High impedance fault feeder selection method based on MPA-VMD and LightGBM in distribution networks

• A new high impedance fault feeder selection mothed of distribution networks. • The HIF features is obtained by using MPA-VMD. • The energy matrix maps the fault features and inputs the lightGBM. • Impact of noise, topology, fault location and other factors. • The effectiveness of the recorded data verification method. The accurate selection of fault feeder during single-phase high impedance fault (HIF) in distribution networks is challenging due to the complex conditions and weak fault characteristics. This paper proposes an HIF feeder selection method that combines the marine predators algorithm-variational mode decomposition (MPA-VMD) and a light gradient boosting machine (LightGBM). Initially, different HIF models are compared, and an enhanced Emanuel model is selected to simulate and analyse the characteristics of the HIF zero-sequence current signals. A data window adaptive algorithm is then designed to capture the significant signal period of the zero-sequence current for each feeder. Subsequently, the VMD parameters are optimized using MPA to decompose the collected zero-sequence current signal into multiple intrinsic mode functions (IMFs). The IMFs, which best characterise the HIF information, are selected to construct an energy matrix, serving as a feature mapping of the original signal. Finally, LightGBM, adjusted through Bayesian parameter optimization, is used to extract the characteristics of the zero-sequence current in the energy matrix, enabling accurate feeder selection. Extensive tests using simulation and recorded data demonstrate that this method offers excellent HIF feeder selection performance and holds significant potential for practical engineering applications.