Research On Image Acquisition And Reconstruction Of Electrical Power System Based On Compressed Sensing

With the continuous expansion and upgrading of electrical power grid construction,the cost of power system management and maintenance is also increasing.How to effectively use the development results of emerging technologies to improve the management and maintenance of electrical power systems is essential.Images of electrical power system play a very important role in the transmission of electrical power system information and the monitoring and maintenance of electrical power system,while large-scale power system will produce a large amount of power system image data,which impacts the transmission bandwidth and storage burden of power system information during the operation of power system.How to effectively avoid the waste of bandwidth resources and the burden of storage space is the current research difficulty.In order to solve this problem,this thesis proposes a method of power system image acquisition and reconstruction based on compressed sensing,which is of great significance to improve the management and maintenance level of power system.Based on the theory of compressed sensing,this thesis firstly models the acquisition and reconstruction of electrical power system image,analyzes the traditional methods of the image of electrical power system reconstruction,classifies the reconstruction methods in the field of compressed sensing,and expounds the theoretical basis and method steps of each algorithm,and focuses on the data-driven method based on deep learning and data-driven and prior knowledge-based mixed methods.Then,this thesis proposed the method of image sampling and reconstruction based on deep learning.At the reconstruction side,the stack de-noising encoder and convolutional neural network are used.At the acquisition side,the neural network is used to sample the power system image.At the same time,the end-to-end training method is used to jointly train the reconstruction network so that the reconstruction network can guide the acquisition layer to acquire the reconstruction needed to improve the quality of power system image reconstruction.The results show that under four measurement rates,the deep learning sampling method can improve the reconstruction effect of power system image.Then,in order to solve the problem that a single neural network cannot give full play to the advantages of neural network in the reconstruction of images of electrical power system,this thesis cascades the stack denoising encoder and convolution neural network to give full play to the advantages of stack de-noising self encoder in processing low-dimensional input and high-dimensional output of data,and the excellent performance of convolution neural network in processing image problems,combining the advantages of the two network,designed a cascaded reconstruction network.Experiments show that the combination of the two improves the reconstruction performance of images of electrical power system.Finally,in view of the shortcomings of the previous neural network that can not make full use of the low-level information of the sampled image,this thesis designs a dense residual compression sensing network,innovatively designs a dense residual block to fine reconstruct the power system image,transfers the low-level information to each layer of the residual block,and makes full use of the information obtained by each layer of the network.Experiments show that the dense residual network obtains better reconstruction results after obtaining more low-level information.