A rapid prediction model for motion responses of moored ship in harbors is developed based on convolutional neural networks (CNN) to enhance mooring safety and operational efficiency.Offshore wave conditions over a decade are simulated using MIKE21-SW.The maximum difference algorithm (MDA) is employed to select 212 groups of representative and return-period scenarios from offshore wave conditions spanning nearly a decade.Numerical simulations are conducted using the MIKE21-BW and MA software to extract motion response data of moored ship and establish a hydrodynamic database for the training and validation of the prediction model.With incident wave height,period,and direction as inputs,the model predicts six-degree-of-freedom motion of the ship.The prediction results indicate that the Pearson correlation coefficients of this model are all above 0.90,with the mean absolute error for translational motion not exceeding 0.015 m and the average absolute error of rotational motion controlled within 0.015°.The model can effectively capture the nonlinear characteristics of mooring motions and enable rapid transformation of massive historical data into refined motion hindcasts.It has been successfully applied to the daily wave-condition-based vessel motion hindcasting in the project’s sea area from 2021 to 2023.The research findings can provide a scientific basis for ports to optimize mooring plans