Font Size: a A A

Study On Automatic Meter Readingof Substation Inspection Robot

Posted on:2022-04-12Degree:MasterType:Thesis
Country:ChinaCandidate:H B ShiFull Text:PDF
GTID:2518306737497874Subject:Control theory and control engineering
Abstract/Summary:PDF Full Text Request
To realize the intelligent inspection of the pointer and digital instruments,and promote the intelligent developments of substation,it is expected to utilize the inspection robots to conduct the automatic inspections of instrument.As such,a inspection robot-based automatic meter reading system is designed,which can be applied to perform the automatic meter reading and data recording of pointer and digital instruments for substation.The specific research work mainly includes the following four aspects.(1)The location of the target instrument is studied,where the RFID technology is applied to search the instrument area,and a lightweight YOLO-v4 network based on the bneck-m model is proposed to detect and locate the instruments.On the basis of the original YOLO-v4 model,by combing with the improved deep separable convolution block(i.e.,bneck-m),a lightweight YOLO-v4 model is proposed.Then,the instrument images are captured from the actual substation environment,which are further expanded by using the Albumentations framework,thus constructing a substation instrument detection data set for the training and testing of our lightweight YOLO-v4 model.Experimental results illustrate that the m AP(mean Average Precision)value and the number of frames processed per second(FPS)of the proposed model are respectively 99.75% and 35.6 frames per second.Compared with the original YOLO-v4 model,the m AP value of the proposed model is only reduced by 0.05%,while the FPS is improved by 11.6 frames per second.Therefore,the proposed lightweight YOLO-v4 model can achieve the real-time detection and location of instruments.(2)The pointer meter reading recognition is analyzed.In particular,the reading recognition algorithm based on key point detection is used to recognize the reading of pointer instrument for substation.This method first applies Gamma transformation to enhance the pointer instrument images and utilizes the perspective transformation to correct these instrument images.After that,the pointer instrument images are captured and expanded to build a pointer key point data set,and by considering the ability of Center Net attitude estimation,the Center Net pointer Key point detection model is trained for the location of the pointer in the instrument images.Finally,the reading of the pointer instrument is recognized by using the angle method.Based on the above analysis,this proposed method can accurately recognize the pointer instruments in substation.(3)Digital meter reading recognition is completed.A lightweight YOLO-v4 model is proposed to identify the digital meter readings of substation.Firstly,by capturing actual digital instrument images from the real substation,Gamma transformation is applied to enhance these images,which are further expanded to build the digital display area location and character recognition data sets.Then,combining with the characteristics of the convolutional neural network,the lightweight YOLO-v4 model is trained to realize the positioning of the display area of the digital instrument and the recognition of digital characters,and the sequential output character classification is carried out to complete the digital instrument reading recognition.Based on the above data sets,our lightweight YOLOv4 model can accurately identify digital meter readings,whose recognition accuracy is97.5%.(4)A inspection robot-based automatic meter reading system is designed for substation.This system contains three parts,such as software system client,database,and data processing terminal,which are respectively applied to complete the instrument detection and recognition,instrument data recording,and real-time display of instrument data for substation.The experimental results conducted in the actual environment demonstrate that the designed automatic meter reading system can accurately identify and display the readings of instruments in real time.The proposed target detection and positioning method improves the location accuracy and real-time performance of instruments for substation,and the pointer and digital instrument recognition algorithms effectively improve the accuracy of the instrument recognition,thus making it possible for substation to be unattended and laying a certain foundation for accurate recognition of the pointer and digital instruments.
Keywords/Search Tags:Instrument detection, Reading recognition, Deep learning, YOLO-v4
PDF Full Text Request
Related items