Research On Fault Detection Of Photovoltaic Panel Inspection Image Based On Deep Learning

Photovoltaic panels are the source of life for photovoltaic power plants and the most important source in power generation systems.Their overall state is directly related to the power generation efficiency and safety of the entire photovoltaic power plant.Once the photovoltaic panel has some unexpected problems,it will cause huge economic losses and even casualties.Therefore,fault detection and diagnosis of photovoltaic panels has become a necessary means to ensure the normal and reliable operation of plant power generation systems in photovoltaic power plants,which has key economic and practical significance.However,with the widespread application of deep learning,the problem of fault diagnosis of inspection images based on deep learning needs to be solved urgently.Owing to rely on large number of sensors to transmit feedback information,or judge through remote sensing images,most of the existing photovoltaic fault detection methods are oriented to manual inspection data,which has the disadvantage of high cost,low accuracy,lack of pertinence,and narrow range of use.For the image data obtained by robot inspection,it is difficult to use the existing fault detection methods.The main problems are as follows:(1)the criteria for foreign object detection differ from person to person,and there is no truly unified fault detection standards;(2)due to the high economic cost,the existing fault detection methods are not suitable for all photovoltaic power plants;(3)there is no real-time diagnosis scheme for detecting foreign objects on photovoltaic panels in autonomous photovoltaic power plants based on patrol robot.Aiming at the above problems,this paper describes the methods of detecting foreign objects in the inspection images of photovoltaic power plant from the three aspects,including photovoltaic panel identification,positioning areas of interest in inspection images,and foreign object detection.This paper focuses on the application research of threshold segmentation detection method based on genetic algorithm in the inspect image of photovoltaic power plant.First of all,based on the national standards,this paper summarizes the existing fault detection methods of photovoltaic power plants,classifies and summarizes the faults and fault detection methods of photovoltaic power plants,expounds in detail the technologies and principles used in each type of fault detection methods,making a comparative analysis.Secondly,the deep learning method was proposed for to locate the photovoltaic panel area in the inspection image.This method uses the Single Shot Multi Box detection technology to accurately locate the area with photovoltaic panels.It can be applied to all photovoltaic power plant inspection,solving the problem of high economic cost and narrow scope of application.Thirdly,the method of threshold segmentation based on genetic algorithm is studied to detect foreign objects on photovoltaic panels,and the simulation experiment is carried out for the test data set.This method uses genetic algorithm-based threshold segmentation to transform the actual problem into an image segmentation problem.Experiments show that this method is an efficient method for foreign objects detection of photovoltaic panels.Finally,the foreign objects detection experiment results of photovoltaic panels are analyzed and summarized.