水上飞机运动极短期预报可以提供未来数秒的摇荡运动姿态，对确保大风浪条件下的起飞与降落阶段的安全有重要意义。目前已有相关水上飞机运动极短期预报方法研究，但是对不同方法适用性差异分析的研究较少。针对该情况，以NACA TN 2929飞机为例，基于势流理论计算其在典型工况下三自由度运动仿真数据，通过构建AR（auto regressive）、LSTM（long short term memory）和TCN（temporal convolutional network）三种典型运动极短期预报模型，对比三种模型在不同预报条件下的预报性能。结果表明：与AR模型相比，以LSTM、TCN为代表的神经网络模型在预报时长较长时具有更好的预报精度，可较好地实现对水上飞机的垂荡、横摇和纵摇运动十秒级的准确预报，为水上飞机运动预报算法的选取提供较好的理论参考价值。

The extremely short-term prediction of seaplane motion can provide the rocking motion posture over the next few seconds, which is considered essential for ensuring safety during take-off and landing phases under adverse wind and wave conditions. Although some research has been conducted on extremely short-term prediction methods for seaplane motion, limited attention has been given to analyzing differences in the applicability of various methods. In this context, the NACA TN 2929 aircraft was taken as an example, and the three degree of freedom motion simulation data under typical working conditions were calculated based on potential flow theory. To compare the forecasting performance under different forecasting conditions, three typical extremely short-term prediction of seaplane motion models, namely AR (auto-regressive), LSTM (long short term memory), and TCN (temporal convolutional network), were constructed. The results show that compared to the AR model, the LSTM and TCN neural network models exhibit superior forecasting accuracy for longer prediction durations, effectively enabling accurate predictions of the heave, roll, and pitch motions of the seaplane at the ten-second level, providing a valuable theoretical reference for the selection of seaplane motion prediction algorithms.