Harbor oscillations is the most common form of wave disaster,which is very harmful.Accurate and efficient prediction of waves in the port is helpful to reduce economic losses and avoid casualties.The numerical simulation based on Boussinesq equation is an important method to study the harbor oscillation,and the calculation cost is high.To cope with the business oriented prediction of harbor oscillations,this paper proposes a parametric model of harbor oscillation and achieves rapid prediction of wave heights in the port,taking the Hambantota Port in Sri Lanka as an example.Based on the measured data of Hambantota port,the wave spectrum is divided into independent wave systems,each wave system is parameterized,the maximum difference selection algorithm (MDA) is used to select the calculation conditions,and the FUNWAVE-TVD numerical model is input to simulate the response of waves in the port under different incident wave parameter combinations,and the dataset is generated.The dataset is divided into generalization set and test set.The former is used to train and select the convolutional neural network (CNN),while the latter is utilized to measure the network performance on unknown cases.The output of the network is significant wave height and low-frequency wave height over the whole computing domain within the port.The results show that the model performs well on the test set and can accurately estimate the full-field wave height under unknown conditions.Then,the model is successfully verified according to the field measurement data,which proves the reliability of the wave numerical model and CNN.Once the wave parameters outside the port are obtained,the wave conditions inside the port can be quickly estimated.