Power System Fault Diagnosis Based On Deep Learning

In recent years,the pace of development of China's power grid has been accelerating.Currently,a gradually complete UHV,AC-DC hybrid,and long-distance transmission network has been formed.Our modern power grid architecture,which is characterized by large-scale renewable energy access and intelligence,has begun to emerge..With the enhancement of grid architecture and operation level,users' demand for power supply reliability has gradually increased.How to establish an efficient and automated grid fault diagnosis strategy under increasingly complex operating conditions,effectively shorten the outage time and reduce the power outage caused by faults.And the economic loss of users has always been a basic problem in the research and application of power grid dispatching operations.Aiming at the problem that the rule-based fault diagnosis system for power grids needs to introduce a large number of protection and equipment operation rules in the early stage of modeling and modeling difficulties,the fault event description word segmentation in the alarm information is used as the object to study its distribution characteristics in the process of fault evolution and extraction.The important text features of the alarm information text,the vector space model of the alarm information is established,and the vectorization of the alarm information text is completed.A power grid fault diagnosis method based on support vector machine is proposed.Simulation results show that the model can accurately diagnose single/multiple faults and line/bus/transformer fault types based on the establishment of a ten-dimensional vector space.Aiming at the shortcomings of text vectorization based on support vector machine for power grid fault diagnosis,which cannot directly extract fault features,combined with deep learning theory,a power grid fault diagnosis model based on convolutional neural network was established,which can automatically extract text features In order to realize the classification and prediction of alarm information text,the modeling process is further simplified,and the accuracy of diagnosis is not limited by the inherent model,which improves the adaptability of the model.Finally,the validity and rationality of the proposed diagnostic method is verified by simulation.