| Demand for power grid reliability is improving because of development of society.Fault identification and protection action of distribution network is important.However,IIDGs not only bring positive effect,but also affects the application effect of traditional fault identification methods and current protection.Based on data driven,this paper proposes a fault identification method of distribution network,and then proposes the fault identification model and adaptive current protection according to the situation that IIDG is connected to the distribution network.Because the steady-state characteristics of single-phase earth fault are not obvious,the identification accuracy is affected when the transition resistance is high.A fault identification method for distribution network based on improved artificial immune network is proposed.Firstly,the characteristics of zero sequence voltage and three-phase current were extracted.The characteristics of zero sequence voltage were extracted by Hilbert Huang transform,which reflect the waveform change degree.Then,the memory cell adaptive adjustment was used to improve the artificial immune network to generate the feature vector that can reflect the fault,which can overcome the uneven training data.Finally,the fault identification model was established based on K-nearest neighbor.This method can identify the fault and normal disturbance accurately,and its adaptability is still good when the neutral grounding mode changes.More and more IIDGs are connected to the distribution network.To overcome the large scale of simulation and the lack of training samples,A fault identification method for distribution network with IIDG based on deep transfer learning is proposed.Firstly,the fault current characteristics of IIDG with low voltage ride through function were analyzed,and the source domain feature matrix was constructed.Then,the sample features in different scenarios were the target domain.The joint distribution adjustment method was used to improve the intra class compactness and inter class discrimination of the data between the domains.Finally,the source domain feature samples were input into the long-term and short-term neural network,which is good at processing sequence data.The network was trained to detect the mapped target domain samples.This method can identify the fault accurately due to the reduction of the feature differences between domains.Compared with other methods,this method has better generalization ability and higher classification accuracy.In view of the limitation of the existing protection principle after IIDG access,an adaptive current quick break protection method for distribution network with IIDG was proposed,which is bases on random forest.Firstly,the influence of IIDG on traditional protection was studied,and the characteristics of short-circuit current were analyzed.Secondly,the short-circuit current without IIDG was proposed as the characteristic quantity,and the current after fault was calculated by random forest.Finally,the setting value of current protection was adaptively adjusted according to the current calculated from the line.The method can calculate the current accurately,and can operate quickly and reliably when the output of IIDG changes. |
