Design And Research Of Manual Misoperation Alarm System For Switchgear Based On Deep Learning

Switchgear is an important equipment in the distribution network,which realizes the on-off control of the circuit,and can be used in many occasions such as industry,transportation,and power stations.During the operation of the switchgear,safe operation is a very important part.Once a misoperation event occurs,it will cause serious consequences.Therefore,the prevention of misoperation of the switchgear is of great significance.In order to further improve the safety of the switchgear operation,the paper designs and implements a manual misoperation alarm system for the switchgear.The manual operation intention detection unit is used to detect the operator’s operation intention,and the alarm judgment output unit detects the operating state of the switchgear,judge whether the manual operation intention is a misoperation,and give voice prompts before the misoperation is performed to prevent misoperation accidents.The main research content of this article includes the following parts.(1)Analyze the functional requirements of the system,complete the overall design of the manual misoperation alarm system for the switchgear.Through the research of manual operation intention detection method,a manual operation intention detection method based on target detection algorithm is proposed.Analyze the operating specifications of the switchgear,design the alarm output logic table.(2)Selection and embedded deployment for Manual operation intention detection algorithm.The operation images of the switchgeart were collected through the USB camera,and a training data set was produced to train for the four algorithms of HOG+SVM,Faster R-CNN,YOLOv3 and YOLOv4.Comparison for model detection accuracy and detection speed,the YOLO model with high detection accuracy and fast detection speed was selected to be deployed to Jetson Nano embedded devices.In order to improve the detection speed of the YOLO model on the Jetson Nano embedded device,carried out two optimization operations of model pruning and Tensor RT inference acceleration.After model pruning,the detection speed of YOLOv3 is 5FPS faster than YOLOv4,and the detection accuracy is only0.0038 lower.Therefore,the YOLOv3 model is selected for Tensor RT inference acceleration.Through Tensor RT inference acceleration,the detection speed of the YOLOv3 model on the Jetson Nano embedded device reached 25 FPS.(3)System hardware and software design and testing.Through the overall design scheme,the circuit schematic diagram and PCB diagram of the alarm judgment output unit are designed,and the circuit board is soldered.According to the functional requirements of the system and the hardware composition,the programming of the system is completed.In order to test the effectiveness of the manual misoperation alarm system,the manual operation intention detection method test,the alarm judgment output unit test,and the overall system test were carried out.The results show that the manual operation intention detection unit can accurately detect the four manual operation intentions of the operator(open the cable compartment door,open the circuit breaker compartment door,move the circuit breaker handcart,and operate the grounding switch)before the misoperation is performed,and alarm The judgment output unit can make judgments on 16 kinds of misoperations and issue a voice alarm before the misoperation is executed.