Transmission Line Fault Detection And Classification Based On Deep Learning

In recent years,with the increasing complexity of power system,its fault analysis needs a lot of data processing.With the continuous development of artificial intelligence technology,the development of intelligent fault detection and classification system is a new trend,in which deep learning method has become the latest technology to deal with fault detection and classification.This paper uses deep learning technology to detect and classify transmission line faults.The main research topics are as follows:1)Aiming at the end-to-end automatic detection of transmission line fault target,this paper proposes two kinds of fault detectors based on deep learning.The first detector is wavelet time scattering support vector machine,which is the integration of wavelet time scattering network and two classifiers of support vector machine.As the name suggests,its wavelet time scattering network is mainly to automatically extract the fault features in the input time series,and the support vector machine classifier identifies the fault based on these features.The second detector is the attention gated circulation unit,and the attention mechanism is to scale the dot product attention.It is also a wavelet time scattering network that automatically extracts fault features,and the attention gated loop unit further extracts features to identify faults with softmax layer.2)For the end-to-end automatic classification of transmission line fault targets,two types of fault classifiers are also proposed based on deep learning.The first classifier is a multichannel convolution bidirectional long-term and short-term memory network,which is also the integration of multi-channel convolution and bidirectional long-term and short-term memory network.It can have the advantages of both.The natural regression coefficient,Shannon entropy and singularity spectrum of the input time series are manually selected as the fault characteristics,and the fault three-phase signals can be input at the same time to ensure the multi classification of faults.The second classifier is a residual network with parallel connection between compression and incentive learning.The compression and incentive learning module is also a special attention mechanism.It is also the integration of parallel link network and residual network,which can have the advantages of both.It extracts the characteristic diagram(time-frequency diagram,tiled diagram of time and frequency of time series input),and splices the three-phase time-frequency diagram into the input characteristic diagram of the system in the order of a,B and C,which ensures the multi classification of faults.In this paper,two kinds of detectors are tested on the public data set of eNet.The simulation results show that compared with other detection methods,this detector can improve the detection accuracy of IOC fault.The other two classifiers are also tested on the simulation data set generated by the 735kv transmission line model.The simulation results show that compared with other classification methods,this classification can achieve the classification accuracy of advanced faults.