To predict regional freight flow distribution more accurately and depict the connections and interactions between regional economies, a method combining a clustering algorithm and an improved gravity model was proposed. First, Pearson correlation analysis is performed on the influencing factors of freight. The K-means ++ algorithm was then used for node clustering, considering the heterogeneity and similarity between different regions, to study origin-destination (OD) pairs in detail. Next, the traditional gravity model was improved by introducing parameters such as freight influencing factors, social connection strength, and an impedance function constructed from distance and time costs. This makes the model more adaptable to different regional traffic characteristics, enhancing its generality and applicability. Finally, the improved gravity model was used to predict freight flow distribution in Yunnan Province, and the results are compared with those of the traditional gravity model. The results show that the improved gravity model improves prediction accuracy by 57.75% and stability by 54.66% compared to the traditional gravity model. This method significantly enhances prediction accuracy, providing a more reliable approach for regional freight flow forecasting.