【作者】 罗权；

【导师】 刘永强；

【作者基本信息】 华南理工大学 ， 电工理论与新技术， 2021， 硕士

【摘要】 变电站内设备繁多,其中二次设备能够对一次系统起到监视、测量、控制、调节与辅助的作用,电力系统的安全稳定十分依赖于二次设备的可靠工作。而二次设备中尤为重要的是继电保护压板,它是用来投入或退出一次设备对应保护功能的,在保护回路中起接入和断开的作用。变电站日常巡检中有一项工作——继电保护压板状态核对,目前一些变电站尝试通过采用巡检机器人核对代替人工核对压板状态,这可以有效的减少电网检修人员的工作量进而提高电网运行效率,并且可以避免电网检修人员业务不足导致人为失误。但是在轮式巡检机器人识别压板过程中,存在如主摄像头拍摄存在视觉盲区、非正常视觉图像获取环境下压板难以识别等问题。首先,针对主摄像头拍摄存在视觉盲区问题,通过装载下位副广角摄像头加以解决,并进行轮式机器人副摄像头可见区域测试实验,实验结果表明装载的下位副广角摄像头下不存在视觉盲区,并给出了机器人与继电保护柜的最佳拍摄距离为40cm。面对副广角摄像头带来的拍摄图片畸变问题,先简单介绍并实现二元多项式畸变校正算法,通过调整校正算法次数,得出多项式四次校正效果最佳的结论。然后,针对非正常视觉图像获取环境下压板识别问题,先对非正常视觉图像获取环境进行分析,并尝试对非正常视觉图像获取环境进行优化,同时使用传统识别算法对其进行测试。经测试,传统识别算法测试结果显示无法达到识别要求后,为突破传统识别算法限制,本文提出采用卷积神经网络对非正常视觉图像获取环境下的压板识别,选用介绍了Faster R-CNN算法。压板图片预处理后,采用Faster R-CNN压板图像状态识别算法进行实现,并对正常情况、广角畸变校正前后、非正常视觉图像获取环境下的压板图片进行训练,并分析了训练测试结果。最后,基于本文需要解决的是轮式巡检机器人变电站现场巡检识别继电保护压板状态遇到的各类问题,因此搭建好测试环境后进行实地巡检测试,并对测试结果进行分析。通过以上一系列问题分析以及算法实现,解决了主摄像头视觉盲区问题以及非正常视觉图像获取环境下的压板识别问题,达到了变电站压板状态识别要求,同时对巡检机器人在变电站的推广应用有促进作用,对各种巡检机器人拍摄、图像识别问题研究具有一定参考价值。更多还原

【Abstract】 There are many equipment in the power substation,including the secondary equipment which can monitor,measure,control,regulate and assist the primary system.The safety and stability of the power system depends on the reliable work of the secondary equipment.The most important thing in the secondary equipment is the relay protection plate,which is used to input or exit the corresponding protection function of the primary equipment,and plays the role of access and disconnection in the protection circuit.There is one task in the daily inspection of substations--check the state of relay protection plate.At present,some substations try to replace the manual check of plate state by using inspection robot,which can effectively reduce the workload of power grid maintenance personnel and improve the operating efficiency of power grid,and can avoid the human error caused by the lack of service of power grid maintenance personnel.However,in the process of recognizing the platen,there are some problems,such as the visual blind area in the main camera and the difficulty in recognizing the platen in the abnormal visual image acquisition environment.Firstly,the problem of visual blind area in the shooting of the main camera was solved by loading the lower wide-angle camera,and the visible area test of the wheeled robot’s secondary camera was carried out.The experimental results showed that there was no visual blind area under the loaded lower wide-angle camera,and the optimal shooting distance between the robot and the relay protection cabinet was given as 40 cm.Face the vice wide Angle camera shooting distortion problem,to briefly introduce and implement binary polynomial distortion correction algorithm,by adjusting the algorithm degree of polynomial,concluded four times polynomial correction effect is best.Then aiming at the problem of pressure plate recognition in the abnormal visual image acquisition environment,firstly analyzes the abnormal visual image acquisition environment,and tries to optimize the abnormal visual image acquisition environment,and at the same time uses the traditional recognition algorithm to test it.After the optimization effect was poor and the test results of the traditional recognition algorithm showed that it could not satisfy the recognition requirements,the convolutional neural network was proposed to identify the platen in the abnormal visual image acquisition environment,and the Fast-R-CNN algorithm was selected and introduced.Faster R-CNN pressure plate image state recognition algorithm was implemented after the preprocessing of the pressure plate image,and the pressure plate images were trained under normal conditions,before and after wide-angle distortion correction,and under abnormal visual image acquisition environment,then the training results were analyzed.Finally,based on this paper,we need to solve all kinds of problems encountered by wheeled inspection robot in on-site inspection and identification of relay protection plate state in substation.Therefore,field inspection and test are conducted after the test environment is set up,then the test results are analyzed.Through the above a series of problem analysis and algorithm implementation,solve the problem of the main camera visual blind area and abnormal visual image acquisition environment pressure plate recognition problems,meet the requirements of the substation linking piece state recognition,as well as the application in substation inspection robot has a promoting effect,for all kinds of inspection robot shooting,image recognition problem study has certain reference value.更多还原