人工神经网络在电力设备红外测温中的应用研究

Research on the Application of Artificial Neural Networks in Infrared Temperature Measurement of Power Equipment

变电站中电力设备发热会对电网运行造成很大隐患,极大地降低电能质量和供电可靠性,因此有必要在变电站常规巡检中监测变压器、高压开关柜、绝缘子、导线接触点的温度,以确保变电站电气设备正常稳定运行。传统的变电站温度采集方式极为费时、费力,并且由于相间及相地电压非常高,使得传统测温方法对人员的安全也有很大的威胁,而且也容易误检、漏检,容易造成人员和资源的浪费。针对这此现象开展无人机巡检红外照片自动测温是保障电气设备稳定运行的有效措施,如何快速且自动地识别无人机红外巡检照片中的设备温度异常点是一个亟待解决的问题。本文提出一种通过人工神经网络近似红外图像像素RGB值到摄氏温度值映射关系的方法,该方法将红外图像像素点的RGB值作为人工神经网络的输入,网络的输出为摄氏温度值。我们使用变电站常规巡检中获取的红外图像为数据源,对具有三个隐藏层的全连接人工神经网络进行训练,测试结果显示该人工神经网络对塔材和绝缘子的拟合程度较好偏差值小于1℃,对树木和天空的拟合能力较差偏差大于1℃。

Heating of power equipment in substations can pose significant risks to the operation of the power grid,greatly reducing power quality and supply reliability.Therefore,it is necessary to monitor the temperature of transformers,high-voltage switchgear,insulators,and wire contact points during routine inspections of substations to ensure the normal and stable operation of substation electrical equipment.The traditional method of temperature collection in substations is extremely time-consuming and laborintensive,and due to the high voltage between phases and ground,the traditional temperature measurement method also poses a great threat to personnel safety.It is also prone to or missed detections,which can cause waste of personnel and resources.To address this phenomenon,conducting unmanned aerial vehicle inspection infrared photo automatic temperature measurement is an effective measure to ensure the stable operation of electrical equipment.How to quickly and automatically identify equipment temperature anomalies in unmanned aerial vehicle infrared inspection photos is an urgent problem to be solved.This article proposes a method of approximating the mapping relationship between RGB values of infrared image pixels and Celsius temperature values through an artificial neural network.The method takes the RGB values of infrared image pixels as the input of the artificial neural network,and the output of the network is the Celsius temperature value.We used infrared images obtained from routine inspections of substations as data sources to train a fully connected artificial neural network with three hidden layers.The test results showed that the artificial neural network had a good fitting degree for tower materials and insulators,with a deviation value of less than 1℃,and a poor fitting ability for trees and the sky with a deviation value of more than 1℃.