Research On Single Phase Grounding Fault Identification Based On Spatial Multi-dimensional Data Drive

The neutral point non-effective grounding is often adopted in distribution networks,which is also called low current grounding system.Single phase grounding fault is the most common fault in these systems,while the fault characteristics are not obvious.Hence,it is difficult to identify the fault.If the fault is not eliminated in time,it may cause further expansion of the fault,lead to other problems about endangering personal and property safety.Therefore,it is the guarantee of safe and stable operation of power grid that identify and eliminate the fault timely.The previous research methods are divided into two categories: model-driven and data driven.The former are based on analysis about the circuit principle and extract the fault characteristics from the perspective of the physical model,such as the method based on zero-sequence current amplitude and phase angle.The accuracy of the model directly affects the applicability of the method.Data-driven methods focuses on data mining and then extract the fault characteristics,such as fault feature data clustering method.These methods use low precision,high density and large dimension data as the foundation,and heighten the influence of the data as well as reduce the dependence on the model.The development of fault indicators is becoming more and more mature,and some products already have data collection functions,which provide data support for data-driven methods.This paper proposes to complete the collection of multi-dimensional fault data of the entire network when the system fails by extracting the fault indicator data,which greatly improves the data dimension.Compared with the traditional data-driven methods,only the initial data of the line in the substation is used.The multi-dimensional data can more fully reflect the system failure status,making it possible to extract the failure characteristics from the redundant data.Based on the analysis of the physical model and the line differential current value as the basis for fault identification,this paper proposes the theory of using line multidimensional data sets to realize line fault identification.In the context of intelligent development,the deep neural network is used to efficiently classify data sets as the basis for distinguishing faulty lines and sections.Moreover,the deep neural network fault recognition model is constructed,and then the network training and testing algorithm is designed.Through design simulation,this paper verifies that the method can accurately identify faults under the conditions of variable transition resistance,unbalanced load,and T-shaped branch.The test experiment carried out in the power system dynamic simulation laboratory of Huazhong University of Science and Technology also verified the feasibility of the method.