Ship abnormal trajectory identification method based on AIS data

Abstract

The invention discloses a ship abnormal track identification method based on AIS data, which belongs to the technical field of AIS data mining processing, is used for identifying an abnormal ship track, and comprises the following steps: collecting and sorting AIS historical abnormal track data and marking, and constructing a graph structure of the track data; dividing the abnormal trajectory data into a training set and a test set, training a graph convolutional neural network model by the training set, evaluating model performance by the test set, and adjusting hyper-parameters of the model; performing matching according to mmsi to obtain complete AIS data, and establishing a graph structure of trajectory data; classifying ship trajectories to be classified by using the model, and finding out individual abnormal trajectories; and processing the ship trajectory points based on a neighbor mapping Haversine grid algorithm, and finding out a group abnormal trajectory. Compared with the prior art, the method has stronger automatic feature learning ability, can effectively capture the space-time relationship and the complex trajectory mode, and improves the recognition performance and adaptability.

Description

Technical Field

[0001] The invention discloses a method for identifying abnormal ship tracks based on AIS data, and belongs to the technical field of AIS data mining and processing. Background Art

[0002] AIS is a global automatic ship identification system, which aims to improve ship navigation safety, monitor ship positions and manage traffic. Through AIS, ships can broadcast their position, speed, heading and identity information in real time, making AIS data a valuable resource for monitoring and managing global shipping activities. In recent years, with the increase in the number of ships worldwide, the monitoring of ship trajectories has become crucial. However, the boundary between normal ship movement and abnormal trajectories is blurred, and abnormal trajectories may indicate safety issues, illegal behavior or unexpected events. A technology that can automatically identify abnormal ship trajectories is essential to improve maritime safety and regulation. Traditional abnormal trajectory detection methods have some limitations, such as the inability to effectively process large-scale AIS data and the inability to identify specific types of abnormal behavior. Therefore, it is necessary to propose an efficient, accurate and adaptive ship abnormal trajectory detection method to analyze AIS data, detect potential abnormal behaviors, and provide alarm and monitoring functions.

[0003] Traditional methods for identifying abnormal ship trajectories mainly include machine learning, ship trajectory clustering, and Gaussian mixture models. These methods are effective in some cases, but they also have some potential disadvantages. Traditional methods require manual selection and extraction of features, which requires a lot of domain knowledge, resulting in the method being highly dependent on feature quality. The graph convolutional neural network model is a deep learning model specifically designed to process graph structured data. Compared with traditional methods for identifying abnormal ship trajectories, the graph convolutional neural network model has stronger automatic feature learning capabilities and can effectively capture spatiotemporal relationships and complex trajectory patterns, thereby improving recognition performance and adaptability. Summary of the invention

[0004] The object of the present invention is to provide a method for identifying abnormal ship trajectories based on AIS data, so as to solve the problems of poor performance and low adaptability in identifying abnormal ship trajectories in the prior art.

[0005] A method for identifying abnormal ship trajectories based on AIS data defines approaching and crossing behaviors as group abnormal behaviors, and defines five behaviors, namely, deviation from standard routes, wandering, unplanned AIS switching, unconventional port arrival, and entry into controlled areas, as individual abnormal behaviors, including:

[0006] S1. Collect and organize historical AIS abnormal trajectory data and annotate them to build a graph structure of trajectory data;

[0007] S2. The collected AIS historical abnormal trajectory data are divided into a training set and a test set. The graph convolutional neural network model is trained with the training set, and the model performance is evaluated with the test set. Based on the performance verification, the hyperparameters of the graph convolutional neural network model are continuously adjusted to improve the performance;

[0008] S3. According to the ship's marine mobile communication system identification code mmsi, the AIS real-time dynamic data and the data in the ship's static data database are matched to obtain complete AIS data and establish a graph structure of trajectory data for each ship;

[0009] S4. Use the trained and tested graph convolutional neural network model to classify the ship trajectories to be classified and find out the individual abnormal trajectories;

[0010] S5. Process the ship trajectory points based on the nearest neighbor mapping Haversine grid algorithm to find out the group abnormal trajectories.

[0011] S1 collects AIS historical abnormal trajectory data, including deviation from standard routes, wandering, AIS unplanned switching, unconventional port arrival, and abnormal trajectory of ships entering controlled areas. Each ship trajectory point has ship type type_aux, heading angle course, speed speed, latitude lat, longitude lon, time post_time, and ship maritime mobile communication system identification code mmsi attribute information. Different types of trajectories are marked with category labels. Departure from the standard route is represented by 0, wandering is represented by 1, AIS unplanned switching is represented by 2, unconventional port arrival is represented by 3, and entry into controlled areas is represented by 4.

[0012] S1 constructs the graph structure of trajectory data, including constructing a graph for each acquired trajectory data. AIS trajectory data exists in the form of time series. The position and trajectory of the ship change over time. The network structure of the graph is constructed according to the time change law of the trajectory.

[0013] Each trajectory point is regarded as a node in the construction graph, and the relationship between two trajectory points is used to create an edge. In the construction graph, type_aux, course, speed, lat, lon, post_time, and mmsi are used as the attribute information of the node and represented by a seven-dimensional vector. The speed difference and heading difference of the nodes at both ends of the edge are used as the attribute information of the edge and represented by a two-dimensional vector.

[0014] S2 includes building a graph convolutional neural network model. The network structure of the convolutional neural network includes the feature dimension of the input data, the number of neurons in the hidden layer, the activation function, and the output layer.

[0015] S2 includes inputting the training set into the graph convolutional neural network model, calculating the predicted value through forward propagation, calculating the performance of the loss metric model under the current parameter setting, calculating the gradient through back propagation, and updating the parameters of the model through the gradient descent algorithm to reduce the loss function;

[0016] Input the test set data into the trained graph convolutional neural network model, calculate the accuracy of the classification results, evaluate the generalization performance of the graph convolutional neural network model, and adjust the graph convolutional neural network hyperparameters based on the test results. The hyperparameters include model learning rate, number of layers, and number of hidden neurons.

[0017] S3 includes obtaining static information of ship types from websites and mobile applications. The static information of ship types includes text description of ship types and MMSI. The ship type description text is parsed according to MMSI to establish a global ship static data database.

[0018] S3 includes obtaining AIS real-time data, cleaning the original data, deleting duplicate AIS track point data, correcting or deleting missing AIS dynamic data, matching the missing static data from the global ship static data database according to the ship's mmsi, completing the ship's static information, and building a graph structure based on the track data of each ship.

[0019] S4 includes inputting graph structured data into the trained graph convolutional neural network model, classifying the trajectory data, and screening out five types of individual abnormal behavior trajectories from the AIS trajectory: deviation from the standard route, wandering, unplanned AIS switching, unconventional port arrival, and entry into the controlled area.

[0020] S5 includes using the nearest neighbor mapping Haversine grid algorithm to identify abnormal group behavior, dividing the monitoring area into rectangular grid units, and using the Haversine grid algorithm in each grid unit to calculate the distance between the trajectories of different ships at similar time points. When the distance is less than the judgment threshold, it is judged as a trajectory close to the intersection abnormal behavior.

[0021] S5 includes setting an overlap area between every two adjacent unit grids when constructing the grid units, and the overlap distance is less than the determination threshold.

[0022] Compared with the prior art, the present invention has the following beneficial effects: it has stronger automatic feature learning capability, can effectively capture spatiotemporal relationships and complex trajectory patterns, and improves recognition performance and adaptability. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 It is a schematic diagram of the process of the present invention;

[0024] Figure 2 It is a schematic diagram of the trajectory that deviates from the standard route;

[0025] Figure 3 It is a schematic diagram of the trajectory of the wandering route;

[0026] Figure 4 This is a schematic diagram of the AIS unplanned switch route trajectory;

[0027] Figure 5 This is a schematic diagram of the trajectory of the route to the unconventional port;

[0028] Figure 6 It is a schematic diagram of the trajectory of the approaching intersection route;

[0029] Figure 7 It is a schematic diagram of the route trajectory entering the control area;

[0030] Figure 8 This is a schematic diagram of the Haversine grid model;

[0031] Fig. 9 Schematic diagram of the structure of the graph convolutional neural network model. DETAILED DESCRIPTION

[0032] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention is described clearly and completely below. Obviously, the described embodiments are part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0033] A method for identifying abnormal ship trajectories based on AIS data defines approaching and crossing behaviors as group abnormal behaviors, and defines five behaviors, namely, deviation from standard routes, wandering, unplanned AIS switching, unconventional port arrival, and entry into controlled areas, as individual abnormal behaviors. The flowchart of the present invention is shown in FIG. Figure 1 As shown, Figure 1 The X in the middle represents five abnormal route trajectories, and the steps include:

[0034] S1. Collect and organize historical AIS abnormal trajectory data and annotate them to build a graph structure of trajectory data;

[0035] S2. The collected AIS historical abnormal trajectory data are divided into a training set and a test set. The graph convolutional neural network model is trained with the training set, and the model performance is evaluated with the test set. Based on the performance verification, the hyperparameters of the graph convolutional neural network model are continuously adjusted to improve the performance;

[0036] S3. According to the ship's marine mobile communication system identification code mmsi, the AIS real-time dynamic data and the data in the ship's static data database are matched to obtain complete AIS data and establish a graph structure of trajectory data for each ship;

[0037] S4. Use the trained and tested graph convolutional neural network model to classify the ship trajectories to be classified and find out the individual abnormal trajectories;

[0038] S5. Process the ship trajectory points based on the nearest neighbor mapping Haversine grid algorithm to find out the group abnormal trajectories.

[0039] S1 collects AIS historical abnormal trajectory data, including deviation from standard routes, wandering, AIS unplanned switching, unconventional port arrival, and abnormal trajectory of ships entering controlled areas. Each ship trajectory point has ship type type_aux, heading angle course, speed speed, latitude lat, longitude lon, time post_time, and ship maritime mobile communication system identification code mmsi attribute information. Different types of trajectories are marked with category labels. Departure from the standard route is represented by 0, wandering is represented by 1, AIS unplanned switching is represented by 2, unconventional port arrival is represented by 3, and entry into controlled areas is represented by 4.

[0040] S1 constructs the graph structure of trajectory data, including constructing a graph for each acquired trajectory data. AIS trajectory data exists in the form of time series. The position and trajectory of the ship change over time. The network structure of the graph is constructed according to the time change law of the trajectory.

[0041] Each trajectory point is regarded as a node in the construction graph, and the relationship between two trajectory points is used to create an edge. In the construction graph, type_aux, course, speed, lat, lon, post_time, and mmsi are used as the attribute information of the node and represented by a seven-dimensional vector. The speed difference and heading difference of the nodes at both ends of the edge are used as the attribute information of the edge and represented by a two-dimensional vector.

[0042] S2 includes building a graph convolutional neural network model. The network structure of the convolutional neural network includes the feature dimension of the input data, the number of neurons in the hidden layer, the activation function, and the output layer.

[0043] S2 includes inputting the training set into the graph convolutional neural network model, calculating the predicted value through forward propagation, calculating the performance of the loss metric model under the current parameter setting, calculating the gradient through back propagation, and updating the parameters of the model through the gradient descent algorithm to reduce the loss function;

[0044] Input the test set data into the trained graph convolutional neural network model, calculate the accuracy of the classification results, evaluate the generalization performance of the graph convolutional neural network model, and adjust the graph convolutional neural network hyperparameters based on the test results. The hyperparameters include model learning rate, number of layers, and number of hidden neurons.

[0045] S3 includes obtaining static information of ship types from websites and mobile applications. The static information of ship types includes text description of ship types and MMSI. The ship type description text is parsed according to MMSI to establish a global ship static data database.

[0046] S3 includes obtaining AIS real-time data, cleaning the original data, deleting duplicate AIS track point data, correcting or deleting missing AIS dynamic data, matching the missing static data from the global ship static data database according to the ship's mmsi, completing the ship's static information, and building a graph structure based on the track data of each ship.

[0047] S4 includes inputting graph structured data into the trained graph convolutional neural network model, classifying the trajectory data, and screening out five types of individual abnormal behavior trajectories from the AIS trajectory: deviation from the standard route, wandering, unplanned AIS switching, unconventional port arrival, and entry into the controlled area.

[0048] S5 includes using the nearest neighbor mapping Haversine grid algorithm to identify abnormal group behavior, dividing the monitoring area into rectangular grid units, and using the Haversine grid algorithm in each grid unit to calculate the distance between the trajectories of different ships at similar time points. When the distance is less than the judgment threshold, it is judged as a trajectory close to the intersection abnormal behavior.

[0049] S5 includes setting an overlap area between every two adjacent unit grids when constructing the grid units, and the overlap distance is less than the determination threshold.

[0050] The schematic diagram of the trajectory deviating from the standard route is as follows Figure 2 As shown in the figure, the trajectory diagram of the wandering route is as follows Figure 3 As shown in the figure, the AIS unplanned switch route trajectory diagram is as follows Figure 4 As shown in the figure, the trajectory diagram of the route to the unconventional port is as follows Figure 5 As shown in the figure, the trajectory diagram of the approaching intersection route is as follows Figure 6 As shown in the figure, the schematic diagram of the route trajectory entering the control area is as follows Figure 7As shown in the figure, the AIS abnormal trajectory is divided into six different abnormal trajectory types, including deviation from the standard route, wandering, unplanned AIS switch, unconventional port arrival, close intersection and entry into the control area. These different abnormal trajectory types correspond to different abnormal behaviors of ships. For example, abnormal trajectories close to the intersection may indicate that the ship has collided, or that the ship has engaged in illegal activities such as smuggling or drug trafficking at sea; while abnormal trajectories entering the control area may represent illegal fishing, illegal mining and other illegal activities at sea.

[0051] In the present invention, in order to more effectively identify the abnormal trajectory of the ship, the nearest neighbor Haversine grid model and the graph convolutional neural network are used to detect the group abnormal trajectory and the individual abnormal trajectory respectively. These methods can more accurately identify the abnormal behavior of the ship, thereby providing more effective protection for maritime safety.

[0052] The structural diagram of the graph convolutional neural network model is as follows Figure 8 As shown, G=(V n ,E n ) represents the graph structure constructed previously. n Represents the nodes of the graph, that is, the trajectory points of the ship; E n Represents the edge connecting the nodes. The constructed graph convolutional neural network model needs to undergo two convolutions, and the convolution formula of the graph is:

[0053]

[0054] Where l represents the number or level of the layer; H is the node representation matrix, each row represents each node in the graph, and each column represents the feature vector of the node; σ represents the activation function; is the normalized adjacency matrix, usually defined as Where I is the identity matrix, A is the adjacency matrix used to represent the graph, and D is the degree matrix; The purpose of is to normalize the adjacency matrix to take into account the degree of the nodes; W is the weight matrix.

[0055] The core idea of ​​the convolutional formula is to propagate information from one node to its neighbors by considering the graph structure and the connection relationship of the nodes, and then transform the features through the weight matrix and activation function. This process can be done layer by layer to gradually learn higher-level node representations, which is convenient for predicting abnormal trajectory types later.

[0056] After two layers of convolution, it enters the fully connected layer. The fully connected layer in the graph convolutional neural network model is the same as the fully connected layer in the convolutional neural network. In this architecture, the probability distribution of the potential categories of the trajectory is obtained by forward propagating the input graph of the ship trajectory, and finally the category with the highest probability is selected as the final classification result. Compare the classification result with the annotation information of the graph, calculate the loss function, and continuously adjust the parameters through the back propagation process. Finally, the model performance is evaluated using the prepared test set to understand the performance of the trained model on data that has never been seen. Through the performance evaluation of the test set, the hyperparameters of the model are adjusted to improve the performance of the model.

[0057] The structural diagram of the graph convolutional neural network model is as follows Fig. 9 As shown in the figure, the arrangement of rectangular grid units is depicted, where L represents the side length of the grid; T is the threshold used to determine whether the distance between trajectory points is abnormal; r represents the overlap distance of adjacent grids, r>T; D represents the distance between ships. For the study area P, it is divided into several subsets P = {p1, p2, …, p k}, p i The geographical range of (i=1,2,…,k) is (N i ,S i ,W i ,E i ). The AIS trajectory point set S = {S1, S2, ..., S n}Distributed to rectangular grids according to geographic range i (i=1,2,…,k). Haversine algorithm is used in each rectangular grid to calculate the distance between every two ship track points in the area. The formula is:

[0058]

[0059] Where D is the distance between the two track points of the two ships; r is the radius of the earth, usually 6371 km; lon1, lat1, lon2, lat2 are the longitude and latitude of track point 1 and track point 2 respectively.

[0060] Finally, the distance D between the ships is compared with the threshold T. If D is less than T, the navigation trajectories of the two ships are determined to be abnormal trajectories close to the intersection class.

[0061] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, a person skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some or all of the technical features may be replaced by equivalents, and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for identifying abnormal ship trajectories based on AIS data, defining approach and intersection behavior as group abnormal behavior, and defining five behaviors, namely, deviation from the standard route, wandering, unplanned AIS switching, unconventional port arrival, and entry into a controlled area, as individual abnormal behaviors, characterized in that: include: S1. Collect and organize historical AIS abnormal trajectory data and annotate them to build a graph structure of trajectory data; S2. The collected AIS historical abnormal trajectory data are divided into a training set and a test set. The graph convolutional neural network model is trained with the training set, and the model performance is evaluated with the test set. Based on the performance verification, the hyperparameters of the graph convolutional neural network model are continuously adjusted to improve the performance; S3. According to the ship's marine mobile communication system identification code mmsi, the AIS real-time dynamic data and the data in the ship's static data database are matched to obtain complete AIS data and establish a graph structure of trajectory data for each ship; S4. Use the trained and tested graph convolutional neural network model to classify the ship trajectories to be classified and find out the individual abnormal trajectories; S5. Process the ship trajectory points based on the nearest neighbor mapping Haversine grid algorithm to find out the group abnormal trajectories.

2. The method for identifying abnormal ship trajectories based on AIS data according to claim 1, characterized in that: S1 collects historical AIS abnormal trajectory data, including deviations from standard routes, wandering, unplanned AIS switching, arrival at unconventional ports, and abnormal trajectories of ships entering controlled areas. Each ship trajectory point has the ship type type_aux, heading angle course, speed speed, latitude lat, longitude lon, time post_time, and ship maritime mobile communication system identification code mmsi attribute information. Different types of trajectories are marked with category labels.

3. The method for identifying abnormal ship trajectories based on AIS data according to claim 2, characterized in that: S1 constructs the graph structure of trajectory data, including constructing a graph for each acquired trajectory data. AIS trajectory data exists in the form of time series. The position and trajectory of the ship change over time. The network structure of the graph is constructed according to the time change law of the trajectory. Each trajectory point is regarded as a node in the construction graph, and the relationship between two trajectory points is used to create an edge. In the construction graph, type_aux, course, speed, lat, lon, post_time, and mmsi are used as the attribute information of the node and represented by a seven-dimensional vector. The speed difference and heading difference of the nodes at both ends of the edge are used as the attribute information of the edge and represented by a two-dimensional vector.

4. The method for identifying abnormal ship trajectories based on AIS data according to claim 3, characterized in that: S2 includes building a graph convolutional neural network model. The network structure of the convolutional neural network includes the feature dimension of the input data, the number of neurons in the hidden layer, the activation function, and the output layer.

5. The method for identifying abnormal ship trajectories based on AIS data according to claim 4, characterized in that: S2 includes inputting the training set into the graph convolutional neural network model, calculating the predicted value through forward propagation, calculating the performance of the loss metric model under the current parameter setting, calculating the gradient through back propagation, and updating the parameters of the model through the gradient descent algorithm to reduce the loss function; Input the test set data into the trained graph convolutional neural network model, calculate the accuracy of the classification results, evaluate the generalization performance of the graph convolutional neural network model, and adjust the graph convolutional neural network hyperparameters based on the test results. The hyperparameters include model learning rate, number of layers, and number of hidden neurons.

6. The method for identifying abnormal ship trajectories based on AIS data according to claim 5, characterized in that: S3 includes obtaining static information of ship types from websites and mobile applications. The static information of ship types includes text description of ship types and MMSI. The ship type description text is parsed according to MMSI to establish a global ship static data database.

7. The method for identifying abnormal ship trajectories based on AIS data according to claim 6, characterized in that: S3 includes obtaining AIS real-time data, cleaning the original data, deleting duplicate AIS track point data, correcting or deleting missing AIS dynamic data, matching the missing static data from the global ship static data database according to the ship's mmsi, completing the ship's static information, and building a graph structure based on the track data of each ship.

8. The method for identifying abnormal ship trajectories based on AIS data according to claim 7, characterized in that: S4 includes inputting graph structured data into the trained graph convolutional neural network model, classifying the trajectory data, and screening out five types of individual abnormal behavior trajectories from the AIS trajectory: deviation from the standard route, wandering, unplanned AIS switching, unconventional port arrival, and entry into the controlled area.

9. The method for identifying abnormal ship trajectories based on AIS data according to claim 8, characterized in that: S5 includes using the nearest neighbor mapping Haversine grid algorithm to identify abnormal group behavior, dividing the monitoring area into rectangular grid units, and using the Haversine grid algorithm in each grid unit to calculate the distance between the trajectories of different ships at similar time points. When the distance is less than the judgment threshold, it is judged as a trajectory close to the intersection abnormal behavior.

10. The method for identifying abnormal ship trajectories based on AIS data according to claim 9, characterized in that: S5 includes setting an overlap area between every two adjacent unit grids when constructing the grid units, and the overlap distance is less than the determination threshold.

Images