Double Terminal Fault Location Study Of 110kv H Igh Voltage Distribution Line Based On One-dimension Al Convolutional-gated Recurrent Unit Algorithm

At present,with the growing scale of power grid operation,new energy grid and ac/dc coupling information to an adverse effect on power grid and the operation of the huge amount of growth,make traditional line fault location method can not be well deal with huge data size,complex fault mechanism,such as timing,coupling problem,cannot be fully to achieve precise fault feature representation,In this way,the line fault can be accurately located.Deep learning network does not need to extract features manually,and can learn deeper data feature connections through data samples.It has strong nonlinear fitting ability,and the learning ability is stronger with the increase of network depth.Based on this,this paper carried out a research on fault location of 110kV high-voltage distribution network based on deep learning network.Firstly,the electrical simulation model was established by using the actual distribution network parameters,and the fault data samples were obtained by fault setting,and the data samples were preprocessed to eliminate the influence of the difference between the two physical quantities.Secondly,on the basis of previous studies,the characteristics of each mainstream neural network are fully considered,and the fault location model of high voltage distribution network line based on one-dimensional convolution gating cyclic unit is proposed.Finally,the deep network is trained based on the data samples and model structure,and the ranging accuracy of the proposed model is calculated and analyzed based on the test samples.The results show that using the model proposed in this paper,the minimum deviation of line fault location is only 0.17m,and the average deviation of test samples is 129.6145m.Compared with the other two positioning models,it is found that the root mean square error of the proposed model is 0.1691,which is 37.83%higher than the maximum error of the control model 0.2720.At the same time,it is found that the change of transition impedance has no great the fault location accuracy,which indicates that the model can cope with the location problem in the case of large range changes of excessive impedance.The research results further show that the deep learning network can obtain deep data features,which has positive significance for improving the fault location accuracy of the line.