Research On Fault Identification Of Distribution Network With Photovoltaic Power Supply Based On Improved Residual Network

Distribution network is the key link of power distribution in power system.The accuracy of fault identification directly affects the reliability of end users.At present,the grid structure of distribution network in China is complex,feeder faults occur frequently,and there are many safety risks.In recent years,with the gradual maturity of new energy power generation technology and the proposal of "double carbon goal",distributed clean energy such as photovoltaic power supply is widely connected to the distribution network,and the topology of the distribution network is changed from a single radial power supply network to a multi power supply network,which increases the complexity and uncertainty of operation.In addition,the fluctuation of photovoltaic power generation makes the output under different working conditions have great differences.These influencing factors reduce the identification effect of commonly used traditional identification methods,which is easy to cause fault misjudgment and missed judgment.Therefore,in the scenario of photovoltaic access,there is an urgent need to accurately identify the fault type and fault feeder for the distribution network.After discussing and analyzing the advantages and disadvantages of the main identification methods,the main research work is as follows:Firstly,based on PSCAD electromagnetic transient simulation software,the distribution network simulation model with photovoltaic power generation system is built.The simulation experiment is used to explore the main environmental factors causing photovoltaic fluctuation,and the experimental data of photovoltaic output fluctuation are obtained by adjusting parameters.Combined with the experimental data characteristics of photovoltaic distribution network fault under different access positions and different output,the impact of photovoltaic access on distribution network fault identification is analyzed.Secondly,in order to obtain a rich set of fault samples,a program is written to realize automatic batch simulation for multi scenario fault experiment,so that the fault data covers a variety of working conditions.In the actual operation,the deployment of feeder acquisition device is limited.Combined with entropy theory and variational mode decomposition theory,an entropy-variational mode decomposition(E-VMD)fault feature extraction method is proposed to mine the fault data features and construct a neural network data set containing the fault features of spatial location and timely frequency domain information.Finally,in the training process of neural network,the classification and characterization effect will be reduced due to the multi-layer back propagation and multi-layer convolution operation of error signal.This paper proposes to improve the residual network to ensure the performance of the model with "identity mapping",strengthen the nonlinear expression ability of the network model through multi-scale feature extraction,and overcome the disadvantages of over fitting and network degradation caused by feature data over extraction.The model training test results show that the improved residual network model is more stable and can accurately identify the fault type and fault feeder.In order to further verify the generalization performance of the model,the model is tested and analyzed in three cases such as data loss.The experimental data show that the model has good adaptability and robustness,and has good classification effect in the task of fault type and fault feeder identification of photovoltaic distribution network.