Abstract：To tackle the issues of low execution efficiency and poor fault tolerance in traditional fault

localization methods for active distribution networks using swarm intelligence optimization algorithms，a two-stage

fault location method was introduced based on the SSA-RF algorithm and cosine similarity. Firstly，the fault current

state equation was used to create a fault feature database of the target distribution network by stochastically

simulating single-point and multi-point faults. Next，an enhanced random forest（RF）classification model that

integrates the sparrow search algorithm（SSA）was introduced. Through model training，a high-dimensional

mapping correlation between the fault current direction matrix and the line segment containing the fault point was

established. This trained SSA-RF classification model was utilized for the initial localization of the faulted line

segment. Subsequently，cosine similarity of fault current direction information of neighboring segmented lines

within the identified segment was computed for precise fault location. Experimental results on the modified IEEE

33-node test distribution network demonstrate that the proposed two-stage fault locatlizaion method achieves

superior accuracy and anti-interference capabilities compared to fault location methods based on swarm intelligent

optimization algorithms.

      Key words：active distribution network；fault localization；two-stage model；random forest（RF）；cosine

similarity