为更准确地预测内河船舶交通流，提出基于注意力机制的CNN-GRU船舶交通流预测模型。该模型主要借助一维卷积单元提取数据的高维特征，GRU单元学习数据中的时序特征，并通过引入注意力机制加强重要特征的学习，实现对超长序列的学习。此外，通过分析内河上下游航道交通流间的关联性，提取长江中下游6个航段的船舶AIS数据，构造多航段船舶交通流序列数据集，并将其输入本文模型中进行训练及测试。结果表明：相比序列预测模型中的SAE、LSTM、GRU、CNN+GRU和GRU+Attention，本文模型在针对不同交通流参数的预测中均具有更高的预测精度，交通流量、交通流密度和交通流速度的预测精度分别达到95.42%、97.33%、94.99%，可更好地满足工程应用需求。

In order to predict inland waterway ship traffic flow more accurately, a CNN-GRU ship traffic flow prediction model was proposed based on attention mechanism. The model mainly extracted the high-dimensional features of the data by means of  one-dimensional convolution unit. The GRU unit learned the temporal characteristics in the data and enhanced the learning of important features by introducing attention mechanism to realize the learning of ultra-long sequences. In addition, by analyzing the correlation between the upper and lower waterway traffic flows, the ship AIS data of 6 sections in  the middle and lower reaches of the Yangtze River were extracted to construct the multi-segment ship traffic flow sequence data set and then input into the proposed model for training and testing. The  results show that compared with SAE, LSTM, GRU, CNN+GRU and GRU+Attention in the series prediction model,the proposed model has higher prediction accuracy in the prediction of different traffic flow parameters, and the prediction accuracy of traffic flow, traffic flow density and traffic flow speed are 95.42%, 97.33% and 94.99% respectively, which can better meet the needs of engineering applications.