Research On Thermal Fault Diagnosis Method Of Substation Equipment Based On Deep Learning

The reliability of substation equipment operation is directly related to the stability of the entire power system operation.With the development and popularity of infrared thermal imaging technology,the state network operation and maintenance departments also widely use this technology for thermal fault detection of substation equipment,but with the increase in the amount of infrared image data of electrical equipment collected by the operation and maintenance departments,the efficiency of using manual thermal fault diagnosis of substation equipment is reduced.To improve the efficiency of infrared inspection,this paper analyzes the shortcomings and improves the Faster R-CNN algorithm based on the deep learning region suggestion algorithm Faster R-CNN for the problem of too many infrared images of substation equipment,high manual screening workload and low efficiency of thermal fault diagnosis.Firstly,it introduces the structure of six types of electrical equipment in substations:disconnect switches,circuit breakers,current transformers,high-voltage bushings,lightning arresters and voltage transformers,as well as the causes of common thermal faults,and explains the application of infrared detection technology;the infrared images of substation equipment collected by an operation and maintenance department in Jilin Province of the State Grid using a German Testo infrared camera are used as the data set for experiments,and the infrared images of electrical equipment The labeled infrared image samples are labeled using the software Labelimg to construct the experimental dataset.Then,for the problem of difficult target recognition due to the small size of substation equipment,this paper adds the feature pyramid network FPN to improve the original Faster R-CNN algorithm based on it,so that the detection accuracy of small size electrical equipment is enhanced greatly,while the Res Net-101 network is selected as the feature extraction network of the improved model and combined with FPN For the ROI pooling in the original algorithm,it is replaced by ROI Align;the generation of the region suggestion box anchor in the RPN is improved by introducing the clustering algorithm k-means++,so that the target prediction box generated by the improved model is closer to the shape of the substation equipment.Through experimental comparison,the detection speed of the improved model is 5frames per second higher than that of the original model,and the average accuracy m AP,accuracy and recall rate and F1 score of the improved Faster R-CNN model are 2.51%,4.49%,2.13% and0.4 higher than that of the original Faster R-CNN model for the detection of infrared images of substation equipment,respectively.The experimental results prove that the improved model has a large improvement in the detection speed and average accuracy for the infrared images of substation equipment.Finally,this paper first combines the classification and localization function of the improved Faster R-CNN model for substation equipment with the grayscale-temperature conversion formula,first classifies and localizes the substation equipment to be identified in the input IR image separately,then intersects and compares the equipment region and the abnormal heating region in the IR image,uses the grayscale processing method for the intersecting region to get the corresponding grayscale value,and calculates the image according to the grayscale-temperature conversion formula-Then,the corresponding highest temperature value and temperature range are calculated based on the grayscale-temperature conversion formula,and then combined with the selected electrical equipment fault diagnosis criterion to derive the equipment fault situation.is feasible.