Research And Development Of 220kV Substation Remote Inspection System Based On Intelligent Image Recognition

With the continuous development of economy and society,the scale of power grid is increasing,the traditional substation operation and maintenance methods are constantly challenged.The difficulties and risk factors of operation and maintenance are increasing.The construction and popularization of intelligent substations have become the development trend of building smart grid.Replacing the manual inspection with unmanned intelligent inspection is an important part.Therefore,the Southern Power Grid Company has proposed the “Southern China Power Grid Smart Technology In The Field Of Production Technology Application Route Program”.Under the guidance of this program,Foshan Power Supply Bureau of Guangdong Power Grid Company is building demonstration areas,including data integration,unmanned inspection,operation procedure and intelligent security.As a basic project of the demonstration substation with integrated application of intelligent technology built by Foshan Power Supply Bureau,this thesis develops a remote inspection system of substation with image recognition.The developed system aims to apply the unmanned inspection and programmed operation.Meanwhile system can help deal with the problems of bad weather and light pollution in the work.The thesis focuses on the identification algorithm of meter reading and the equipment status.The difficulties and technical solutions of meter reading recognition are analyzed.Using SF6 air pressure meter recognition as an example,the implementation methods of template making,template matching based on SURF features,pointer recognition and meter reading are studied.The algorithm has good robustness to the changes of the size of the meter and the rotation angle.At the same time,the implementation scheme is highly portable.According to the characteristics of different meters,it can formulate pointer screening methods which can help us to recognize data.After analyzing the difficulties of equipment status recognition and the shortcomings of commonly used target detection algorithms,an efficient method is proposed.Two-step equipment status recognition scheme is presented,which based on Mobile Net-v3 lightweight network learning.The methods of sample image preparation and expansion,convolutional neural network designation,and classification model training are studied.The above algorithm has been empowered to Songxia Operation and Inspection Center.After testing,the meter reading recognition algorithm consumes 300 ms each time and the accuracy rate is 96.5%.Meanwhile the equipment status recognition algorithm consumes 5-6ms on average,and the accuracy rate is 98.99%.The remote inspection system was tested for100 days in the Songxia Operation and Inspection Center for its efficiency and stability.The test results show that the system runs stably and meets the engineering requirements of actual remote inspection.It is expected to replace 25.5% of manual operations.As a comprehensive demonstration application of intelligent technology,the remote inspection system developed in this thesis has high scalability and can be applied to other mature substations.