Research On Accurate Detection Of The State Of Isolator Based On Radar Point Cloud And Image

The safe operation of power system is a significant support for the development of national economy,and the isolator is an extremely important equipment in substation.The operation and status judgment of isolator is related to the normal operation of power grid.Long-term use will lead to abnormal structure of isolator,causing inadequate closing and resulting in serious safety problems.At present,the judgment of closing in place mainly depends on the auxiliary contacts of the isolator itself,but the abnormal transmission part will lead to wrong judgment.Traditionally,it is very inconvenient to send staff to the field for inspection.Relevant power departments have issued notifications that better dual checking methods are urgently needed.The technical difficulties are as follows:(1)There is little difference in appearance between the inadequate closing state and the in-place state;(2)The plan cannot affect the normal operation of other equipment in the substation;(3)There are many interference factors in substation.Most of the current studies on the judgment of the state of the isolator stay on the judgment of the on-off state,ignoring the inadequate state of the isolator,although some methods have mentioned that,they have not yet achieved the expected results.Based on the above,this paper intends to research three typical isolators in the substation:horizontal isolator,vertical isolator and scissors isolator.The image of the isolator and radar point cloud data are combined to detect the accurate state of the isolator in the substation.The main idea of the plan is to use the camera to capture the image data,using the more mature twodimensional image neural network detection model to preliminarily determine the state of the isolator and locate the target,and then use the three-dimensional point cloud detection model for further accurate state detection combined with the point cloud data of the isolator collected by lidar,combined with the scope of the previous positioning and relevant prior information.The main contents of this paper are as follows:1)Due to the lack of public data of the isolator,this paper observed the shape characteristics of three kind of isolators with different states in the field.The appropriate location of the camera is analyzed,and a large number of images and radar point cloud data are collected.After a series of preprocessing,a standardized data set is made.2)In order to preliminarily distinguish the state of the isolator and locate the target,experiments are carried out on the collected data set using multiple two-dimensional image neural network detection models.The experimental results show that the inference speed of YOLO-V5 is much faster than other models on the premise of 95.4% accuracy and 95.1%average pixel recall.The GUI application of the isolator positioning is designed and implemented.3)The study uses the radar point cloud data of the isolator to further discriminate the state of the isolator,and proposes an optimized distance-based point cloud discrete point removal algorithm,which is more capable of detecting discrete point cloud groups;constructs a multifeature ISO-Point Net network structure,and adds Lalonde features and reflectivity features to the original network.The detection accuracy of the knife gate is improved and reaches 89.7%.