Abstract: Due to the limitation of communication bandwidth, the acquisition of ship navigation information such as main engine speed and draft is distorted and missing, which often leads to the uncertainty of ship speed prediction at shore end, and also brings potential safety hazards to the ship. In order to predict ship speed under different sea conditions with poor information, in-depth investigation is conducted on 1101 voyages of 87 ocean bulk carriers sailing on China-Southeast Asia route, including a total of 1.45 million samples. A ship speed prediction model is established driven by marine meteorological information. Considering that the difference of meteorological conditions between different voyages will lead to the lack of generalization ability of the data-driven model, a representative method is used in the field of transfer learning domain adaptation (DA) to optimize the characteristic space of training data, so that distribution characteristics of meteorological data in the training set and the test set tend to be consistent, ensuring that the training model is also robust on new test data. DA method is integrated with LightGBM to develop the training model and conduct the independent test. 50-route closest to the departure time of the voyage are selected as the test set. Prediction results indicate that the mean of error absolute value between the speed predicted by LightGBM model and AIS speed is 0.47, and the root mean square error is 0.65, which increases by 32.85% and 30.53% respectively compared with prediction results of empirical equation. The mean of error absolute value between the speed predicted by DA-LightGBM method and AIS speed is 0.42, while the root mean square error is 0.56. These values represent increases of 10.63% and 13.85%, respectively, when compared to the speeds predicted by the LightGBM model. Additionally, determination coefficient is improved to 0.82. The sample deviation of the speed predicted by DA-LightGBM is nearly 70% lower than 0.5 knots. From the perspective of geographical waters, in narrow waterways, islands and waypoints close to land waters, DA-LightGBM compared with LightGBM, speed predictions of 3 methods have significant deviations in the waters around Taiwan Island, Hainan Island, the Strait of Malacca and the eastern Philippines. The significant deviation does not preclude the use of motorized navigation near the reef islands. DA-LightGBM demonstrates superior model interpretability and predictive speed, effectively meeting demands of ship speed prediction, and it enhances the capabilities of ship navigation service support.