Research On Switch State Recognition System Of Substation Switchboard Based On Machine Vision

Unmanned substation is an important link in the process of substation intelligent transformation,and the running state recognition of key electrical cabinet switch equipment is crucial to the safe and stable operation of substation.At present,manual inspection is used to monitor the running state of switch of electric cabinet in substation.This manual way requires the staff to spend a lot of manpower,material resources to the electrical control cabinet for periodic inspection,and check the running state of switch one by one.With the continuous development of electric power network,there are more and more electrical control cabinets in substations.The workload of the staff increases greatly,and it is very easy for some equipment to malfunction or switches to run in wrong state without timely detection,which will result in major power accidents.Aiming at the problem of automatic recognition of switch state of electrical control cabinet in substation,this paper studies the method of recognition of switch state of electrical control cabinet in substation based on machine vision technology.The main research contents are as follows:(1)The switch state recognition system of electrical control cabinet based on machine vision technology is designed.The whole system includes the switch panel image acquisition and switch control information acquisition system,switch position detection and status recognition algorithm,post-recognition result processing and alarm system.The system uses industrial cameras or smart phones to collect switch panel images and the two-dimensional code which saves the switch control information.The images are then transferred to the switch detection and recognition system for state recognition.Finally,the results will be uploaded to the substation background management system for comparison.Early warning if the switch working state is wrong.(2)Aiming at the problem of switch position detection on switch panel,two algorithms are studied and proposed.The first algorithm uses SVM and sliding window technology with the image HOG feature to detect whether the target window contains switches.The second algorithm uses convolutional neural network to extract the features of the input image,and then nominates the target region,and submits the nominated region to the RPN network for target detection and position regression.Through a large number of data sets training,the algorithm can accurately detect the position of the switch.(3)Aiming at the switch state recognition problem on the switch panel,the switch state recognition algorithms based on SVM and convolutional neural network are studied and proposed respectively.Both algorithms transform the problem of target switching state recognition into the task of image classification.SVM based switch state recognition algorithm uses kernel function to map the two-dimensional image of switch to the high-dimensional space and then classifies it.The algorithm based on convolutional neural network extracts the features of the switch by multi-layer convolution operation,and finally outputs the classification results by using the full connection layer.Both algorithms can accurately recognize the state of the switch and adapt to the changeable scene.(4)Proposing a switch position detection and state recognition algorithm based on YOLOv3.This algorithm combines the target switch position detection and state recognition tasks,and is an end-to-end target detection and recognition algorithm.The algorithm can detect the position of the switch and recognize the state with high precision by training with a large scale of sample set,and the algorithm can meet the application requirements of current real-time detection in substations.