| With the continuous enhancement of China’s comprehensive national strength,the scale of the State Grid is also expanding,which puts forward high requirements for the stability of the operation of the power system.In the power system,substations are the key links connecting power plants and users,and have the role of converting and distributing electrical energy.If the equipment in the substation fails,it will have some degree of impact on all walks of life.Therefore,the equipment in the substation needs to be regularly inspected.The traditional method of thermal fault inspection of equipment in substations has the disadvantages of relying on the experience of staff,and the detection speed and accuracy are not high.However,infrared detection technology has the advantages of high sensitivity,fast detection speed,and no direct contact,and is widely used in the field of thermal fault detection of substation equipment.With the increase in the number of infrared images of substation equipment collected by the operation and maintenance department of the State Grid,deep learning technology has played a certain role in this regard in order to improve the efficiency of thermal fault detection of infrared images of substation equipment.In view of the problems of increasing the number of infrared images of substation equipment collected,low manual detection efficiency,and difficulty in identifying small targets in complex backgrounds,this paper uses the infrared map of typical electrical equipment collected by the State Grid operation and maintenance department using infrared thermal imager as the experimental data set.In this paper,the basic structure and thermal fault types of the research objects are analyzed by taking the common electrical equipment of high-voltage bushings,current transformers,voltage transformers,disconnectors and circuit breakers in substations as research objects,and the temperature acquisition methods and thermal fault diagnosis basis of substation equipment are studied.In this paper,a deep learning model is used to compare the detection effect,which improves the efficiency and accuracy of thermal fault detection in infrared images of substation equipment.In order to improve the training speed and detection accuracy of the model,this paper performs two optimizations on the basis of the original Mask R-CNN model.First,this paper improves its backbone network structure.In this paper,Res Ne Xt-50 is used as the feature extraction network to improve the detection accuracy and improve the performance of the network while reducing the number of parameters.Secondly,the ROI Alignment of the region of interest pooling in the original algorithm is replaced by Pr Ro I Pooling,which eliminates the influence of the quantification operation of the ROI Align structure and improves the network accuracy.The improved Mask RCNN model has 2.04%,1.72%,1.30% and 0.02 higher average accuracy m AP,accuracy,recall and F1 scores for infrared image detection of substation electrical equipment than the original Mask RCNN model,respectively,and the detection speed of the improved model is 4 frames per second higher than that of the original model.The results show that the improved Mask R-CNN model significantly improves the detection speed and accuracy of infrared images of substation equipment,and improves the detection efficiency.Finally,the improved Mask R-CNN model is used to identify objects in substation equipment.The grayscale-temperature model is used for the infrared image,which is converted to the temperature range corresponding to the target equipment,and the fault level is determined according to the thermal fault diagnosis criterion of the substation equipment.In this paper,the effect test of infrared images of disconnectors and current transformers is carried out by using this model,and the results verify the feasibility of improving the Mask R-CNN model in the thermal fault diagnosis of infrared images of substation equipment. |
