Abstract: Ship operation time is a crucial factor in developing berth plans. While most existing researches focus on container terminals, few scholars conduct in-depth studies on dry bulk terminals due to the unique nature of their cargo and complex operational processes. Considering the similarities and differences among berths at dry bulk terminals and the limitations of a single machine learning model, this paper first proposes a berth clustering method based on K-means algorithm to cluster berths with similar operational characteristics. Then, an adaptive Stacking ensemble learning model combined with simulated annealing algorithm (SA-Stacking) is proposed for predicting ship operation time in various types of berths at dry bulk terminals. Real historical operation data from Dry Bulk Terminal of Qingdao Port is utilized in this paper. Experimental results demonstrate that the SA-Stacking model can adapt to select the most appropriate combination of base models according to the data distribution characteristics of different berth categories, yielding superior prediction accuracy and generalization ability compared to single machine learning models. Furthermore, the prediction approach based on berth characteristic classification can better capture the operational patterns of different berths. Compared to prediction without berth clustering, the mean absolute error is reduced by nearly 2 hours, and both the root mean square error and the mean absolute percentage error are significantly reduced. Simultaneously, the R-square is increased, enhancing both the interpretability and prediction accuracy of the model.