Research On Single-phase-to-ground Fault Line Selection Of Small-current Grounding System Based On Improved Deep Confidence Network

Small current grounding system is widely adopted in domestic and foreign medium voltage power grid,and single-phase grounding fault is the main fault form in this operating mode.In order to avoid power supply interruption caused by tripping,it is necessary to locate and handle fault line in time.Due to the complexity and changeability of modern distribution networks,traditional transient and steady-state fault line selection methods are greatly affected by external factors,which causes low fault line selection accuracy.With the development of artificial intelligence and modern signal processing technology,the intelligent wire selection method based on information fusion has emerged and has a good prospect of development.Based on the transient and steady-state characteristic quantities of single-phase grounding faults in small current grounding system,this thesis focuses on the shortcomings of single fault characteristic quantity line selection methods.Deep belief network(DBN)and variational modal decomposition(VMD)are applied to line selection.In this work,we use the fault transient and steady-state eigenvolumes to construct the line selection model.The main researches are as follows:(1)The commonly used single-fault characteristic quantity line selection methods are realized through principle analysis and software simulation,and the factors that affect line selection accuracy are analyzed.Meanwhile,this thesis determine the application scope of each method under two cases: unearthed neutral system and Peterson coil compensated neutral system.The advantages and disadvantages of the existing selection methods are summarized,and information fusion is indicated as the development trend of the fault selection method.(2)The signal decomposition process of the VMD algorithm is described in detail,and its ability to deal with the modal aliasing is proved.A method is proved to determine VMD decomposition levels K,and it is verified by actual fault signals simulation.Based on the intrinsic mode functions obtained by VMD,two fault transient characteristic quantities: transient energy and transient energy entropy are proposed to quantify the difference between fault lines and non-fault lines.(3)A fault line selection method based on DBN is designed,and PSO algorithm is used to select the number of neurons in each hidden layer.It will improve the feature extraction ability,and determine the most optimal line selection model.Two transient characteristic quantities of transient energy and transient energy entropy are combined with two steady-state characteristic quantities of high-order harmonic synthesis and zero-sequence active components of each line to form a fault data set,it is used as the input of DBN network to realize fault line selection target.(4)The superiority and feasibility of this algorithm is verified through classic models.Firstly,Simulink is used to generate fault signals,and Matlab is used to process the raw data and obtain fault data set.Then,the line selection model of this thesis is trained,and it is compared with the original DBN algorithm,BP algorithm and PSOBP algorithm,the results will verify its superiority.Finally,real environment is simulated to prove the feasibility of the algorithm in engineering application.