Faulty Feeder Classification Algorithm Based on Cost-sensitive Learning for Distribution Network
To achieve fault line identification in distribution network,the paper proposes a fault location method based on big data for distribution network.High dimensional combined characteristics in time-frequency domain are constructed by using correlation dimension algorithm and various time-frequency characteristic indices.Then,aiming at the unbalanced characteristics of the sample data set of single-phase-to-ground(SPG)fault,a faulty feeder identification based on AdaCost is proposed.This method trains cost factor based on true fault data,which can effectively reduce the false negative rate of fault samples.Finally,the proposed method is verified by physical model simulation.The simulation results show that the proposed method can effectively improve the accuracy of faulty feeder identification,and is not affected by the fault type,the proportion of fault samples and the neutral grounding modes.It provides a new solution for the fault location of SPG fault.
distribution networkSPG faultcost-sensitive learningunbalanced sample data setfault location
NETL
NSTL
万方数据
