压器故障。在分析传统k-均值聚类算法RBFNN的缺点和最优聚类特性的基础上,介绍了RBFNN的新算法-自适应k-均值聚类算法,它既能避免传统k-均值聚类算法的局部收敛的缺点,又能动态调整学习率。最后,大量聚类实验结果显示自适应k-均值聚类算法在收敛速度和聚类性能上比传统k-均值聚类算法显著提高;故障诊断实验结果显示所提出的模型故障诊断准确度高于传统BPNN、RBFNN及IEC三比值法。

Dissolved gas analysis （DGA） is an important method to judge the insulation condition and find out the potential faults inside the oil-immersed transformer, and the Multi-Layer Feed-forward Neural Network （MLFNN） is a widely used model in transformer fault diagnosis. In this paper, the application of Radial Basis Function Neural Network （RBFNN）, which uses the DGA data as the characteristic parameters to fault diagnose of the power transformer, is presented. The drawbacks of RBFNN based on the traditional k-means clustering algorithm and the mechanism of optimal cluster are analyzed. Moreover, a new adaptive k-means clustering algorithm is put forward, which can overcome the traditional k-means clustering algorithm＇s shortcomings of local convergence and dynamically adjust the learning rate based on the quality of the current clustering. This adaptive k-means clustering algorithm is based on the optimality criterion for the k-means partition, all the regions＇in an optimal k-means partition have the same variations if the number of regions in the partition is large and the underlying distribution for generating input patterns is smooth. The goal of equalizing these variations is introduced in the competitive function that assigns each new pattern vector to the ＂appropriate＂ region. Quantities of fault samples are analyzed by this method. Considering the influence of transformer form, capacity and working environment, the transformer DGA test is recorded and corresponding fault results of different manufactories and different regions are widely collected. Among them, 72 representative samples are selected as training samples, other 60 samples are used for recognition. The results of clustering test show that the new adaptive k-means clustering algorithm outperforms the traditional k-means clustering algorithm both in clustering performance and convergence rate. The results of verification show this method has higher diagnosis precision than the Back Propagation Neural Network（BPNN）.