Intelligent Fault Diagnosis Technology Based On Computer-visualized Power Flow

Fault diagnosis is an important basis for accident treatment and analysis,and is of great significance for improving the reliability of the grid.With the gradual increase of installed capacity of new energy and the acceleration of UHV transmission line construction,the characteristics of faults have become increasingly complex.This requires dispatchers to monitor the weak links of the grid in real time,analyze the massive abnormal measurement data in time after the fault,accurately and quickly judge the cause and take measures.Therefore,it is necessary to study the intelligence technology used to help dispatchers quickly and comprehensively control the information of power flow and provide the corresponding decision support.Therefore,with the aim of intelligent fault diagnosis,using deep learning technology,considering the variation of power flow in continuous periods and the realization of comprehensive diagnosis of multi-agent in large grids,this paper designed a fault diagnosis method for power grid dispatching based on the idea of computer visualization of power.To analyze the distribution of power flow in the grid before and after the fault,a visualization method was used to compare the active and reactive power flow when different faults occur in different locations.On this basis,a transformation method from digital power flow to computer-visualized power flow(CVPF)was proposed.By transforming the power of the electrical equipment and the topology of the grid,this method converted the change in power flow into the change in the outline of the pixel block on the image.Finally,the goal of converting structured and digitized currents into pictures that can be recognized by convolutional neural networks was completed.Based on the CVPF transformation method,by extending the CVPF transformation data source to multiple time sections in series or difference form,this paper proposed two specific sample generation schemes,dynamic CVPF and difference CVPF.At the same time,in order to make up for the shortcomings of fewer fault samples in the real-time data,a scheme for generating a massive CVPF sample sets was proposed,which considered the fluctuation of power sources and the change in topology.Then,by taking the classic structure as a blueprint and considering the latest research progress in the field of image recognition,a convolutional neural network that can identify and diagnose CVPF samples was constructed.Finally,the effectiveness of dynamic CVPF for fault diagnosis was demonstrated by tracking the convergence of the neural network and reducing the dimensionality of the output features of each layer.Furthermore,the problems with difference CVPF were highlighted through the comparison of case studies.For the large grid,a comprehensive fault diagnosis method based on multi-agent perception of the local CVPF was proposed.This method first splits the entire network into a number of small sub-networks,and then transforms them into a local CVPFs.Local CVPF is then used to train the convolutional neural networks separately,and finally a multi-agent cluster is formed for comprehensive consultation.First,the retrieval and generation of the radial network was accomplished by defining nodes and branches at all levels.Then,using the fluctuation of the power flow on the branch as an indicator,the multi-agent diagnosis startup strategy was designed.The case study highlighted the problem of false starts of agents in a small observation range,and verified the feasibility of using multi-agent clusters to perceive local CVPF to achieve a comprehensive diagnosis within the jurisdiction and across the jurisdiction.In this process,the Precision was used to evaluate the agent’s online cross-jurisdiction diagnosis.