Wave information forecast is of great significance to ship navigation safety. Aiming at the problem of significant wave height forecast, a forecast method based on extreme learning machine is proposed to improve the capability of wave information forecast. Based on the measurement data of the real ship, the optimal hidden layers of the extreme learning machine forecast model are first determined, and then the forecast results of the extreme learning machine algorithm and other three classical algorithms are compared to verify the forecast advantages of the model. The forecast accuracy of the proposed models is tested under the condition of small sample. The generalization ability of the models is verified and analyzed by using the sample data of different time periods. The results show that the significant wave height forecast model based on extreme learning machine has obvious advantages over the other three algorithms, and can provide support and reference for the navigation safety of ships.