Research On Several Key Technologies And System Design Of Defect Module Location In Photovoltaic Power Plant

With the rapid growth of installed capacity of photovoltaic(PV)power plants both domestically and internationally,the maintenance of PV modules is becoming increasingly important.Due to the huge area of large-scale PV power plants,it is time-consuming and laborious with low efficiency to locate faults by manual inspection,which cannot accurately locate faults effectively.Therefore,a more intelligent and efficient method is needed to locate defective modules to improve efficiency of PV power plants intelligent maintenance.This paper focuses on the key technology research of defective module localization algorithms in practical PV power scenarios.The main contents of the work are as follows:(1)Aiming at the ineffective extraction of correct matching feature pairs for unmanned aerial vehicle(UAV)aerial dual-spectral images,a coarse-to-fine hierarchical matching method is proposed.By utilizing the characteristics of multi-sensor coaxial imaging carried by UAVs,the imaging parameters are used to achieve coarse matching.An intensity structure and statistical similarity-based criterion is constructed to evaluate the similarity of overlapping areas,which refer to the area between visible images/rectified infrared images and the fusion images,by considering the grayscale statistics,brightness,contrast and structure information synthetically.The quantum particle swarm optimization algorithm is implemented to obtain optimized registration parameters by maximizing the criterion.The experiment shows that this method is competitive compared with the joint area-feature based methods and facilitates accurate positioning of subsequent faulty components.(2)To address the difficulty of PV panel recognition in multi-scenario aerial images of solar power plants,a hybrid model based on U-Net and Efficient Net-B7 is proposed for dual-spectral images PV panel recognition.Firstly,a custom-designed data augmentation strategy is used to improve the accuracy and generalization performance of the model for multi-scenario aerial images.Considering the advantages of EfficientNet-B7 in nonlinear feature extraction,Efficient Net-B7 is introduced into the encoder of the U-Net network to construct a hybrid model for panel recognition.The robustness and reliability of the hybrid model are analyzed and validated on multiple scenario inspection images.(3)Aiming at the inconsistency of the coordinate system and the difficulty of feature matching between the UAV aerial image and the base map of the PV power plants,a module localization method based on the panel mask is proposed.Firstly,the position information and coordinate transformation of the PV aerial images and base map are utilized to extract the overlapping area on the base map.Then,the coarse-grained geographical registration of the aerial image and the base map is completed by optimizing the registration parameters.The panel with defects is matched using the mask contour centroid shortest distance method and the optimal corner localization method of the panels is used to determine the position information of the defective module in the fullscene base map.(4)A software system for locating defective modules in a PV power plant has been designed and developed.The system includes functions such as accurate registration of infrared and visible images,overview of the PV defective modules location information,and inspection reports automatic generation with defect module positioning.This system provides information support for subsequent operation and maintenance of PV power plants,and can improve operational efficiency.