A Ship Trajectory Prediction Framework Based on a Recurrent Neural Network

Abstract

Ship trajectory prediction is a key requisite for maritime navigation early warning and safety, but accuracy and computation efficiency are major issues still to be resolved. The research presented in this paper introduces a deep learning framework and a Gate Recurrent Unit (GRU) model to predict vessel trajectories. First, series of trajectories are extracted from Automatic Identification System (AIS) ship data (i.e., longitude, latitude, speed, and course). Secondly, main trajectories are derived by applying the Density-Based Spatial Clustering of Applications with Noise (DBSCAN) algorithm. Next, a trajectory information correction algorithm is applied based on a symmetric segmented-path distance to eliminate the influence of a large number of redundant data and to optimize incoming trajectories. A recurrent neural network is applied to predict real-time ship trajectories and is successively trained. Ground truth data from AIS raw data in the port of Zhangzhou, China were used to train and verify the validity of the proposed model. Further comparison was made with the Long Short-Term Memory (LSTM) network. The experiments showed that the ship's trajectory prediction method can improve computational time efficiency even though the prediction accuracy is similar to that of LSTM.

Keywords: DBSCAN; GRU; LSTM; deep learning; redundant data; trajectory prediction.

Figure 1

Flowchart of the vessel trajectory prediction framework.

Figure 2

A visualization of the original AIS data in Zhangzhou, Fujian, China. The green line indicates the trajectory, and darker color indicates greater trajectory density. For example, the red area indicates the area with a large trajectory density.

Figure 3

This is the heat map of the trajectory data processed by the algorithm; the trajectory in the state of drifting, anchoring, etc. is cleared, and the trajectory data in the normal sailing state is saved.

Figure 4

The two forms of AIS data ship speed’s distribution picture: (a) box diagram of ship speed, from which it can be seen that the speed is mainly distributed around 15, and a speed greater than 60 is an abnormal speed; and (b) probability distribution diagram, where normal data accounts for 98.98% of the total data.

Figure 5

Plot of the comparison of the original trajectory and optimized trajectory of a same vessel.

Figure 6

Trajectory clusters by DBSCAN. Each color cluster represents a different densest waterway.

Figure 7

After gridding, the shaded area represents the grid for counting the number of times the ship has passed through the cross section.

Figure 8

The gridded area shows the number of crossings recorded by each grid in the form of frequency.

Figure 9

The three representative common vessel trajectories: (a) straight trajectory entering a port; (b) circuitous trajectory in narrow seas; and (c) trajectory with large turns in wide seas.

Figure 10

Distance from point $p_1^2$ to $T^1$. $S_i^2$ represents segments. The distance of $P_1^2$ between $S_3^2$ is the minimum.

Figure 11

SPD distance from the trajectory $T^1$ to the trajectory $T^2$.

Figure 12

Schematic diagram of the trajectory prediction model based on the GRU neural network: (a) GRU unit at the last time; (b) GRU unit of the current time, which describes the detailed information of the transfer process; and (c) GRU unit at the next time.

Figure 13

Schematic diagram of the GRU neural network hidden layer structure.

Figure 14

Plot of all parameters setting of GRU recurrent neural network model.

Figure 15

Plot of comparison of iterative convergence of two models.

Figure 16

Plot of comparison of prediction accuracy diagram of two models.

Figure 17

Plot of comparison of a real trajectory and fitting trajectory uses the GRU model.

Figure 18

Prediction errors of the three methods calculated by equation (25): gray solid circle markers indicate the prediction results of LSTM, blue solid triangle indicate those of EKF, and the green solid stars markers denote those of GRU model.

Figure 19

Visualization of the vessel’s real trajectory and predicted trajectory that uses the GRU model.

Figure 20

A detailed representation of the vessel’s real trajectory and predicted trajectory that uses the GRU model.