A Two-Stage Fault Localization Method for Active Distribution Networks Based on COA-SVM Model and Cosine Similarity

To address the issues of low efficiency and poor noise immunity in traditional active distribution network (ADN) fault location methods based on swarm intelligent optimization algorithms, this paper proposes a two-stage fault location method utilizing the COA-SVM model and cosine similarity. First, this paper constructs the fault signature database for the target distribution network by randomly simulating single- and multi-point faults using the fault current state equation. Next, this paper introduces the COA-SVM classification model, establishing the high-dimensional mapping relationship between the fault current direction matrix and the fault zones through model training. The well-trained COA-SVM classification model is used to identify the fault zones, which include the fault line segments. Finally, for each identified fault zone, this paper calculates the cosine similarity of the fault current direction information of adjacent line segments, accurately pinpointing the fault line segments by identifying mutation points of the cosine similarity. Using the modified IEEE 33 node test distribution network as an example, simulation results demonstrate that the proposed two-stage fault location method offers higher accuracy and resistance to signal interference compared to fault location methods based on swarm intelligence optimization algorithms. The COA-SVM classification model surpasses conventional models, achieving high accuracy and excellent noise resilience. It accurately identifies fault segments within the test distribution network with a remarkable 100% precision. Moreover, the accuracy of fault localization remains above 83% when the FTU encounters fewer than three abnormal signals.