Abstract: Autonomous sailboats, as cutting-edge platforms for ocean observation, offer unparalleled capabilities for long-duration, large-scale, and high spatiotemporal resolution marine monitoring tasks. However, it is difficult to predict the trajectories of autonomous sailboats accurately by fixed models, due to the complex and variable nature of the marine environment, coupled with the extensive temporal and spatial scope of their operations. To address this challenge, a path prediction method based on VFOSELM-SRP (variable-forgetting-factor-based online sequential extreme learning machine) model is proposed, to tackle the trajectory prediction difficulties of the Seagull autonomous navigation sailboat. The proposed method leverages the VFOSELM for real-time prediction of the sailboat's coordinates, and utilizes SRP to compensate for prediction errors. To validate the effectiveness of this method, an experiment is conducted in the South China Sea using the Seagull, during which relevant data are collected. These data are fed into the model in a streaming form, and the prediction performance of the model is rigorously evaluated. Experimental results indicate that, compared to an online sequential extreme learning machine with a fixed forgetting mechanism, the proposed method reduces the mean distance error by 77.4% and the root mean square error by 78.3%, thereby significantly enhancing the prediction accuracy. Furthermore, it is compared with other advanced trajectory prediction algorithms in terms of training time and prediction accuracy, to comprehensively assess the prediction performance of the proposed VFOSELM-SRP model. The results indicate that the VFOSELM-SRP model demonstrates remarkable advantages in both training time and precision aspects.