基于红外图像特征与BP神经网络的绝缘子串低零值故障和污秽故障识别方法

A Detection Method of Low/Zero Fault and Contamination Fault of Insulator String Based on Infrared Image Feature and BP Neural Network

提出一种基于红外图像温度分布特征和BP神经网络(BPNN, back-propagation neural networks)的绝缘子串低零值故障和污秽故障识别方法。首先利用图像处理技术分割提取绝缘子串红外图像中钢帽和盘面目标区域,得到对应温度数据;之后引入K-means聚类算法剔除分割目标区域中背景像素温度数据的干扰,并计算每个分割区域温度平均值,形成反映绝缘子运行状态的钢帽和盘面温度特征向量;在此基础上,建立以温度特征向量为输入的BPNN模型,实现绝缘子串低零值故障和污秽故障的识别及故障定位。最后通过将模型应用于某500 kV变电站绝缘子串故障诊断,验证所提出方法的准确性。

A low/zero fault and contamination fault detection method based on infrared image temperature distribution and BP neural network is proposed in this paper. Firstly, the image processing technique is used to segment the steel cap and the disk area of insulator string infrared image, and the corresponding temperature data are obtained. Then the K-means clustering algorithm is used to eliminate the interference of the background pixel temperature data in the segmentation target region and the average temperature of each divided area is calculated to form the steel cap and the disk surface temperature characteristic vector which reflects the operation state of the insulator. On this basis, a BP neural network model with temperature characteristic vector input is established to realize the detection and fault location of zero/low value fault and contamination fault of insulator string.Finally. the model is applied to the fault diagnosis of insulator string in a 500 kV substation to verify the accuracy of the proposed method.