Multi-ship meeting similar scene identification method based on topological graph sequence

Abstract

The invention discloses a multi-ship meeting similar scene identification method based on a topological graph sequence. The method comprises the following steps: 1) collecting multi-ship meeting scene information; 2) extracting a multi-ship meeting scene; the method comprises the following steps: constructing a multi-ship meeting topological graph sequence, and extracting a multi-ship meeting scene; 3) identifying similar scenes encountered by multiple ships; and 4) arranging a plurality of graph sequences in an identification result queue obtained by scene hierarchical identification from small to large according to the similarity of the graph sequences to obtain a final result, and completing the acquisition of the similar scene of the multi-ship encounter scene. Aiming at the problem that a similar scene is difficult to find through similarity measurement in a multi-ship meeting scene, the invention provides a multi-ship meeting scene identification method based on a topological graph sequence, and a research result can provide a method reference for related research of ship collision avoidance decision and meeting risk analysis.

Description

TECHNICAL FIELD

[0001] The present application relates to ship collision avoidance technology, and in particular to a multi-ship encounter similar scenario recognition method based on a topological graph sequence. BACKGROUND

[0002] Multi-ship encounter situations often occur in busy waterways, and such complex ship encounter scenarios often bring difficulties to ship collision avoidance. An effective means for this problem is to learn good collision avoidance habits from historical similar scenarios, and the core is the extraction of encounter scenarios and the recognition of similar scenarios. By extracting ship encounter scenarios from AIS data and comparing similar scenarios, ships can learn from others' experiences and decisions and take appropriate collision avoidance measures.

[0003] Current research on ship encounter scenarios has attracted widespread attention. Ma Jie et al. proposed an improved support vector machine-based encounter situation recognition model based on relative distance and heading difference between two ships, which effectively identified ship encounter situations. Long Hui et al. extracted ship encounter scenarios by calculating relative motion parameters and according to collision avoidance rules and maritime experience. Rong et al. proposed an improved sliding window algorithm-based encounter scenario extraction method that automatically identified ship maneuvering behavior based on ship trajectories. Zhang Jinfen et al. identified encounter scenarios based on ship trajectories and quantified real-time ship collision risk. Hwang T et al. proposed a data-driven navigation situation clustering model that classified ship encounter situations into ordinary and special encounter scenarios. Zhang et al. constructed a ship encounter recognition model based on ship encounter characteristics, designed a collision avoidance behavior extraction algorithm, and established a behavior knowledge base. The above researches mainly focus on the recognition and extraction of encounter scenarios, and the similarity measurement between encounter scenarios is relatively less, which makes it difficult to develop accurate and effective collision avoidance strategies. Therefore, it is necessary to explore and develop a multi-ship encounter scenario similarity measurement method. SUMMARY

[0004] The technical problem to be solved by the present application is to provide a multi-ship encounter similar scenario recognition method based on a topological graph sequence in view of the defects in the prior art.

[0005] The technical solution adopted by the present application to solve the technical problem is: a multi-ship encounter similar scenario recognition method based on a topological graph sequence, comprising the following steps: 1) Multi-ship encounter scenario information collection; Multi-ship encounter situation is defined as: when 3 or more ships approach each other, any ship at a certain time forms an encounter with at least one ship, which constitutes a multi-ship encounter situation; Multi-ship encounter scenario is defined as: a combination of consecutive multi-ship encounter situations in chronological order; Obtain multi-ship encounter scene information, including ship type, ship quantity, ship MMSI, and ship navigation information; 2) Multi-ship encounter scene extraction; By constructing a multi-ship encounter topology graph sequence, multi-ship encounter scene extraction is performed; Ship information and encounter relationship are taken as node features and edge attributes of the topology graph, and the topology graph with time and space information is defined as a time sequence graph; a multi-ship encounter topology graph sequence is constructed by combining multiple time sequence graphs in time sequence to represent the dynamic changes of ship encounters in the scene; 3) Multi-ship encounter similar scene identification; 3.1) Use of graph convolutional neural network for similar scene overall identification; By compressing the topology relationship of the scene graph sequence, a representative topology graph representing the entire scene is generated as input; The GCN model is used to classify the input representative topology graph, and the representative graph category is determined according to the output graph category probability distribution, thereby realizing the similarity measurement of the overall topology relationship of the scene; 3.2) Similar scene hierarchical identification The hierarchical identification method based on graph structure is used to realize the identification of the internal state of the scene: The graph sequence model is split into graph units, topology graphs, and graph sequences according to the graph structure, and the similarity of the three-layer graph structure is calculated layer by layer to measure the similarity between dynamic graph models, thereby completing the hierarchical identification of similar scenes; 4) The identification result queue obtained by the hierarchical identification of the scene has multiple graph sequences, which are arranged in order of increasing similarity of the graph sequences to obtain the final result, thereby completing the acquisition of similar scenes of multi-ship encounter scenes.

[0006] According to the above scheme, in step 1), for the ships included in the earliest multi-ship encounter situation in the scene, the time when the ships form an encounter is recorded as the starting time of the multi-ship encounter scene; for the ships included in the starting time encounter situation ship s If at least one ship does not form an encounter relationship with ship s another ship in the scene at a certain time, the time is recorded as the end time of the scene.

[0007] According to the above scheme, in step 2), the multi-ship encounter topology graph sequence is constructed as follows: 2.1) Topology graph edge relationship construction Three encounter characteristic parameters, relative distance, DCPA, and TCPA, are used to determine the encounter relationship between ships: at i the time, when j the ship andj When the encounter characteristic parameters between the ships all satisfy the threshold value, it is determined that the ships form an encounter relationship at the current time; 2.2) Node attribute definition MMSI and ship type are selected as the node characteristics of the topology graph, and the topology graph is represented by an undirected graph , where X is the node set of the graph G, containing ship information and the number of ships contained in the current graph; E is the edge set, containing the encounter relationship between ships; 2.3) Scene graph sequence construction; The Jaccard coefficient is calculated to determine the encounter scene to which it belongs: When the Jaccard coefficient of the two graphs is greater than the threshold value, it indicates that they exist in the same encounter scene; The topology graphs belonging to the same scene are clustered to construct a graph sequence model representing the complete scene , where Y represents the graph sequence characteristics of C, including the duration of the scene, the starting time; S represents the number of time sequence graphs contained in C.

[0008] According to the above scheme, in step 3.1), the representative topology graph generation process is as follows: The representative topology graph containing all graph structures in the scene is obtained by aggregating the point and edge sets of all time sequence graphs in the scene; Feature information extraction: The extraction of the representative topology graph adjacency matrix is used to complete the extraction of the ship encounter relationship in the scene; Each node of the representative topology graph has N features, denoted as , and MMSI and ship type are input as node characteristics.

[0009] According to the above scheme, in step 3.2), the similar graph unit identification is as follows: 3.2.1) Similar graph unit identification; 3.2.1.1) Extract graph units; A graph unit is a partial representation of the internal structure of a time sequence graph, and a graph unit is composed of a source node and multiple adjacent nodes , denoted as ; where the node attribute represents the ship type; For a time sequence graph , each node contained in the time sequence graph is sequentially extracted as a source node to extract a graph unit; 3.2.1.2) Calculate the similarity of graph units; The node properties and the connection between nodes are selected as the elements of the graph unit to calculate the similarity of the graph units. For two graph units and The similarity of the graph units is defined as follows:

[0010] wherein, , represent the number of adjacent nodes of and respectively; is the number of adjacent nodes shared by the two graph units; is the number of edges shared by the two graph units, which is calculated as follows:

[0011] is the similarity of the source nodes , of the two graph units, which is calculated as follows:

[0012] 3.2.1.3) Similar graph unit identification; The similarity of the query graph unit with all the graph units in the database is calculated and screened to obtain a queue of graph units with the smallest distance. Then, the adaptive evaluation score is used to eliminate the graph units in the queue with a distance less than , so as to screen the graph units in the database that may have similarity with the query graph unit. According to the above scheme, in the step 3.2), the similar time sequence graph identification is as follows: 3.2.2) Similar time sequence graph identification; 3.2.2.1) Time sequence graph similarity; Each time sequence graph is an unordered set composed of graph units. The calculation of the time sequence graph similarity is the best allocation of two sets of graph units. 3.2.2.2) Similar time sequence graph identification; For the time sequence graph , wherein each graph unit finds its similar graph unit queue through the graph unit identification, denoted as , so M similar graph unit queues are obtained; according to the elements in the queue, the time sequence graph containing the similar graph unit is found from the database, so that L time sequence graphs with similar relationship with the query graph are screened out, and a graph association list is formed therefrom; ​Similarity calculation is performed on each time sequence diagram in the association list, and screening is performed, and a qualified similar diagram sequence is obtained based on the evaluation score According to the above scheme, in the step 3.2), the similar diagram unit identifies as follows: 3.2.3) Similar diagram sequence identification 3.2.3.1) Similarity of diagram sequence Diagram sequence containing n diagrams The time sequence can be regarded as being composed of a plurality of time slices, so the similarity of the diagram sequence is regarded as a problem of calculating the similarity of the time sequence, and dynamic time warping is introduced to define the similarity between the diagram sequences. For two diagram sequences 、 The similarity of the diagram sequence is calculated as follows:

[0013] 3.2.3.2) Similar diagram sequence identification For the diagram sequence , firstly, n similar time sequence diagram queues are found through a time sequence diagram identification algorithm, and an association list of diagram sequences is formed according to the time sequence diagrams in the queues; then The association list is traversed, and a similar diagram sequence queue is obtained based on the evaluation score The beneficial effects of the present application are as follows: The present application proposes a multi-ship encounter scene identification method based on a topological diagram sequence, aiming at the problem that it is difficult to find a similar scene through similarity measurement in a multi-ship encounter scene, and the research results can provide method reference for ship collision avoidance decision and encounter risk analysis related research. BRIEF DESCRIPTION OF DRAWINGS

[0014] The present application will be further described below in combination with the drawings and embodiments, and the drawings are as follows: Figure 1 is a method flowchart of an embodiment of the present application; Figure 2 is a multi-ship encounter scene diagram sequence construction schematic diagram of an embodiment of the present application; Figure 3 is a diagram convolutional neural network schematic diagram of an embodiment of the present application; Figure 4 is a scene representative diagram generation schematic diagram of an embodiment of the present application; Figure 5 is a GCN model accuracy rate curve schematic diagram of an embodiment of the present application; Figure 6 is a GCN model loss function curve schematic diagram of an embodiment of the present application; Figure 7 is a multi-ship encounter scene hierarchical identification schematic diagram of an embodiment of the present application;​​ Figure 8 is a graph unit extraction schematic diagram of an embodiment of the present application; Figure 9 is a graph similarity calculation schematic diagram of an embodiment of the present application; Figure 10 is a weight matrix schematic diagram of an embodiment of the present application. DETAILED DESCRIPTION

[0015] In order to make the objects, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with reference to embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and not used to limit the present application.

[0016] The present application aims at the problem that a multi-ship encounter scene is difficult to find a similar scene through similarity measurement, and proposes a multi-ship encounter scene recognition method based on a topological graph sequence. First, multi-ship encounter scenes are identified and extracted from AIS data and converted into a graph sequence model; then, a similar scene recognition algorithm is used for graph sequence similarity calculation to recognize multi-ship encounter scenes similar to the original scene. The experimental results show that the proposed method can recognize encounter scenes with similar encounter evolution trends, and can recognize the encounter situation of corresponding ships in the scene.

[0017] As shown in Figure 1 , a multi-ship encounter similar scene recognition method based on a topological graph sequence includes the following steps: 1) Multi-ship encounter scene information acquisition; The multi-ship encounter situation is defined as: when 3 or more ships approach each other, at a certain moment, any ship forms an encounter with at least one ship, which constitutes a multi-ship encounter situation; The multi-ship encounter scene is defined as: the combination of continuous multi-ship encounter situations in chronological order; The multi-ship encounter scene information includes ship type, ship number, ship MMSI, and ship navigation information; the navigation information includes speed, heading, and position information; 2) Multi-ship encounter scene extraction; The multi-ship encounter scene extraction is performed by constructing a multi-ship encounter topological graph sequence; The ship information and encounter relationship are used as the node features and edge attributes of the topological graph, and this type of topological graph with space-time information is defined as a time sequence graph; a plurality of time sequence graphs are combined in chronological order to represent the dynamic changes of ship encounters in the scene, and a multi-ship encounter topological graph sequence is constructed, as shown in Figure 2 ; 2.1) Topological graph edge relationship construction Three encounter characteristic parameters, relative distance, DCPA, and TCPA, are used to judge the encounter relationship between ships: in tAt that moment, when i boat and j If all the encounter characteristic parameters between ships meet the threshold, then they are determined to have an encounter relationship at the current moment. t Relative distance between ships at any time The calculation is as follows:

[0018] , , , They are i boat and j The ship is t Location information at any given time. Based on relative distance. The two ships can be calculated at... t Moment and :

[0019]

[0020] yes i boat and j The ship in time t relative velocity, It is the cosine of the angle between the line connecting the two ships and their relative velocities, and it is calculated as follows:

[0021]

[0022] in , They are t The two ships' course at that moment.

[0023] 2.2) Node attribute definition MMSI and ship type are selected as node features of the topology graph, and the topology graph is represented by an undirected graph. Let X be the set of nodes in graph G, containing ship information and the number of ships currently in the graph; E is the set of edges, containing the encounter relationships between ships.

[0024] 2.3) Scene Graph Sequence Construction The Jaccard similarity coefficient is used to determine the encounter scenario to which two images belong: when the Jaccard similarity coefficients of two images are greater than a threshold, it indicates that they exist in the same encounter scenario. The calculation formula is as follows: (6) in, , These are topology diagrams , The similarity between two topological graphs is calculated by determining the similarity of node features (MMSI, ship type) contained in the two node sets. The threshold is typically set between 0.4 and 0.8, and the specific threshold setting needs to be selected through experimental comparison and analysis.

[0025] Clustering topological graphs belonging to the same scene to construct a graph sequence model representing the complete scene. , where Y represents the graph sequence features of C, including the duration of the scene and the start time; S represents the number of time sequence graphs contained in C.

[0026] 3) Identification of similar scenarios when multiple ships encounter each other; 3.1) Use graph convolutional neural networks for overall recognition of similar scenes, such as... Figure 3 ; By compressing the topological relationships of the scene graph sequence, a representative topological graph representing the entire scene is generated as input; The GCN model is used to classify the input representative topology graph. The category of the representative graph is determined based on the probability distribution of the output graph category, thereby measuring the similarity of the overall topological relationships in the scene. 3.1.1) GCN Model Construction For the graph convolutional network model used It has the following hierarchical propagation rules: (7) Let h represent the output features of the h-th layer, and ,and It is a picture Input features; ,in yes The adjacency matrix, It is the identity matrix; It is the degree matrix of a graph in an undirected graph. It equals the number of edges connected to the vertex; It is the trainable weight matrix of layer h; This represents the activation function.

[0027] 3.1.2) Input Dataset Construction ① Represents topology graph generation The process of generating a scene-representative topology diagram is as follows: Figure 4 As shown.

[0028] Depend on Figure 4It can be seen that by aggregating the points and edges of all the time series graphs in the scene, a representative topology graph containing all the graph structures in the scene can be obtained to summarize the changes in the entire scene.

[0029] ② Scene feature information extraction a. Graph adjacency matrix The extraction of the representative graph adjacency matrix is used to complete the extraction of the meeting relationship between ships in the scene. For the representative graph C Figure 4 Generated scene representative graph C R -1, its adjacency matrix As follows:

[0030] b. Graph node feature matrix For each node of the topology graph There are N features, denoted as . Taking MMSI and ship type as node feature input, for the scene representative graph in Figure 4 Its node feature matrix As follows:

[0031] 3.1.3) GCN model training To accurately control the update of model parameters and speed up the convergence speed, according to the size of the data set and the complexity of the model, 30 samples are processed for each batch (batch-size = 30), and 500 iterations (epoch = 500) are performed, while the dropout layer discard probability p is 0.5, and the 10-fold cross-validation (N-fold = 10) method is used for training. After training, the best learning rate of the model is determined to be 0.001.

[0032] During the training process, the accuracy and loss function curves of the model are plotted, as shown in Figure 5 , Figure 6 .

[0033] The results show that the model starts to show the best performance at iteration 350, with an accuracy of more than 90%, and the loss curve also shows a smooth trend with a small fluctuation range, indicating that the model has good stability and excellent performance.

[0034] 3.2) Similar scene hierarchical identification The meeting scene hierarchical identification algorithm based on graph structure is used to realize the identification of the internal state of the scene: The graph sequence model is split into three layers of graph structure, graph unit, topological graph and graph sequence, and the similarity between dynamic graph models is measured layer by layer by calculating the similarity of the three layers of graph structure, and the hierarchical identification of similar scenes is completed. The hierarchical identification process is shown in Figure 7 ; 2.2.1) Similar graph unit identification 1) Graph unit similarity The internal structure of the topological graph can be represented by factors such as node, edge attributes and connection mode, so the internal structure of the time sequence graph is represented by the combination of multiple graph units.

[0035] ① Graph unit The graph unit is a partial representation of the internal structure of the time sequence graph, and a graph unit is composed of a source node and multiple adjacent nodes , denoted by ; wherein the node attribute represents the ship type, wherein: 6: passenger ship, 7: general cargo ship, 8: oil tanker; for the time sequence graph , the extraction process of its graph unit is shown in Figure 8 .

[0036] As can be seen from Figure 8 , each node contained in is extracted as a source node in turn to obtain , , , four graph units, and since the order of adjacent nodes in the graph unit does not affect the result, it is not considered for the time being.

[0037] ② Graph unit similarity According to the above definition of the graph unit, the higher the similarity of the internal graph unit of the topological graph, the more similar the two graphs. The node attributes and the connection between nodes are selected as the elements of the graph unit to calculate the similarity of the graph unit. For two graph units and , the graph unit similarity is defined as follows:

[0038] wherein , represent the number of adjacent nodes of and respectively; is the number of adjacent nodes common to the two graph units; is the number of edges common to the two graph units, which is calculated as follows:

[0039] is the similarity of two graph units , ,

[0040] For example, for graph units Figure 8 and , : , ; ; ; their unit similarity .

[0041] 2) Graph unit recognition algorithm By calculating the similarity of the query graph unit with all graph units in the database and screening, the graph unit queue with the smallest distance is obtained, and then the adaptive evaluation score is used to eliminate graph units with a distance less than in the queue, so as to screen the graph units in the database that may have similarity with the query graph unit. The calculation formula of the evaluation score is as follows:

[0042] 2.2.2) Similarity of timing charts 1) Timing chart similarity Each timing chart is essentially an unordered set composed of graph units, so the calculation of graph similarity is regarded as solving the best assignment of two graph unit sets. The similarity between timing charts , is calculated as follows:

[0043] wherein is a weight matrix, is a Boolean matrix. The graph similarity calculation process is shown in Figure 9 .

[0044] The core of solving the best matching problem is to obtain the weight matrix , and the weight matrix obtained according to Figure 9 is as follows.

[0045] According to the weight matrix in Figure 10 , the graph similarity , of is obtained by using the Hungarian algorithm.

[0046] 2) Timing chart recognition algorithm ① Graph association list For the timing graph , where each graph unit can find its similar graph unit queue by the graph unit recognition algorithm, denoted as , so M similar graph unit queues can be obtained; according to the elements in the queue, the timing graph containing the similar graph unit can be found from the database, thereby screening out L timing graphs that may have a similar relationship with the query graph , and constructing a graph association list as the basis for similar timing graph recognition.

[0047] 2) Similar timing graph recognition algorithm Calculate the similarity between and each timing graph in the association list and screen, and obtain the similar graph queue that meets the conditions based on the evaluation score . The calculation formula of the evaluation score is as follows:

[0048] where is the qth graph unit of the query graph ; and is the graph unit of the current graph in the similar graph queue. If is the similar unit of , calculate the unit similarity and accumulate it. As can be seen from equation (15), the evaluation score indicates the number of similar graph units contained in the two graphs.

[0049] 2.2.3) Similar graph sequence recognition 1) Graph sequence similarity A graph sequence containing n graphs can be regarded as a time sequence composed of multiple time slices, so the similarity of the graph sequence is regarded as a time sequence similarity calculation problem, and therefore dynamic time warping (DTW) is introduced to define the similarity between the graph sequences. For two graph sequences , , the graph sequence similarity is calculated as follows:

[0050] 2) Graph sequence recognition algorithm A similar graph sequence recognition algorithm is proposed: for a graph sequence , first find n similar timing graph queues by the timing graph recognition algorithm, and construct a graph sequence association list according to the timing graphs in the queue; then traverse the association list , and based on the evaluation score Get the similar graph sequence queue. Evaluation score Calculated by the following formula.

[0051]

[0052] Wherein, is the nth time sequence graph of the query graph sequence Therefore Indicates the number of similar time sequence graphs contained in the two graph sequences.

[0053] The similar graph unit recognition of the encounter scenario can obtain the corresponding graph association list respectively, the similar time sequence graph of each graph in the two scenarios can be obtained by traversing the graph association list for similar time sequence graph recognition, and the graph sequence association list can be established, the similar graph sequence recognition is carried out by traversing the two graph sequence association lists, and the recognition result is obtained.

[0054] 3) There are multiple graph sequences in the recognition result queue obtained by the hierarchical recognition of the scene, the final result is obtained by arranging the graph sequence similarity from small to large, and the similar scene of the multi-ship encounter scenario is obtained.

[0055] The method provided by the application can recognize the encounter scenarios with similar encounter evolution trends, and can identify the encounter situation of the corresponding ship in the scene.

[0056] It should be understood that those skilled in the art can make improvements or changes according to the above description, and all these improvements and changes shall belong to the protection scope of the appended claims of the application.

Claims

1. A method for identifying similar multi-ship encounter scenes based on topological graph sequences, characterized in that, Includes the following steps: 1) Information collection for multi-ship encounter scenarios; A multi-ship encounter situation is defined as follows: when three or more ships approach each other, and at any given moment any ship encounters at least one other ship, a multi-ship encounter situation is constituted. The multi-ship encounter scenario is defined as: a combination of consecutive multi-ship encounter situations in chronological order; Acquire information on multi-ship encounter scenarios, including ship type, number of ships, ship MMSI, and ship navigation information; 2) Extraction of multi-ship encounter scenarios; Multi-ship encounter scenarios are extracted by constructing a sequence of multi-ship encounter topology maps. Ship information and encounter relationships are used as node features and edge attributes of the topology graph, and this type of topology graph with spatiotemporal information is defined as a time sequence graph. Multiple time sequence graphs are combined in chronological order to characterize the dynamic changes of ship encounters in the scene and construct a multi-ship encounter topology graph sequence. 3) Identification of similar scenarios when multiple ships encounter each other; 3.1) Use graph convolutional neural networks for overall identification of similar scenes; By compressing the topological relationships of the scene graph sequence, a representative topological graph representing the entire scene is generated as input; The GCN model is used to classify the input representative topology graph. The category of the representative graph is determined based on the probability distribution of the output graph category, thereby measuring the similarity of the overall topological relationships in the scene. 3.2) Layered identification of similar scenes A graph-based hierarchical encounter scene identification method is used to identify the internal states of the scene: The graph sequence model is split into three layers according to the graph structure: graph unit, topological graph, and graph sequence. The similarity between dynamic graph models is measured layer by layer by calculating the similarity of the three layers of graph structure, so as to complete the hierarchical identification of similar scenes. 4) The queue of identification results obtained from scene layer identification contains multiple graph sequences. The graph sequences are sorted in ascending order of similarity to obtain the final result, thus completing the acquisition of similar scenes in multi-ship encounter scenarios.

2. The method for identifying similar multi-ship encounter scenes based on topological graph sequences according to claim 1, characterized in that, In step 1), in the multi-ship encounter scenario, for the ships included in the earliest multi-ship encounter situation in the scenario, the time when the encounter occurs is recorded as the start time of the multi-ship encounter scenario; for the ships included in the encounter situation at the start time... ship s If at any given time at least one ship is not in contact with ship s If other ships in the scene encounter each other, this moment will be recorded as the end of the scene.

3. The method for identifying similar multi-ship encounter scenes based on topological graph sequences according to claim 1, characterized in that, In step 2), the multi-ship encounter topology sequence is constructed as follows: 2.1) Construction of edge relationships in the topological graph The encounter relationship between ships is determined using three encounter characteristic parameters: relative distance, DCPA, and TCPA. t At that moment, when i boat and j If all the encounter characteristic parameters between ships meet the threshold, then they are determined to have an encounter relationship at the current moment. 2.2) Node attribute definition MMSI and ship type are selected as node features of the topology graph, and the topology graph is represented by an undirected graph. Let X be the set of nodes in graph G, containing ship information and the number of ships currently in the graph; E is the set of edges, containing the encounter relationships between ships. 2.3) Scene graph sequence construction; The encounter scenario is determined by calculating the Jaccard coefficient: When the Jaccard coefficients of two graphs are greater than a threshold, it indicates that they exist in the same encounter scenario. Clustering topological graphs belonging to the same scene allows for the construction of a graph sequence model representing the complete scene. , where Y represents the graph sequence features of C, including the duration of the scene and the start time; S represents the number of time sequence graphs contained in C.

4. The method for identifying similar multi-ship encounter scenes based on topological graph sequences according to claim 1, characterized in that, In step 3.1), the process of generating the topology graph is as follows: By aggregating the union of points and edges of all time-series graphs in the scene, a representative topological graph containing all graph structures within the scene is obtained. Feature information extraction: The encounter relationships between ships in the scene are extracted by extracting the adjacency matrix that represents the topological graph. For each node representing the topology graph Each has N features, denoted as MMSI and ship type are used as node feature inputs.

5. The method for identifying similar multi-ship encounter scenes based on topological graph sequences according to claim 1, characterized in that, In step 3.2), the similarity graph unit is identified as follows: 3.2.1) Similarity graph unit identification; 3.2.1.1) Extract graph cells; A graph cell is a partial representation of the internal structure of a time series diagram; a graph cell... From the source node and multiple adjacent nodes Composition, using express; Node attributes Indicates the type of vessel; For timing diagrams , time sequence diagram Each node contained herein is used sequentially as a source node to extract graph units; 3.2.1.2) Calculate the similarity of graph elements; Node attributes and inter-node connections are selected as elements of graph units to calculate graph unit similarity. 3.2.1.3) Similarity graph unit identification; By query graph cells Similarity is calculated and filtered among all graph units in the database to obtain a queue of graph units with the minimum distance, and then an adaptive evaluation score is applied. Remove queue members whose distance is less than The graph cells are used to filter graph cells in the database that may be similar to the query graph cell.

6. The method for identifying similar multi-ship encounter scenes based on topological graph sequences according to claim 5, characterized in that, In step 3.2.1.2), for two graph units... and The similarity of the graph units is calculated as follows: in, , They represent and The number of adjacent nodes; It is the number of adjacent nodes shared by the two graph units; This is the number of edges shared by the two graph units, calculated as follows: in, It is the source node of the two graph units. , The similarity is calculated using the following formula: 。 7. The method for identifying similar multi-ship encounter scenes based on topological graph sequences according to claim 1, characterized in that, In step 3.2), the similar time series diagrams are identified as follows: 3.2.2) Identification of similar time series diagrams; 3.2.2.1) Time series diagram similarity; Each time series graph is an unordered set of graph units, and the similarity of time series graphs is calculated as the optimal allocation of two sets of graph units; 3.2.2.2) Identification of similar time series diagrams; For timing diagrams Each graph unit All of them are identified by graph unit identification to find their similar graph unit queues, denoted as Therefore Obtain a queue of M similar graph units; based on the elements in the queues, find the time series graphs containing the similar graph units from the database, thereby filtering out those that match the query graph. L time series diagrams that have similar relationships are used to form a graph association list. ; Will Similarity is calculated and filtered for each time series graph in the associated list, and then based on the evaluation score. Obtain a queue of similar graphs that meet the criteria.

8. The method for identifying similar multi-ship encounter scenes based on topological graph sequences according to claim 1, characterized in that, In step 3.2), the similarity graph unit is identified as follows: 3.2.3) Similar graph sequence identification; 3.2.3.1) Graph sequence similarity; A graph sequence containing n graphs A graph sequence can be viewed as a time series composed of multiple time slices. Therefore, measuring the similarity of graph sequences is considered as a problem of calculating time series similarity. Hence, dynamic time warping is introduced to define the similarity between graph sequences. For two graph sequences... , The similarity of their graph sequences is calculated as follows: 3.2.3.2) Similar graph sequence identification; For graph sequences First, a sequence graph identification algorithm is used to find n similar sequence graph queues. Then, a graph sequence association list is constructed based on the sequence graphs within each queue. Finally, [the algorithm is then used to...] Iterate through the associated list and based on the evaluation score Obtain a queue of similar graph sequences.

9. An electronic device, characterized in that, include: One or more processors; as well as Storage device for storing one or more programs. Wherein, when the one or more programs are executed by the one or more processors, the one or more processors perform the method according to any one of claims 1 to 8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method described in any one of claims 1 to 8.

Images