Figure 1.
Trajectory examples for various vessel types, illustrate significant differences in navigation trajectories under the influence of various vessel attributes and types.
Figure 1.
Trajectory examples for various vessel types, illustrate significant differences in navigation trajectories under the influence of various vessel attributes and types.
Figure 2.
The process of data preprocessing.
Figure 2.
The process of data preprocessing.
Figure 3.
Label-level modeling and analysis process.
Figure 3.
Label-level modeling and analysis process.
Figure 4.
The trajectory cluster result. Different colors represent different clusters, and the black labels indicate cluster centers.
Figure 4.
The trajectory cluster result. Different colors represent different clusters, and the black labels indicate cluster centers.
Figure 5.
The structure of VEPO-S2S comprises the Multi-level Vessel Trajectory Representation Module (Multi-Rep) and the Feature Fusion and Decoding Module (FFDM). The Multi-Rep is designed to obtain trajectory information and Multi-level Vessel Characteristics, applying distinct encoders for encoding. The FFDM is targeted to select and integrate the above characteristics from Multi-Rep for prediction.
Figure 5.
The structure of VEPO-S2S comprises the Multi-level Vessel Trajectory Representation Module (Multi-Rep) and the Feature Fusion and Decoding Module (FFDM). The Multi-Rep is designed to obtain trajectory information and Multi-level Vessel Characteristics, applying distinct encoders for encoding. The FFDM is targeted to select and integrate the above characteristics from Multi-Rep for prediction.
Figure 6.
The Portrait Selection Component consists of prior distribution and posterior distribution. Prior distribution expresses the trajectory coding vector and portrait feature , and the posterior distribution incorporates the label y to improve the accuracy of selection. Meanwhile, the KLD is designed to bridge the gap between the prior distribution and the posterior distribution, allowing the prior distribution to benefit from the posterior distribution and generate more accurate results.
Figure 6.
The Portrait Selection Component consists of prior distribution and posterior distribution. Prior distribution expresses the trajectory coding vector and portrait feature , and the posterior distribution incorporates the label y to improve the accuracy of selection. Meanwhile, the KLD is designed to bridge the gap between the prior distribution and the posterior distribution, allowing the prior distribution to benefit from the posterior distribution and generate more accurate results.
Figure 7.
The Multi-head Decoder Component consists of two GRU blocks and a fusion unit. It can flexibly adjust the weighting between the trajectory information and the vessel characteristics during the prediction process.
Figure 7.
The Multi-head Decoder Component consists of two GRU blocks and a fusion unit. It can flexibly adjust the weighting between the trajectory information and the vessel characteristics during the prediction process.
Figure 8.
The distribution of vessel type in our processed dataset.
Figure 8.
The distribution of vessel type in our processed dataset.
Figure 9.
An example of dividing a dataset using the sliding window method, and the red dots represent trajectory points.
Figure 9.
An example of dividing a dataset using the sliding window method, and the red dots represent trajectory points.
Figure 10.
The influence of different GRU layers on the prediction of the VEPO-S2S model, according to the RMSE. The X-axis represents the number of layers, and the Y-axis represents the RMSE loss value.
Figure 10.
The influence of different GRU layers on the prediction of the VEPO-S2S model, according to the RMSE. The X-axis represents the number of layers, and the Y-axis represents the RMSE loss value.
Figure 11.
The influence of different hidden sizes on the prediction of the VEPO-S2S model, according to the RMSE. The X-axis represents the number of hidden sizes, and the Y-axis represents the RMSE loss value.
Figure 11.
The influence of different hidden sizes on the prediction of the VEPO-S2S model, according to the RMSE. The X-axis represents the number of hidden sizes, and the Y-axis represents the RMSE loss value.
Figure 12.
The predictions of cargo ships and container ships under various models, with the difficulty of predictions increasing from (a) to (b). Our model performs the best in both scenarios. In (a), which involves straight-line navigation of cargo ships, all models except GRU achieve decent prediction results. In (b), which involves container ship turning. Additionally, the METO-S2S model is also able to accomplish the prediction tasks to some extent. However, other models struggle to achieve satisfactory prediction performance.
Figure 12.
The predictions of cargo ships and container ships under various models, with the difficulty of predictions increasing from (a) to (b). Our model performs the best in both scenarios. In (a), which involves straight-line navigation of cargo ships, all models except GRU achieve decent prediction results. In (b), which involves container ship turning. Additionally, the METO-S2S model is also able to accomplish the prediction tasks to some extent. However, other models struggle to achieve satisfactory prediction performance.
Figure 13.
The trajectory predictions of the cargo ship and oil tanker using different structures of the VEPO-S2S model, with the predictive difficulty increasing gradually from (a) to (b). As depicted in the graph, VEPO-GRU-GRU achieves the best predictive performance.
Figure 13.
The trajectory predictions of the cargo ship and oil tanker using different structures of the VEPO-S2S model, with the predictive difficulty increasing gradually from (a) to (b). As depicted in the graph, VEPO-GRU-GRU achieves the best predictive performance.
Figure 14.
The predicted trajectories of a tugboat, where the green and red lines are the prediction results using VEPO-S2S with and without Shallow-level Attributes, respectively. The Shallow-level Attributes are associated with the vessel’s inertia and turning capabilities. The model without Shallow-level Attributes is not able to grasp this ability well, which may cause errors.
Figure 14.
The predicted trajectories of a tugboat, where the green and red lines are the prediction results using VEPO-S2S with and without Shallow-level Attributes, respectively. The Shallow-level Attributes are associated with the vessel’s inertia and turning capabilities. The model without Shallow-level Attributes is not able to grasp this ability well, which may cause errors.
Figure 15.
The predicted trajectories of a tugboat, where the green and red lines represent the prediction results of VEPO-S2S with and without considering the Sailing Location Preference, respectively. The Sailing Location Preference helps the model identify the adaptability of the vessel to the geographical environment. When the Sailing Location Preference is not considered, the VEPO-S2S model is not correct.
Figure 15.
The predicted trajectories of a tugboat, where the green and red lines represent the prediction results of VEPO-S2S with and without considering the Sailing Location Preference, respectively. The Sailing Location Preference helps the model identify the adaptability of the vessel to the geographical environment. When the Sailing Location Preference is not considered, the VEPO-S2S model is not correct.
Figure 16.
The predicted trajectories of a tugboat, where the green and red lines represent the prediction results of the VEPO-S2S model with and without considering the Voyage Time Preference, respectively. The Voyage Time Preference is related to the habits of the crews. When the Voyage Time Preference is not considered, the VEPO-S2S model may produce inaccurate predictions.
Figure 16.
The predicted trajectories of a tugboat, where the green and red lines represent the prediction results of the VEPO-S2S model with and without considering the Voyage Time Preference, respectively. The Voyage Time Preference is related to the habits of the crews. When the Voyage Time Preference is not considered, the VEPO-S2S model may produce inaccurate predictions.
Figure 17.
The predicted trajectories of a tugboat, where the green and red lines represent the prediction results of the VEPO-S2S model with and without considering the Anchoring Time Preference, respectively. The Anchoring Time Preference helps the model identify the working and resting habits of the vessel. In the absence of the Anchoring Time Preference, the model produces an incorrect estimation for each timestamp.
Figure 17.
The predicted trajectories of a tugboat, where the green and red lines represent the prediction results of the VEPO-S2S model with and without considering the Anchoring Time Preference, respectively. The Anchoring Time Preference helps the model identify the working and resting habits of the vessel. In the absence of the Anchoring Time Preference, the model produces an incorrect estimation for each timestamp.
Figure 18.
The predicted trajectories of a tugboat, where the green and red lines are the prediction results using VEPO-S2S with and without the Portrait Selection Component, respectively. The Portrait Selection Component is responsible for selecting the most relevant characteristics for prediction. Without this component, the model cannot select appropriate characteristics to assist in prediction, leading to a decrease in model robustness.
Figure 18.
The predicted trajectories of a tugboat, where the green and red lines are the prediction results using VEPO-S2S with and without the Portrait Selection Component, respectively. The Portrait Selection Component is responsible for selecting the most relevant characteristics for prediction. Without this component, the model cannot select appropriate characteristics to assist in prediction, leading to a decrease in model robustness.
Figure 19.
The predicted trajectories of a tugboat, where green and red lines are the prediction results by VEPO-S2S with and without Feature Fused Component, respectively. The Feature Fused Component effectively integrates trajectory information and vessel characteristics and increases the correlation between both. Without the Feature Fused Component, vessel characteristics are difficult to express adequately in the VEPO-S2S model, leading to a decrease in model accuracy.
Figure 19.
The predicted trajectories of a tugboat, where green and red lines are the prediction results by VEPO-S2S with and without Feature Fused Component, respectively. The Feature Fused Component effectively integrates trajectory information and vessel characteristics and increases the correlation between both. Without the Feature Fused Component, vessel characteristics are difficult to express adequately in the VEPO-S2S model, leading to a decrease in model accuracy.
Figure 20.
The predicted trajectories of a tugboat, where the green and red lines are the prediction results using VEPO-S2S with and without the Multi-head Decoder Component, respectively. The Multi-head Decoder Component regulates the involvement of the trajectory information and vessel characteristics in the prediction process. Without it, the adaptability of the VEPO-S2S model decreases.
Figure 20.
The predicted trajectories of a tugboat, where the green and red lines are the prediction results using VEPO-S2S with and without the Multi-head Decoder Component, respectively. The Multi-head Decoder Component regulates the involvement of the trajectory information and vessel characteristics in the prediction process. Without it, the adaptability of the VEPO-S2S model decreases.
Table 1.
The detail of our dataset.
Table 1.
The detail of our dataset.
Dataset | Region | Track Points Count | Type Count | Vessel Count |
---|
Total | Coast of the United States | 144,445,580 | 68 | 28,645 |
Ours | Coast of the United States | 4,930,061 | 45 | 6194 |
Table 2.
Comparison results of VEPO-S2S with various baselines under RMSE evaluation metric. Here, 10->5 represents the RMSE value of the last 5 trajectories predicted by 10 historical trajectories.
Table 2.
Comparison results of VEPO-S2S with various baselines under RMSE evaluation metric. Here, 10->5 represents the RMSE value of the last 5 trajectories predicted by 10 historical trajectories.
Model Name | 10->1 | 10->2 | 10->3 | 10->4 | 10->5 |
---|
Kalman | | | | | |
VAR | | | | - | - |
ARIMA | | | | | |
LSTM | | | | | |
BiLSTM | | | | | |
GRU | | | | | |
BiGRU | | | | | |
LSTM-LSTM | | | | | |
BiLSTM-LSTM | | | | | |
GRU-GRU | | | | | |
BiGRU-GRU | | | | | |
Transformer | | | | | |
METO-S2S | | | | | |
Ours | | | | | |
Table 3.
Comparison results of VEPO-S2S with various baselines under MAE evaluation metric. Here, 10->5 means predicted through 10 historical trajectories.
Table 3.
Comparison results of VEPO-S2S with various baselines under MAE evaluation metric. Here, 10->5 means predicted through 10 historical trajectories.
Model Name | 10->1 | 10->2 | 10->3 | 10->4 | 10->5 |
---|
Kalman | | | | | |
VAR | | | | - | - |
ARIMA | | | | | |
LSTM | | | | | |
BiLSTM | | | | | |
GRU | | | | | |
BiGRU | | | | | |
LSTM-LSTM | | | | | |
BiLSTM-LSTM | | | | | |
GRU-GRU | | | | | |
BiGRU-GRU | | | | | |
Transformer | | | | | |
METO-S2S | | | | | |
Ours | | | | | |
Table 4.
Comparison results of VEPO-S2S with various baselines under ADE evaluation metric. Here, 10->5 means predicted through 10 historical trajectories.
Table 4.
Comparison results of VEPO-S2S with various baselines under ADE evaluation metric. Here, 10->5 means predicted through 10 historical trajectories.
Model Name | 10->1 | 10->2 | 10->3 | 10->4 | 10->5 |
---|
Kalman | | | | | |
VAR | | | | - | - |
ARIMA | | | | | |
LSTM | | | | | |
BiLSTM | | | | | |
GRU | | | | | |
BiGRU | | | | | |
LSTM-LSTM | | | | | |
BiLSTM-LSTM | | | | | |
GRU-GRU | | | | | |
BiGRU-GRU | | | | | |
Transformer | | | | | |
METO-S2S | | | | | |
Ours | | | | | |
Table 5.
Exploration results on different Seq2Seq structure.
Table 5.
Exploration results on different Seq2Seq structure.
Model Name | Encoder | Decoder | RMSE | MAE | ADE | FDE |
---|
VEPO-S2S | BiGRU | GRU | | | | |
VEPO-S2S | LSTM | LSTM | | | | |
VEPO-S2S | BiLSTM | LSTM | | | | |
VEPO-S2S | GRU | GRU | | | | |
Table 6.
The quantitative analysis results of each trajectory point under the RMSE evaluation index.
Table 6.
The quantitative analysis results of each trajectory point under the RMSE evaluation index.
Model Name | First | Second | Third | Fourth | Fifth |
---|
Kalman | | | | | |
VAR | | | | - | - |
ARIMA | | | | | |
LSTM | | | | | |
BiLSTM | | | | | |
GRU | | | | | |
BiGRU | | | | | |
LSTM-LSTM | | | | | |
BiLSTM-LSTM | | | | | |
GRU-GRU | | | | | |
BiGRU-GRU | | | | | |
Transformer | | | | | |
METO-S2S | | | | | |
Ours | | | | | |
Table 7.
The quantitative analysis results of each trajectory point under the MAE evaluation index.
Table 7.
The quantitative analysis results of each trajectory point under the MAE evaluation index.
Model Name | First | Second | Third | Fourth | Fifth |
---|
Kalman | | | | | |
VAR | | | | - | - |
ARIMA | | | | | |
LSTM | | | | | |
BiLSTM | | | | | |
GRU | | | | | |
BiGRU | | | | | |
LSTM-LSTM | | | | | |
BiLSTM-LSTM | | | | | |
GRU-GRU | | | | | |
BiGRU-GRU | | | | | |
Transformer | | | | | |
METO-S2S | | | | | |
Ours | | | | | |
Table 8.
The quantitative analysis results of each trajectory point under the FDE evaluation index.
Table 8.
The quantitative analysis results of each trajectory point under the FDE evaluation index.
Model Name | First | Second | Third | Fourth | Fifth |
---|
Kalman | | | | | |
VAR | | | | - | - |
ARIMA | | | | | |
LSTM | | | | | |
BiLSTM | | | | | |
GRU | | | | | |
BiGRU | | | | | |
LSTM-LSTM | | | | | |
BiLSTM-LSTM | | | | | |
GRU-GRU | | | | | |
BiGRU-GRU | | | | | |
Transformer | | | | | |
METO-S2S | | | | | |
Ours | | | | | |
Table 9.
Quantitative results on different ablation studies.
Table 9.
Quantitative results on different ablation studies.
Ablation | RMSE | MAE | ADE | FDE |
---|
VEPO-S2S | | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |