Abstract: As a key infrastructure in the flood control and disaster reduction system of river basins, the existing risk prevention and control units of embankment projects are mismatched with their linear engineering spatial forms, making it difficult to achieve precise spatial quantification of potential risks in different sections of the embankment. [Objective] To improve the accuracy of identifying the spatial distribution pattern of potential vulnerable sections, [Method] an evaluation unit division method along the embankment axis line with multiple scales was proposed by introducing the vulnerability evaluation modeling paradigm. Taking the typical embankment in Poyang Lake area as an example, an embankment risk vulnerability evaluation model was constructed, and the prediction performance of five spatial resolution scales (10 m, 15 m, 20 m, 40 m, and 2000 m²) and two machine learning models (RF and SVM) was systematically compared. [Result] The results show that: (1) The prediction results of embankment risks are highly sensitive to the spatial scale of the evaluation units, and the prediction accuracy is significantly positively correlated with the refinement of the units; (2) When dealing with multi-source structured data, the RF model, with its ensemble learning mechanism and strong nonlinear fitting ability, has significantly better overall prediction accuracy and robustness than the SVM model. (3) The RF-10m model has the best performance, with an AUC value of 0.952 and an accuracy rate of 92.29%, and the spatial distribution of the vulnerability index is more reasonable, with the highest spatial matching degree between the extremely high and high vulnerability areas and the historical risk sections. [Conclusion] Vulnerability evaluation can be effectively applied to the identification of embankment risks. Fine segmentation along the embankment axis line and coupling with high-performance machine learning models can significantly improve the evaluation accuracy, providing scientific support for embankment risk early warning and disaster reduction planning.