Research On Power Artificial Intelligence Image Recognition Technology And Its Application In Overhead Transmission Line Patrol Service

With the vigorous promotion of the global energy internet,the construction of AC and DC power grid in China has developed rapidly.With the continuous innovation in the field of artificial intelligence technology,the use of new technologies such as image recognition to promote the development of power transmission patrol business has become a major issue in the power grid industry in recent years.Intelligent identification of transmission line operation status includes three aspects:intelligentized identification of transmission line operation status,intelligentized identification of transmission line defects and intelligentized identification of transmission line hidden dangers.Traditional pattern recognition technology has been unable to meet the development needs of three-dimensional transmission inspection under the new situation.It is impossible to identify widgets,complex background defects and hidden dangers accurately.Artificial intelligence technology with self-learning ability is urgently needed to identify and analyze the defects of overhead transmission l ine equipment and the abnormalities and risks in the channel environment.This paper focuses on the research of hidden trouble detection and fault location technology of transmission line ontology devices and channel patrol objects based on artificial intelligence image recognition technology.Combined with prior knowledge,a three-dimensional patrol mode of transmission line is proposed.Based on the application mode research,a set of transmission line image data sharing and application demonstration system is designed and developed,which is fast and accurate.Research fault location and defect type,and then carry out timely and effective elimination of defects and potential risks to ensure the safe and stable operation of the power system.The following are the main research contents of this paper:1.Around the basic support technology of artificial intelligence image processing,the design of heterogeneous high-performance computing cluster and deep learning framework integrated optimization technology framework and the research of sample image processing technology are carried out.The main results include the following two points:(1)A development integration scheme based on the deep learning framework is proposed.The integration of samples,models and resources is realized through interface calls of data processing system and deep learning system,and the end-to-end training ecological chain is completed.The in-depth learning system is composed of sample management,model optimization and resource allocation.It can realize the functions of sample feature analysis,image preprocessing,sample classification and statistics,model training monitoring and resource operation monitoring.(2)The data expansion,image defogging and image restoration technologies are studied respectively.The training samples are increased by image transformation.The super-resolution restoration technology of sequence images is adopted,and the low-resolution degraded images are processed by signal processing method.Thus,one or more high-resolution restoration images are realized.2.Based on the characteristics of overhead transmission line patrol service,this paper classifies and combs the risk identification scenarios of overhead transmission line equipment and channel defects,and carries out research work on Key Technologies of image recognition,puts forward the technical route of patrol image target detection,and further combs the current terminal acquisition mode of overhead transmission line based on video image patrol inspection in China,aiming at fixed-point photography.The status quo of patrol applications such as image heads,unmanned aerial vehicles and helicopters is analyzed.3.Aiming at the application mode of AI image recognition technology in transmission line patrol business,this paper studies and designs the three-dimensional patrol mode of transmission line,and puts forward the channel risk assessment system and security early warning mode.The main results include the following two points:(1)Through the research of quality evaluation technology of inspection defect samples,format transformation method of training data set based on deep learning framework and database storage and retrieval technology,the standard database of image samples for transmission lines has been successfully constructed,and the standardized collection of inspection defect samples with key features of multi-source data and the standardized management of transmission equipment information,unified labeling transformation and defect information have been realized.Efficient storage and retrieval of information.(2)Based on the intelligent patrol and artificial patrol of UAV,helicopter and robot,supplemented by the online monitoring technology of automatic image recognition technology,the special investigation and classification of mountain fires,mechanical breakage,foreign bodies and bird hazards are carried out.The transmission channel is constructed by combining the frequency of hazard events,as well as the information of topography,vegetation,grid structure and meteorological conditions.Risk assessment system and then formed the design of security early warning model.4.The standardized architecture design is carried out around the transmission line image data sharing and application demonstration system,which mainly includes four aspects:business architecture design,application architecture design,technical architecture design and data architecture design.The core functions of the system,such as image management,video management,analysis and identification,fault management and repair,are developed and implemented,and based on the application of system functions.Field scenario verification work has been carried out.