Intelligent Diagnosis Of Defects Of High Voltage Equipment Based On Edge Computing And Improved SSD Algorithm

Transmission line equipment is in a complex environment,affected by factors such as electricity,magnetism,heat,light,mechanical force,temperature,humidity and pollution,it is prone to defects and deterioration,which may affect the safe and stable operation of the power system in severe cases.Realizing the daily live line inspection of transmission line equipment and ensuring its good operating status is of great significance to reducing the incidence of power system failures.At present,the inspection and evaluation of transmission line defects are mostly concentrated in the back-end,which is prone to insecure factors such as high computing pressure on the server and data loss.Aiming at the above problems,this paper proposes an intelligent inspection plan for transmission lines based on edge computing.Based on the image characteristics of transmission line equipment defects,an improved method of Single Shot Multibox Detector(SSD)is proposed.By collecting pictures of a large number of defective equipment on site,establishing a sample database and using geometric transformation methods to expand the sample database,the generalization ability and robustness of the model are improved.Use the low-dimensional Core convolutional group sub-network to replace the high-dimensional VGG network to extract the shallow features of the defect parts of the insulators,wires and the equalizing ring,then use the auxiliary feature extraction sub-network to predict the target,reduce the amount of model parameters and improve the hardware equipment.The memory requirement is low and it has good embedded detection performance.Research the optimization plan of-the improved neural network training process.By adopting the strategy of dynamically adjusting the learning rate and attenuating the learning rate at a certain rate for a fixed number of iterations,the extraction of deep features of defect images is strengthened.At the same time,the Momentum algorithm is adopted to optimize the learning mechanism of the model.Compared with the traditional learning rates of 1e-4,1e-5 and 1e-6,the self-updating learning rate mechanism used in this paper effectively improves the convergence speed of the convolutional neural network model.Study the real-time detection of video stream images at the edge and the model allocation scheme.Nvidia Jetson nano is used to build an edge computing platform,which is equipped with high-speed vision sensors.Develop intelligent monitoring software for defects in transmission line equipment.integrate deep learning algorithms and OpenCV to achieve real-time detection of video stream images.Through the background software.the pre-trained neural network models are distributed to edge devices through key encryption and destination list transmission to complete corresponding defect detection tasks,realize distributed computing and effectively improve data security and execution tasks The efficiency reduces the pressure on the back-end server.By improving the original SSD convolutional neural network algorithm model structure and training strategy,the paper effectively reduces the size of the network model,optimizes the training of the model and makes the model more suitable for the edge.Load the model to the edge and develop background management software,build a communication network between the edge and the server,realize the intelligent distribution of inspection tasks,effectively reduce inspection delays,improve data security and facilitate the construction of smart grids.