Research On Substation Dynamic Patrol Inspection Algorithm Based On Inspection Robot

The power grid plays a pivotal role in China’s modern energy supply system,which is related to the energy security of the whole country.As a key part in the power grid system,the substation is of great significance to ensure the safe and stable operation of the substation.The traditional manual inspection method of substation is inefficient,and the detection data cannot be shared in real time,which is in contradiction with the requirements of high intelligence,informatization and interaction of power grid.It is the general trend to promote the intelligent operation and maintenance of substation.In this paper,based on the dynamic patrol inspection of substation inspection robot and guided by the actual problems in the inspection process,the related problems encountered in substation inspection are deeply studied.In this paper,the image quality problems and common equipment abnormal defects of inspection robot in substation inspection are studied.In the reading task of pointer type instrument of inspection robot,the inspection image is blurred due to many factors,which brings great trouble to the reading of meter.In addition,in the indoor inspection task,there are many rectangular pointer instrument in one inspection image,some of which have large disparity with the template image,and the traditional image correction algorithm based on feature points has large error.There is also a very difficult problem in the inspection task is the abnormal detection of the inspection image.Because of the various types of defects in the substation,it is not realistic to establish the detection target database for all cases.For the above problems,the following research has been carried out in the paper:(1)The reading of pointer instrument needs better image quality.Two kinds of evaluation indexes of pointer instrument image clarity based on re blur are proposed.After obtaining the dial image,the relevant indexes are used to evaluate the image clarity.For the image that does not meet the conditions,the Generative Adversarial networks are used to repair the blurred image,which reduces the error of intelligent reading algorithm of pointer type instrument.(2)This paper studied and analyzed the frequently-used SIFT feature points algorithm,SURF feature points and ORB feature points and its registration algorithm,deeply studies the Super Point feature points extraction algorithm,builds the algorithm trained environment,and compares the registration effect of several feature point algorithms in the instrument image through experiments.The experimental results demon-strate that in the patrol image registration task,the Super Point algorithm has better performance in the correction time and accuracy and the results of Super Point are better than SIFT algorithm,SURF algorithm and orb algorithm.(3)This paper studies the related algorithms of target detection,and proposes the fusion of YOLOv4 target detection algorithm and depth feature difference algorithm for substation abnormal state detection.Firstly,the common defect types or equipment are detected by YOLOv4 algorithm,and then the image depth features are extracted and fused by VGG-19 pre-trained network,and the fused depth features are subtracted from the image features of template library,so as to pass the detection.Finally,the effectiveness of the algorithm is verified by experiments,which provides a novel idea for substation anomaly detection task.