Substation Human-machine Status Monitoring System Based On Deep Learning

With the rapid development of China’s communication technology and the rapid spread of 5G network,IoT technology is playing an increasingly important role in urbanization.The future trend of IoT technology is to integrate data mining analysis technology and artificial intelligence algorithm into IoT technology to complete the intelligent application of IoT data.As essential components of urbanization,power facilities provide energy and information support for the development of urbanization.In the event of a power facilities failure,the equipments need to be tested and repaired by operations and maintenance personnel.The standardization of the personnel equipments and the operational processes will directly affect the operation and maintenance efficiency and personnel safety factor.This thesis takes electric cabinet equipment and operation and maintenance personnel as the research object,and design a deep learning-based IoT platform architecture for electric cabinet equipment to address the problems of low maintenance efficiency.This platform mainly focuses on two application scenarios: switch operation status of electric cabinet and equipment status monitoring of operation and maintenance personnel,and investigates the equipment operation status monitoring algorithm integrating LSD(Line Segment Detector)and deep learning,and the personnel equipment status monitoring algorithm optimizing deep learning,the main research contents and innovation points are as follows.(1)In order to achieve the intelligent monitoring of power facilities operation status and equipment status of operation and maintenance personnel,this thesis designs a deep learningoriented substation IoT platform by integrating artificial intelligence models and algorithms.Combining the characteristics of power facilities environment,the difficulties of image-based human-equipment monitoring algorithm are analyzed.The comparison of detection real-time and accuracy of target detection algorithms is completed.(2)To solve the problems of high maintenance costs of equipment in complex industrial environments,and the long implementation cycle of machine vision,this paper proposes a method to detect the switching status of equipment under BIM(Building Information Modeling)environment by integrating LSD linear detection and deep learning based on the advantages of BIM.BIM technology has the advantage of being consistent with the real scene space and can simulate various lighting conditions.This thesis generates image data sets by using the LSD algorithm to quickly detect the target of the electrical cabinet switch,solving the problem that deep learning image data is difficult to generate quickly,and then inputs the image data sets into convolutional neural network to train the target detection model of electrical cabinet switch status,realizing intelligent monitoring of the working status of electric facilities and solving the problem that the traditional target detection model cannot combine accuracy and real-time.The experiments results show that the proposed algorithm is well adapted to monitor the operating status of the electrical cabinet equipment in the field.(3)To solve the problem of operation and maintenance personnel work status monitoring,this thesis proposes an operation and maintenance personnel work status monitoring model based on the optimized YOLOv3(You Only Look Once version3)network.Aiming at the problems that the operation and maintenance personnel have different imaging,multiple personnel overlapping,equipment obscuring,and equipped with clothing easily confused in the monitoring image data,this thesis uses the DBSCAN clustering algorithm to strengthen the training ability of neural network for difficult samples.For the easily confused clothing equipped situation,the Focal Loss loss function is introduced instead of Sigmod function.This loss function optimizes the target detection ability of deep learning model.This method realizes intelligent monitoring of personnel equipment status.This thesis designs a deep learning-based substation man-equipment status monitoring system.It realizes real-time monitoring of the switch status of electrical cabinets and personnel staffing status.The optimized monitoring method improves power facilities maintenance efficiency and reduces the equipment operation and maintenance costs.This monitoring system can be further extended to the field of intelligent monitoring of equipment and personnel status.