基于注意力机制的TCN-BiLSTM船舶轨迹预测

Ship Trajectory Prediction of TCN-Bi-LSTM Based on Attention Mechanism

摘要:

针对现有船舶轨迹预测模型预测准确度低的问题，提出一种基于注意力机制的时域卷积网络和双向长短时记忆网络（TCN-ABiLSTM）的船舶轨迹预测模型。首先搭建TCN网络提取船舶轨迹的序列特征，之后将注意力机制引入网络调整不同属性特征的权值，凸出对轨迹预测影响更大的特征，最后搭建Bi-LSTM网络学习轨迹序列的前后状况来提取序列中更多的信息，实现对船舶未来轨迹的预测；通过实际船舶AIS数据对网络进行训练与测试实验，实验结果表明，TCN-ABiLSTM模型相比LSTM、Bi-LSTM、TCN、BiLSTM-Attention、TCN-Attention模型船舶轨迹预测精度更高，拟合程度更好，验证了所设计的TCN-ABiLSTM模型在船舶轨迹预测方面的的有效性和实用性。

Abstract:

A ship trajectory prediction model based on attention mechanism time-domain convolutional network and bidirectional long short memory network (TCN-ABiLSTM) is proposed to address the issue of low prediction accuracy in existing ship trajectory prediction models. Firstly, TCN network is constructed to extract the sequence features of ship trajectories. Then, attention mechanism is introduced into the network to adjust the weights of different attribute features, highlighting the features that have a greater impact on trajectory prediction. Finally, Bi-LSTM network is constructed to learn the pre and post situation of trajectory sequences to extract more information from the sequences, achieving prediction of future ship trajectories; Training and testing experiments are conducted on the network using actual ship AIS data. The experimental results show that the TCN-ABiLSTM model has higher accuracy and better fit in predicting ship trajectories compared to LSTM, Bi LSTM, TCN, BiLSTM Attention, and TCN-Attention models. This verifies the effectiveness and practicality of the designed TCN-ABiLSTM model in predicting ship trajectories.