Abstract: In the context of large-scale renewable energy integration, the intermittency and volatility of renewable energy generation superimposed with load variations have exacerbated voltage fluctuation issues. To accurately locate disturbance sources in distribution networks, this paper proposes a method for voltage fluctuation source localization based on Bayesian-optimized XGBoost. The method extracts the disturbance energy, system trajectory slope, equivalent impedance, and the real part of current as key localization features. The XGBoost ensemble classification learning is employed to mine the mapping relationship between disturbance feature quantities and disturbance location information, constructing a multi-feature fusion voltage fluctuation source localization model. The Bayesian probability prediction is utilized to realize adaptive optimization of the XGBoost model hyperparameters, further enhancing the model's accuracy and robustness. Simulation results on the IEEE 33-node system demonstrate that the proposed method achieves a fluctuation source localization accuracy of up to 99.6%, significantly outperforming conventional methods. This provides an effective assurance for the stable operation and high-quality power supply of modern distribution systems.