基于Kohonen网络的典型绝缘缺陷局部放电模式识别

Partial discharge recognition of typical insulation defect based on Kohonen network

针对变电站环境局部放电识别面临的不可控干扰多、初始参数确定难的问题,提出将具有自组织竞争识别、抗干扰性强的Kohonen网络用于干扰环境下绝缘缺陷的局部放电识别。首先,通过设计变电站放电典型绝缘缺陷获取多个样本数据,提取统计特征用于Kohonen网络训练。其次,分析Kohonen网络竞争层节点数对识别效果的影响,得出针对样本数据的最佳识别参数。最后,将Kohonen网络与常用的模式识别算法在同等条件下进行对比,验证其在多种放电识别时具有高稳定性与高识别率,及用于变电站环境局部放电识别的优越性。

Aim at the problem that the uncontrollable interference faced by the partial discharge identification in substation,and the initial parameters of the existing identification method are difficult to determine. Design defects that meet the discharge characteristics of the substation. Multiple sample datas are collected combined with the statistical characteristic parameters extraction method. Based on the Kohonen network with self-organizing competition recognition and strong anti-interference characteristics,new method suitable for partial discharge identification in substation is presented. By exploring the influence of the Kohonen network’s parameters on its recognition effect,the recognition effect is optimized. Then by comparing the network with the commonly used pattern recognition algorithm under the same conditions,high stability and high recognition rate of Kohonen network are proved,and excellent performance in partial discharge identification of substation is verified.