Ship track prediction method, device, equipment and medium based on knowledge drive

Through the construction and dynamic update of space-time knowledge base of multi-platform multi-modal data, the problem of low accuracy in ship trajectory prediction in traditional methods is solved, and more efficient track prediction and regular understanding is achieved.

CN118797279BActive Publication Date: 2024-11-26TSINGHUA UNIVERSITY
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411288428.0
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-09-14
Publication Date
2024-11-26
Estimated Expiration
2044-09-14

AI Technical Summary

Technical Problem

When traditional ship trajectory prediction methods lack real data and environmental instability, the prediction accuracy is low, making it difficult to master the ship's navigation rules, and the data demand is large and the training time is long.

Method used

Collect multimodal data through multiple platforms, perform spatiotemporal big data preprocessing and knowledge discovery, build a spatiotemporal knowledge base, and use knowledge-driven ship track prediction modules, combining long-term and short-term track coding, intention prediction and knowledge mapping, Dynamically update the knowledge base to realize ship trajectory prediction.

Benefits of technology

It improves the accuracy and stability of ship track prediction, reduces the difficulty of model learning, and enhances the understanding and prediction ability of navigation laws.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118797279B_ABST
    Figure CN118797279B_ABST
Patent Text Reader

Abstract

The present application provides a ship track prediction method, device, equipment and medium based on knowledge-driven, which relates to the field of ship track prediction technology, collects and integrates spatiotemporal big data of different sources and types, performs preprocessing operations, obtains the historical track information of the target ship and the surrounding ships, extracts the route knowledge of different categories of ships, integrates common sense information, expert experience, route laws and environmental information, models multivariate spatiotemporal relationships, builds a spatiotemporal knowledge base according to the knowledge tuple model and knowledge representation learning, builds a ship track prediction model and uses the model to obtain the ship track prediction result of the target ship according to the historical track information of the target ship and the spatiotemporal knowledge base, and dynamically learns and updates the data, route laws, and interactive correlation relationships of this ship track prediction. The above method is based on multi-source real navigation data and learns the knowledge in the navigation data, which improves the accuracy of track prediction.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the technical field of ship trajectory prediction, and in particular to a knowledge-driven ship track prediction method, device, equipment and medium. Background Art

[0002] Ships are key elements and actors in the field of maritime transportation. The ship navigation data collected by the maritime traffic network, including the location, direction, and speed of the ship's trajectory, is crucial for sensing the dynamic changes of maritime traffic and predicting the ship's navigation trajectory. Traditional track prediction methods mainly include methods based on physical models, methods based on learning frameworks, and methods based on hybrid models, but they have problems such as high prediction difficulty, large data requirements, and long training time.

[0003] At present, due to the widespread application of deep learning, recurrent neural networks have also been applied to solve time series prediction problems such as trajectory prediction due to their excellent performance in the field of natural language processing. Such as grid-based seq2seq (sequence to sequence) methods, seq2seq methods based on attention mechanisms, methods based on TREAD (traffic path extraction and anomaly detection), ship track prediction methods based on spatiotemporal feature fusion, etc. This type of method is based on historical navigation data, using the nonlinear architecture of recurrent neural networks to implicitly characterize the complex characteristics of ships during navigation, and through training to learn the navigation rules, generate trajectories that conform to their statistical characteristics for prediction.

[0004] However, the actual performance of such methods is often related to the quality of the data. As far as the problem of ship track prediction in adversarial environments is concerned, the main problem is the lack of relevant real data. Without the support of real data, the model will find it difficult to grasp the navigation laws of ships, and the performance in track prediction will decline. In addition, due to factors such as unstable communication links, strong concealment of target movement, and clock offset of receiving stations, the real track data has problems such as discontinuity, redundancy, and missing, resulting in low trajectory prediction accuracy of the model trained based on the above data.

[0005] Based on this, this application proposes a knowledge-driven ship track prediction method, which mines and extracts semantic representations and knowledge models from multimodal data collected from multiple platforms to further improve the accuracy of ship track prediction. Summary of the invention

[0006] The present application provides a knowledge-driven ship track prediction method, device, equipment and medium to solve the above-mentioned problems.

[0007] In a first aspect of an embodiment of the present application, a ship track prediction method based on knowledge-driven is provided, the method comprising:

[0008] Through the raw spatiotemporal big data preprocessing module, spatiotemporal big data of different sources and types are collected and integrated, and preprocessing operations are performed around the three levels of physical integrity, spatial coverage and time accuracy. The preprocessing operations include at least one of the following: filtering, segmentation, and splicing; the different sources include: satellites, ship automatic identification systems, monitoring, and mobile phone signals; the different types include: images, texts, and tables;

[0009] Through the data mining and knowledge discovery module, the historical trajectory information of the target ship and surrounding ships is obtained, and the route knowledge of different types of ships is extracted. Common sense information, expert experience, route rules and environmental information are integrated, and multivariate spatiotemporal relationships are quantitatively modeled through spatiotemporal relationship calculation. According to the knowledge tuple model and knowledge representation learning, a spatiotemporal knowledge base is constructed to realize spatiotemporal knowledge association;

[0010] A ship track prediction model is constructed through a knowledge-driven ship track prediction module, and a ship track prediction result of the target ship is obtained by using the ship track prediction model according to the historical track information of the target ship and the spatiotemporal knowledge base;

[0011] Through the dynamic update module of the spatiotemporal knowledge base, the data, route rules, and interactive correlation relationships of this ship track prediction are dynamically learned and updated, and the spatiotemporal knowledge base is continuously adjusted, while promoting the adaptive iteration of the knowledge-driven ship track prediction module.

[0012] The knowledge-driven ship track prediction module includes: a long-term track encoding module, a short-term track encoding module, an intention prediction module, a knowledge mapping module and a track prediction decoding module;

[0013] Long-term track encoding module is used to convert the current Before the moment The track information of the target ship in a long time range is encoded to obtain the spatiotemporal representation of the long-term historical track of the target ship;

[0014] Short-term track coding module is used to Before the moment The track information of the target ship and surrounding ships in a short time range is processed to obtain the spatiotemporal representation of the short-term historical track of the target ship and the surrounding environment information;

[0015] The intention prediction module is used to encode the long-term historical track of the target ship, the short-term historical track of the target ship and the surrounding environment information respectively, and then decode them based on the attention mechanism to obtain the current intention of the target ship;

[0016] A knowledge mapping module is used to call the pre-established spatiotemporal knowledge base to obtain target track knowledge associated with the target ship according to the spatiotemporal representation of the long-term historical track of the target ship, the spatiotemporal representation of the short-term historical track of the target ship, and the current intention of the target ship;

[0017] The track prediction decoding module is used to obtain the ship track prediction result of the target ship according to the long-term historical track of the target ship, the short-term historical track of the target ship and the target track knowledge.

[0018] Wherein, the intention prediction module includes a historical behavior intention recognition module, an interactive scene environment information encoding module and a current intention prediction module;

[0019] A historical behavior intention recognition module is used to encode the historical trajectory information of the target ship to obtain the historical behavior sequence of the target ship;

[0020] An interactive scene environment information encoding module is used to encode the surrounding environment information and the interactive track of the target ship and the surrounding ships to obtain a spatiotemporal representation of the surrounding environment information of the target ship;

[0021] The current intention prediction module is used to fuse and decode the historical behavior sequence of the target ship and the spatiotemporal representation of the surrounding environment information of the target ship based on the attention mechanism to obtain the current intention of the target ship.

[0022] The establishment process of the spatiotemporal knowledge base is as follows:

[0023] According to the ship ID, relevant data in the multi-source data is associated with the ship, and the multi-source data includes AIS data, radar data, remote sensing data, maritime early warning ship data, news, and video;

[0024] The multi-source data are integrated to obtain the position data, route patterns, and interactive relationships of the same ship at different times to obtain the ship's trajectory information.

[0025] The updating process of the spatiotemporal knowledge base is as follows:

[0026] After obtaining the real multi-source data of the target ship, obtaining the real track information of the target ship according to the real multi-source data of the target ship;

[0027] The actual trajectory information of the category to which the target ship belongs is statistically analyzed based on the overall duration, speed and heading, and the route knowledge of the category is updated.

[0028] In a second aspect of an embodiment of the present application, a ship track prediction device based on knowledge-driven is provided, the device comprising:

[0029] The raw spatiotemporal big data preprocessing module is used to collect and integrate spatiotemporal big data of different sources and types, and perform preprocessing operations around the three levels of physical integrity, spatial coverage and time accuracy. The preprocessing operations include at least one of the following: filtering, segmentation, and splicing; the different sources include: satellites, ship automatic identification systems, monitoring, and mobile phone signals; the different types include: images, texts, and tables;

[0030] The data mining and knowledge discovery module is used to obtain the historical trajectory information of the target ship and surrounding ships, extract the route knowledge of different types of ships, integrate common sense information, expert experience, route rules and environmental information, and quantitatively model the multivariate spatiotemporal relationship through spatiotemporal relationship calculation. According to the knowledge tuple model and knowledge representation learning, a spatiotemporal knowledge base is constructed to realize spatiotemporal knowledge association;

[0031] A knowledge-driven ship track prediction module is used to construct a ship track prediction model, and use the ship track prediction model to obtain a ship track prediction result of the target ship according to the historical track information of the target ship and the spatiotemporal knowledge base;

[0032] The spatiotemporal knowledge base dynamic update module is used to dynamically learn and update the data, route rules, and interactive correlation relationships of this ship track prediction, continuously adjust the spatiotemporal knowledge base, and promote the adaptive iteration of the knowledge-driven ship track prediction module.

[0033] In a third aspect of an embodiment of the present application, an electronic device is proposed, comprising a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the knowledge-driven ship track prediction method described in any one of the first aspects above.

[0034] In a fourth aspect of an embodiment of the present application, a computer-readable storage medium is proposed, on which a computer program / instruction is stored. When the computer program / instruction is executed by a processor, the knowledge-driven ship track prediction method described in any one of the first aspects above is implemented.

[0035] The present application has the following advantages: The present application provides a ship track prediction method, device, equipment and medium based on knowledge-driven, which collects and integrates spatiotemporal big data of different sources and types through the original spatiotemporal big data preprocessing module, and performs preprocessing operations around the three levels of physical integrity, spatial coverage and time accuracy. The preprocessing operations include at least one of the following: filtering, segmentation, and splicing; the different sources include: satellites, ship automatic identification systems, monitoring, and mobile phone signals; the different types include: images, texts, and tables; through data mining and knowledge discovery modules, the historical trajectory information of the target ship and the surrounding ships is obtained, and the route knowledge of different categories of ships is extracted, and the preprocessing operations are integrated. Common sense information, expert experience, route rules and environmental information are quantitatively modeled through spatiotemporal relationship calculation to model multivariate spatiotemporal relationships. According to the knowledge tuple model and knowledge representation learning, a spatiotemporal knowledge base is constructed to realize spatiotemporal knowledge association; a ship track prediction model is constructed through a knowledge-driven ship track prediction module, and the ship track prediction result of the target ship is obtained by using the ship track prediction model according to the historical track information of the target ship and the spatiotemporal knowledge base; through the spatiotemporal knowledge base dynamic update module, the data, route rules and interactive correlation of this ship track prediction are dynamically learned and updated, and the spatiotemporal knowledge base is continuously adjusted, while promoting the adaptive iteration of the knowledge-driven ship track prediction module. Multi-source data are associated and fused, and the correlation between the ship's route category and route knowledge is learned. The ship track prediction model is based on the historical track information and the target route knowledge to obtain accurate track prediction results. The above method is based on real navigation data and learns the rules in the navigation data, which improves the accuracy of track prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the description of the embodiments of the present application will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0037] Figure 1 It is a flow chart of the steps of a knowledge-driven ship track prediction method provided in an embodiment of the present application;

[0038] Figure 2 It is a flow chart of a knowledge-driven ship track prediction method provided in an embodiment of the present application;

[0039] Figure 3 It is a structural schematic diagram of a knowledge-driven ship track prediction module provided in an embodiment of the present application;

[0040] Figure 4 This is an architecture diagram of a knowledge-driven ship track prediction device provided in an embodiment of the present application;

[0041] Figure 5 It is a schematic diagram of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0042] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0043] Ships are key elements and behavioral subjects in the field of maritime transportation. The ship navigation data collected by the maritime traffic network, including information such as the position, direction, and speed of the ship's motion trajectory, is crucial for perceiving the dynamic changes of maritime traffic and predicting the ship's navigation trajectory. Traditional track prediction methods mainly include methods based on physical models, methods based on learning frameworks, and methods based on hybrid models. The physical model-based method mainly uses the law of motion and applies mathematical formulas to describe the movement characteristics of ships. This type of method includes curve models, lateral models, and ship models. In addition to the ship's speed and heading data, these methods also involve the size of the ship and other constant coefficients related to the environment. Due to the limitations of assumptions in actual scenarios, they are often not easy to obtain and difficult to calculate, so the physical model-based method is often used in theoretical analysis and simulation. The learning framework-based method uses the powerful fitting ability of neural networks. In theory, it can learn any function without prior information, such as ship parameters, weather, etc. However, the training of the network model requires massive data support and takes a long time. In addition, building a general network prediction model with excellent performance requires continuous attempts and verification. The hybrid model-based approach combines the physical model-based approach with the learning framework-based approach.

[0044] At present, due to the widespread application of deep learning, recurrent neural networks have also been applied to solve time series prediction problems such as trajectory prediction due to their excellent performance in the field of natural language processing. Such as grid-based seq2seq (sequence to sequence) methods, seq2seq methods based on attention mechanisms, methods based on TREAD (traffic path extraction and anomaly detection), ship track prediction methods based on spatiotemporal feature fusion, etc. This type of method is based on historical navigation data, using the nonlinear architecture of recurrent neural networks to implicitly characterize the complex characteristics of ships during navigation, and through training to learn the navigation rules, generate trajectories that conform to their statistical characteristics for prediction.

[0045] However, the actual performance of such methods is often related to the quality of the data. As far as the problem of ship track prediction in adversarial environments is concerned, the main problem is the lack of relevant real data. Without the support of real data, the model will find it difficult to grasp the navigation laws of ships, and the performance in track prediction will decline. In addition, due to factors such as unstable communication links, strong concealment of target movement, and clock offset of receiving stations, the real track data has problems such as discontinuity, redundancy, and missing, resulting in low trajectory prediction accuracy of the model trained based on the above data.

[0046] Based on this, this application collects multimodal data collected from multiple platforms for information association and fusion, and extracts a unified semantic representation and knowledge model. By encoding the long-term and short-term historical trajectory characteristics of the target ship and the surrounding ships respectively, the dynamic adjustment of the input length of the trajectory of the surrounding target ships is realized, and the fine features of the ship are reused through historical behavior intentions to reduce the difficulty of overall model learning, extract the semantic features driven by time series information, target features, and domain knowledge, and obtain the predicted track of the ship.

[0047] In a first aspect of the present application, a ship track prediction method based on knowledge drive is provided. Figure 1 , Figure 1 : is a flow chart of the steps of a knowledge-driven ship track prediction method provided in an embodiment of the present application, the method comprising:

[0048] Step 101: collect and integrate different sources and different types of spatiotemporal big data through the original spatiotemporal big data preprocessing module, and perform preprocessing operations around the three levels of physical integrity, spatial coverage and time accuracy. The preprocessing operations include at least one of the following: filtering, segmentation, and splicing; the different sources include: satellites, automatic identification systems for ships, monitoring, and mobile phone signals; the different types include: images, texts, and tables;

[0049] Step 102: Through the data mining and knowledge discovery module, the historical trajectory information of the target ship and the surrounding ships is obtained, the route knowledge of different types of ships is extracted, common sense information, expert experience, route rules and environmental information are integrated, and multivariate spatiotemporal relationships are quantitatively modeled through spatiotemporal relationship calculations. According to the knowledge tuple model and knowledge representation learning, a spatiotemporal knowledge base is constructed to realize spatiotemporal knowledge association;

[0050] Step 103: constructing a ship track prediction model through a knowledge-driven ship track prediction module, and using the ship track prediction model to obtain a ship track prediction result of the target ship according to the historical track information of the target ship and the spatiotemporal knowledge base;

[0051] Step 104: Through the spatiotemporal knowledge base dynamic update module, the data, route rules, and interactive correlation relationships of this ship track prediction are dynamically learned and updated, the spatiotemporal knowledge base is continuously adjusted, and the adaptive iteration of the knowledge-driven ship track prediction module is promoted.

[0052] In order to clearly describe a knowledge-driven ship track prediction method provided in the embodiment of the present application, Figure 2 The method is described in detail. Figure 2 It is a flow chart of a knowledge-driven ship track prediction method provided in an embodiment of the present application. In addition, in the embodiment of the present application, track and track have the same meaning, both of which are the route taken by the ship.

[0053] When implementing step 101, since the ship's trajectory is affected by different spatial, physical and temporal factors, the original spatiotemporal big data preprocessing module is used to collect and integrate spatiotemporal big data from different sources and different types. Spatiotemporal big data refers to data around the three levels of physical integrity, spatial coverage and time accuracy. The acquired spatiotemporal big data is preprocessed, and the above preprocessing operations include at least one of the following: filtering, segmentation, and splicing; the above different sources include: satellites, automatic ship identification systems, monitoring, and mobile phone signals; the above different types include: images, texts, and tables. After the above preprocessing operations, effective spatiotemporal big data is obtained. Furthermore, based on the above preprocessed spatiotemporal big data, the ship's trajectory can be accurately predicted, and the above spatiotemporal big data is preprocessed to obtain the historical trajectory information of each ship.

[0054] When implementing step 102, the lack of relevant real data will make it difficult for the trajectory prediction model to grasp the law of ship navigation. Therefore, it is very important to obtain the real historical trajectory information of the target ship and the surrounding ships. The navigation information related to the dissemination is obtained through different platforms, and the historical trajectory information of the target ship and the surrounding ships is obtained. The historical trajectory information includes mobile communication service identification code, longitude and latitude, speed, heading, timestamp, ship type and other information. Then, the route knowledge of different types of ships is extracted, and common sense information, expert experience, route laws and environmental information are integrated. The multivariate spatiotemporal relationship is quantitatively modeled through spatiotemporal relationship calculation. According to the knowledge tuple model and knowledge representation learning, a spatiotemporal knowledge base is constructed to realize spatiotemporal knowledge association.

[0055] The computational quantitative modeling of multivariate spatiotemporal relationships involves the analysis and modeling of the interdependence of multiple variables in space and time, including at least: data collection, data visualization and exploratory analysis, multivariate time series data analysis, spatiotemporal correlation considerations, modeling method selection, model training and verification, and model optimization to achieve the dynamic behavior of multiple variables in space and time. In this application, the above variables can be common sense information, expert experience, route rules, environmental information, and preprocessed spatiotemporal big data, etc. The knowledge tuple model is a structured method for representing and storing knowledge. In natural language, knowledge can usually be expressed in the form of subject, predicate, and object. This expression corresponds to the knowledge tuple in the knowledge base. The knowledge tuple model can provide a structured and standardized way to represent and store knowledge, so that knowledge can be processed by computers more efficiently, and a complex knowledge network can be constructed to achieve the learning of complex knowledge. Through knowledge representation learning, the relationship in the knowledge tuple can be fully learned and embedded in a low-dimensional space. By learning the relationship between different variables in space and time, spatiotemporal knowledge association is achieved.

[0056] Exemplarily, a process for extracting navigation knowledge and navigation categories from historical trajectory data provided by an embodiment of the present application includes: for historical trajectory data, different waypoint areas are delineated, and the same trajectories are grouped into one category according to the order in which the ship sails through the waypoints. A characterization trajectory is extracted for the trajectories in each category, and the characterization trajectory is the "middle" trajectory of all trajectories in this category. In addition to the characterization trajectory, the speed and heading of different sections can also be analyzed to describe the trajectory of this category in more detail and extract effective ship navigation knowledge.

[0057] Building a spatiotemporal knowledge base means creating a database that systematically stores and manages information about entities in time and space. This type of knowledge base can support complex queries and analyses, such as predicting the future trajectory of pedestrians, analyzing the spatial distribution of historical events, or understanding the characteristics of a location over time. Building a spatiotemporal knowledge base according to the knowledge tuple model and knowledge representation learning usually includes: collecting data sets containing spatiotemporal information, such as ship trajectory data, historical event records, etc., and cleaning, standardizing and normalizing them for subsequent analysis. Extracting spatiotemporal relationships from raw data and building a knowledge graph, which may involve natural language processing and computer vision technology to identify and parse entities, relationships and their spatiotemporal attributes. Using graph databases or relational databases to store knowledge tuples, designing efficient indexing strategies to speed up queries. Using the results of knowledge representation learning to perform reasoning and prediction, such as predicting the future behavior trajectory of an entity, or analyzing the development trend of events in a specific area. Through the above steps, building a spatiotemporal knowledge base can not only store rich spatiotemporal information, but also support complex queries, pattern recognition and future predictions, providing decision support for ship track prediction.

[0058] In an optional embodiment of the present application, the establishment process of the above-mentioned spatiotemporal knowledge base is as follows: according to the ship ID, the relevant data in the multi-source data is associated with the ship, and the above-mentioned multi-source data includes AIS data, radar data, remote sensing data, maritime warning ship data, news, and video; the position data, route rules, and interactive correlation relationships of the above-mentioned multi-source data for the same ship at different times are integrated to obtain the ship's track information. The ship's track information includes: mobile communication service identification code, longitude and latitude, speed, heading, timestamp, ship type and other information; with the overall duration, speed, and heading as dimensions, the various track information belonging to the same route category is counted to obtain the knowledge of the route category. The knowledge of the route category includes: common sense information accumulated in the track prediction task, expert experience, navigation rules, environmental information, historical behavior intentions, etc. The above-mentioned ship's track information and route category knowledge are used to construct a spatiotemporal knowledge base.

[0059] In an optional embodiment of the present application, the updating process of the spatiotemporal knowledge base is as follows: after obtaining the real multi-source data of the target ship, the real track information of the target ship is obtained according to the real multi-source data of the target ship; the real track information of the category to which the target ship belongs is counted with the overall duration, speed and heading as dimensions, and the route knowledge of the category is updated. Based on the updated route knowledge, the spatiotemporal knowledge base is updated.

[0060] When step 103 is specifically implemented, a ship track prediction model is constructed through a knowledge-driven ship track prediction module. According to the historical track information of the target ship and the spatiotemporal knowledge base, combined with the constructed ship track prediction model, the ship track prediction result of the target ship is obtained.

[0061] In an optional embodiment of the present application, the structure of a knowledge-driven ship track prediction module proposed in the present application can be found in Figure 3 , Figure 3 1 is a schematic diagram of the structure of a knowledge-driven ship track prediction module provided in an embodiment of the present application. The above-mentioned knowledge-driven ship track prediction module includes: a long-term track encoding module, a short-term track encoding module, an intention prediction module, a knowledge mapping module and a track prediction decoding module; wherein the long-term track encoding module is used to convert the current Before the moment The target ship’s track information in the long-term range is encoded to obtain the spatiotemporal representation of the target ship’s long-term historical track; the short-term track encoding module is used to encode the current Before the moment The track information of the target ship and the surrounding ships within a short time range is processed to obtain the spatiotemporal representation of the short-term historical track of the target ship and the surrounding environment information; the intention prediction module is used to encode the long-term historical track of the target ship, the short-term historical track of the target ship and the surrounding environment information respectively, and then decoded through fusion based on the attention mechanism to obtain the current intention of the target ship; the knowledge mapping module is used to call the pre-established spatiotemporal knowledge base according to the spatiotemporal representation of the long-term historical track of the target ship, the spatiotemporal representation of the short-term historical track of the target ship, and the current intention of the target ship to obtain the target track knowledge associated with the target ship; the track prediction decoding module is used to obtain the ship track prediction result of the target ship according to the long-term historical track of the target ship, the short-term historical track of the target ship and the target track knowledge.

[0062] In an optional embodiment of the present application, the above-mentioned intention prediction module includes a historical behavior intention recognition module, an interactive scene environment information encoding module and a current intention prediction module; in order to enhance the track characteristics of the target ship so that the behavior and trajectory information of the target ship can be more easily learned during decoding, the historical trajectory information of the target ship is encoded through the historical behavior intention recognition module to obtain the historical behavior sequence of the target ship; the surrounding environment information and the interactive track of the target ship and the surrounding ships are encoded through the interactive scene environment information encoding module to obtain the spatiotemporal representation of the surrounding environment information of the target ship; the historical behavior sequence of the target ship and the spatiotemporal representation of the surrounding environment information of the target ship are fused and decoded based on the attention mechanism through the current intention prediction module to obtain the current intention of the target ship and realize the inference of the historical behavior of the target ship.

[0063] When the step 104 is specifically implemented, in order to dynamically predict the ship's trajectory, in the process of using the above-mentioned knowledge-driven ship track prediction module, the data, route rules, and interactive correlation relationships of this ship track prediction are used for dynamic learning and updating, and the spatiotemporal knowledge base is continuously adjusted, while promoting the adaptive iteration of the knowledge-driven ship track prediction module to achieve accurate dynamic prediction of the ship's trajectory.

[0064] In another optional embodiment of the present application, the above-mentioned ship track prediction model includes a ship track processing network and a ship track prediction network. The ship track processing network is used to pre-process the acquired historical track information, and perform data mining and knowledge discovery. The knowledge-driven track prediction module at least includes: a long-term track encoding module, a short-term track encoding module, an intention prediction module, a knowledge mapping module, and a track prediction decoding module. The modules in the ship track processing network and the modules included in the above-mentioned knowledge-driven ship track prediction module have the same function, which will not be repeated here. The above-mentioned ship track prediction network is used to predict the track of knowledge. First, the ship track processing network is used to determine the route category of the target ship according to the historical track information of the target ship, and further determine the target route knowledge associated with the route category of the target ship. The knowledge here refers to the common sense information, expert experience, environmental information, navigation rules, historical behavior intentions, abnormal event warnings, etc. involved in the field of track prediction tasks, and its expression forms include text, charts, files, images, videos, etc. Different types of data and the knowledge they cover are multi-source and heterogeneous in form but have consistent internal semantics. They can enrich the features mined from single trajectory data and help to accurately predict the trajectory.

[0065] In another optional embodiment of the present application, the ship track processing network also includes a route category determination module, and the route category is determined by the route category determination module. Specifically, the historical navigation data of the target ship is first mined to extract the route knowledge of the target ship; then, based on the route knowledge of the target ship, the data points of the target ship's berthing are extracted according to the navigation speed; based on the density, the data point distribution is clustered into different waypoints, and the route tracks passing through similar waypoints are classified into a route category according to the time sequence. This type of track has a similar route and has similar speeds and headings at positions adjacent to each other. That is, the tracks in each type of route have statistical laws in terms of overall duration, speed, and heading. Therefore, in the process of training the ship track prediction model, the most representative type of track category in the water area can be integrated, and the relationship between variables can be found from a large amount of data. Then, the ship track prediction model can be associated with the navigation characteristics of this type of track according to the track category. When predicting the ship track, the route can be classified and then predicted according to the route category, thereby ensuring the accuracy of the track prediction.

[0066] In another optional embodiment of the present application, the ship track prediction network is trained using the historical track information of multiple ships and the route categories of multiple ships as training samples, and the association between the route categories of multiple ships and the route knowledge is learned to obtain a trained ship track prediction network. The trained ship track prediction network can obtain the track prediction result of the target ship based on the historical track information of the target ship and the target route knowledge.

[0067] In another optional embodiment of the present application, the above-mentioned ship track prediction network is a generative network obtained by training a generative adversarial network, wherein the generative network can adopt a recurrent neural network structure, or a transformer structure based on a self-attention mechanism. In the embodiment of the present application, the training of the generative network with a transformer structure is exemplified. The training process of a ship track prediction generative network based on a generative adversarial network provided in the embodiment of the present application is as follows: randomly generate a starting position according to the real spatial distribution of the ship; input the real trajectory data of the ship and the randomly generated starting position into the generative network to be trained for Pre-training is performed, and the pre-trained generating network can be used to generate a string of trajectory data; random positions and random noises are input into the pre-trained generating network to generate simulated trajectory data, and the simulated trajectory data is a string of trajectory data; the real trajectory data and the simulated trajectory data are input into the discriminant network to be trained to obtain the discrimination result, and the network parameters of the discriminant network are updated based on the loss function of the discriminant network to improve the discrimination accuracy of the discriminant network, and the network parameters of the generating network are updated based on the loss function of the generating network to improve the simulation of the data generated by the generating network; when the discrimination result of the discriminant network for the simulated data is true, the training is terminated, and the trained generating network is used as the ship track prediction network.

[0068] The embodiment of the present application provides a ship track prediction method based on knowledge-driven, which collects and integrates spatiotemporal big data of different sources and types through the original spatiotemporal big data preprocessing module, and performs preprocessing operations around the three levels of physical integrity, spatial coverage and time accuracy. The preprocessing operations include at least one of the following: filtering, segmentation, and splicing; the different sources include: satellites, ship automatic identification systems, monitoring, and mobile phone signals; the different types include: images, texts, and tables; through the data mining and knowledge discovery module, the historical trajectory information of the target ship and the surrounding ships is obtained, and the route knowledge of different categories of ships is extracted, and common sense information, expert experience, and other information are integrated to predict the ship's track. Experience, route rules and environmental information, quantitatively model multivariate spatiotemporal relationships through spatiotemporal relationship calculation, build a spatiotemporal knowledge base according to the knowledge tuple model and knowledge representation learning, and realize spatiotemporal knowledge association; through the knowledge-driven ship track prediction module, build a ship track prediction model, use the ship track prediction model, according to the historical track information of the target ship and the spatiotemporal knowledge base, obtain the ship track prediction result of the target ship; through the spatiotemporal knowledge base dynamic update module, dynamically learn and update the data, route rules, and interactive correlation relationship of this ship track prediction, continuously adjust the spatiotemporal knowledge base, and promote the adaptive iteration of the knowledge-driven ship track prediction module. Information association and fusion of multi-source data, and learning the correlation between the ship's route category and route knowledge, the ship track prediction model is based on the historical track information and the target route knowledge to obtain accurate track prediction results. The above method is based on real navigation data and learns the rules in navigation data, which improves the accuracy of track prediction.

[0069] In a second aspect of the embodiment of the present application, a ship track prediction device based on knowledge drive is provided, referring to Figure 4 , Figure 4 : is an architecture diagram of a knowledge-driven ship track prediction device provided in an embodiment of the present application, the device comprising:

[0070] The original spatiotemporal big data preprocessing module 401 is used to collect and integrate spatiotemporal big data of different sources and types, and perform preprocessing operations around the three levels of physical integrity, spatial coverage and time accuracy. The preprocessing operations include at least one of the following: filtering, segmentation, and splicing; the different sources include: satellites, automatic identification systems for ships, monitoring, and mobile phone signals; the different types include: images, texts, and tables;

[0071] The data mining and knowledge discovery module 402 is used to obtain the historical trajectory information of the target ship and the surrounding ships, extract the route knowledge of different types of ships, integrate common sense information, expert experience, route rules and environmental information, quantitatively model the multivariate spatiotemporal relationship through spatiotemporal relationship calculation, build a spatiotemporal knowledge base according to the knowledge tuple model and knowledge representation learning, and realize spatiotemporal knowledge association;

[0072] The knowledge-driven ship track prediction module 403 is used to construct a ship track prediction model, and use the ship track prediction model to obtain a ship track prediction result of the target ship according to the historical track information of the target ship and the spatiotemporal knowledge base;

[0073] The spatiotemporal knowledge base dynamic update module 404 is used to dynamically learn and update the data, route rules, and interactive correlation relationships of this ship track prediction, continuously adjust the spatiotemporal knowledge base, and promote the adaptive iteration of the knowledge-driven ship track prediction module.

[0074] The knowledge-driven ship track prediction module includes: a long-term track encoding module, a short-term track encoding module, an intention prediction module, a knowledge mapping module and a track prediction decoding module;

[0075] Long-term track encoding module is used to convert the current Before the moment The track information of the target ship in a long time range is encoded to obtain the spatiotemporal representation of the long-term historical track of the target ship;

[0076] Short-term track coding module is used to Before the moment The track information of the target ship and surrounding ships in a short time range is processed to obtain the spatiotemporal representation of the short-term historical track of the target ship and the surrounding environment information;

[0077] The intention prediction module is used to encode the long-term historical track of the target ship, the short-term historical track of the target ship and the surrounding environment information respectively, and then decode them based on the attention mechanism to obtain the current intention of the target ship;

[0078] A knowledge mapping module is used to call the pre-established spatiotemporal knowledge base to obtain target track knowledge associated with the target ship according to the spatiotemporal representation of the long-term historical track of the target ship, the spatiotemporal representation of the short-term historical track of the target ship, and the current intention of the target ship;

[0079] The track prediction decoding module is used to obtain the ship track prediction result of the target ship according to the long-term historical track of the target ship, the short-term historical track of the target ship and the target track knowledge.

[0080] Wherein, the intention prediction module includes a historical behavior intention recognition module, an interactive scene environment information encoding module and a current intention prediction module;

[0081] A historical behavior intention recognition module is used to encode the historical trajectory information of the target ship to obtain the historical behavior sequence of the target ship;

[0082] An interactive scene environment information encoding module is used to encode the surrounding environment information and the interactive track of the target ship and the surrounding ships to obtain a spatiotemporal representation of the surrounding environment information of the target ship;

[0083] The current intention prediction module is used to fuse and decode the historical behavior sequence of the target ship and the spatiotemporal representation of the surrounding environment information of the target ship based on the attention mechanism to obtain the current intention of the target ship.

[0084] The data mining and knowledge discovery module includes a spatiotemporal knowledge base establishment module, and the spatiotemporal knowledge base establishment module includes:

[0085] An association submodule is used to associate relevant data in multi-source data with the ship according to the ship ID, wherein the multi-source data includes AIS data, radar data, remote sensing data, maritime early warning ship data, news, and video;

[0086] The integration submodule is used to integrate the multi-source data for the location data, route rules, and interactive correlation of the same ship at different times to obtain the ship's trajectory information.

[0087] Wherein, the data mining and knowledge discovery module includes a spatiotemporal knowledge base update module, and the spatiotemporal knowledge base update module includes:

[0088] A real track acquisition submodule is used to obtain the real multi-source data of the target ship and then obtain the real track information of the target ship according to the real multi-source data of the target ship;

[0089] The statistical update submodule is used to collect statistics on the real trajectory information of the category to which the target ship belongs based on the overall duration, speed and heading, and update the route knowledge of the category.

[0090] Based on the same application concept, the embodiment of the present application discloses an electronic device in a third aspect. Figure 5 A schematic diagram of an electronic device disclosed in an embodiment of the present application is shown. Figure 5As shown, the electronic device 100 includes: a memory 110 and a processor 120, the memory of the electronic device is not less than 12G, the main frequency of the processor is not less than 2.4GHz, the memory 110 and the processor 120 are connected via a bus communication, and a computer program is stored in the memory 110. The computer program can be run on the processor 120 to implement a knowledge-driven ship track prediction method disclosed in an embodiment of the present application.

[0091] Based on the same application concept, the fourth aspect of the embodiment of the present application discloses a computer-readable storage medium, on which a computer program / instruction is stored. When the computer program / instruction is executed by a processor, a knowledge-driven ship track prediction method disclosed in the embodiment of the present application is implemented.

[0092] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.

[0093] The embodiments of the present application are described with reference to the flowcharts and / or block diagrams of the methods, devices, electronic devices, and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing terminal device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing terminal device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0094] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing terminal device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce a manufactured product including an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0095] These computer program instructions can also be loaded onto a computer or other programmable data processing terminal device so that a series of operating steps are executed on the computer or other programmable terminal device to produce a computer-implemented process, thereby providing instructions for executing on the computer or other programmable terminal device to implement the process. Figure 1 A process or multiple processes and / or boxes Figure 1The steps for the functions specified in one or more boxes.

[0096] Although the preferred embodiments of the present application have been described, those skilled in the art may make additional changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the embodiments of the present application.

[0097] Finally, it should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or terminal device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or terminal device. In the absence of further restrictions, the elements defined by the sentence "including one..." do not exclude the existence of other identical elements in the process, method, article or terminal device including the elements.

[0098] The above is a detailed introduction to the knowledge-driven ship track prediction method, device, equipment and medium provided by the present application. This article uses specific examples to illustrate the principles and implementation methods of the present application. The description of the above embodiments is only used to help understand the method of the present application and its core idea; at the same time, for general technical personnel in this field, according to the ideas of the present application, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as a limitation on the present application.

Claims

1. A knowledge-driven ship track prediction method. It is characterized in that The method comprises: Through the raw spatiotemporal big data preprocessing module, spatiotemporal big data of different sources and types are collected and integrated, and preprocessing operations are performed around the three levels of physical integrity, spatial coverage and time accuracy. The preprocessing operations include at least one of the following: filtering, segmentation, and splicing; the different sources include: satellites, ship automatic identification systems, monitoring, and mobile phone signals; the different types include: images, texts, and tables; Through the data mining and knowledge discovery module, the historical trajectory information of the target ship and surrounding ships is obtained, and the route knowledge of different types of ships is extracted. Common sense information, expert experience, route rules and environmental information are integrated, and multivariate spatiotemporal relationships are quantitatively modeled through spatiotemporal relationship calculation. According to the knowledge tuple model and knowledge representation learning, a spatiotemporal knowledge base is constructed to realize spatiotemporal knowledge association; A ship track prediction model is constructed through a knowledge-driven ship track prediction module, and a ship track prediction result of the target ship is obtained by using the ship track prediction model according to the historical track information of the target ship and the spatiotemporal knowledge base; Through the spatiotemporal knowledge base dynamic update module, the data, route rules, and interactive correlation relationships of the ship track prediction are dynamically learned and updated, the spatiotemporal knowledge base is continuously adjusted, and the adaptive iteration of the knowledge-driven ship track prediction module is promoted; The knowledge-driven ship track prediction module includes: a long-term track encoding module, a short-term track encoding module, an intention prediction module, a knowledge mapping module and a track prediction decoding module; Long-term track encoding module is used to convert the current Before the moment The track information of the target ship in a long time range is encoded to obtain the spatiotemporal representation of the long-term historical track of the target ship; Short-term track coding module is used to Before the moment The track information of the target ship and surrounding ships in a short time range is processed to obtain the spatiotemporal representation of the short-term historical track of the target ship and the surrounding environment information; The intention prediction module is used to encode the long-term historical track of the target ship, the short-term historical track of the target ship and the surrounding environment information respectively, and then decode them based on the attention mechanism to obtain the current intention of the target ship; A knowledge mapping module is used to call the pre-established spatiotemporal knowledge base to obtain target track knowledge associated with the target ship according to the spatiotemporal representation of the long-term historical track of the target ship, the spatiotemporal representation of the short-term historical track of the target ship, and the current intention of the target ship; The track prediction decoding module is used to obtain the ship track prediction result of the target ship according to the long-term historical track of the target ship, the short-term historical track of the target ship and the target track knowledge.

2. According to the knowledge-driven ship track prediction method of claim 1, It is characterized in that The intention prediction module includes a historical behavior intention recognition module, an interaction scene environment information encoding module and a current intention prediction module; A historical behavior intention recognition module is used to encode the historical trajectory information of the target ship to obtain the historical behavior sequence of the target ship; An interactive scene environment information encoding module is used to encode the surrounding environment information and the interactive track of the target ship and the surrounding ships to obtain a spatiotemporal representation of the surrounding environment information of the target ship; The current intention prediction module is used to fuse and decode the historical behavior sequence of the target ship and the spatiotemporal representation of the surrounding environment information of the target ship based on the attention mechanism to obtain the current intention of the target ship.

3. According to the knowledge-driven ship track prediction method of claim 1, It is characterized in that The process of establishing the spatiotemporal knowledge base is as follows: According to the ship ID, relevant data in the multi-source data is associated with the ship, and the multi-source data includes AIS data, radar data, remote sensing data, maritime early warning ship data, news, and video; The multi-source data are integrated to obtain the position data, route rules, and interactive correlations of the same ship at different times to obtain the ship's trajectory information.

4. The ship track prediction method based on knowledge drive according to claim 3, It is characterized in that The updating process of the spatiotemporal knowledge base is as follows: After obtaining the real multi-source data of the target ship, obtaining the real track information of the target ship according to the real multi-source data of the target ship; The actual trajectory information of the category to which the target ship belongs is statistically analyzed based on the overall duration, speed and heading, and the route knowledge of the category is updated.

5. A ship track prediction device based on knowledge drive, It is characterized in that The device comprises: The raw spatiotemporal big data preprocessing module is used to collect and integrate spatiotemporal big data of different sources and types, and perform preprocessing operations around the three levels of physical integrity, spatial coverage and time accuracy. The preprocessing operations include at least one of the following: filtering, segmentation, and splicing; the different sources include: satellites, ship automatic identification systems, monitoring, and mobile phone signals; the different types include: images, texts, and tables; The data mining and knowledge discovery module is used to obtain the historical trajectory information of the target ship and surrounding ships, extract the route knowledge of different types of ships, integrate common sense information, expert experience, route rules and environmental information, and quantitatively model the multivariate spatiotemporal relationship through spatiotemporal relationship calculation. According to the knowledge tuple model and knowledge representation learning, a spatiotemporal knowledge base is constructed to realize spatiotemporal knowledge association; A knowledge-driven ship track prediction module is used to construct a ship track prediction model, and use the ship track prediction model to obtain a ship track prediction result of the target ship according to the historical track information of the target ship and the spatiotemporal knowledge base; A spatiotemporal knowledge base dynamic update module is used to dynamically learn and update the data, route rules, and interactive correlation relationships of the ship track prediction, continuously adjust the spatiotemporal knowledge base, and promote the adaptive iteration of the knowledge-driven ship track prediction module; The knowledge-driven ship track prediction module includes: a long-term track encoding module, a short-term track encoding module, an intention prediction module, a knowledge mapping module and a track prediction decoding module; Long-term track encoding module is used to convert the current Before the moment The track information of the target ship in a long time range is encoded to obtain the spatiotemporal representation of the long-term historical track of the target ship; Short-term track coding module is used to Before the moment The track information of the target ship and surrounding ships in a short time range is processed to obtain the spatiotemporal representation of the short-term historical track of the target ship and the surrounding environment information; The intention prediction module is used to encode the long-term historical track of the target ship, the short-term historical track of the target ship and the surrounding environment information respectively, and then decode them based on the attention mechanism to obtain the current intention of the target ship; A knowledge mapping module is used to call the pre-established spatiotemporal knowledge base to obtain target track knowledge associated with the target ship according to the spatiotemporal representation of the long-term historical track of the target ship, the spatiotemporal representation of the short-term historical track of the target ship, and the current intention of the target ship; The track prediction decoding module is used to obtain the ship track prediction result of the target ship according to the long-term historical track of the target ship, the short-term historical track of the target ship and the target track knowledge.

6. An electronic device comprising a memory, a processor and a computer program stored in the memory, It is characterized in that The processor executes the computer program to implement the knowledge-driven ship track prediction method according to any one of claims 1 to 4.

7. A computer-readable storage medium, It is characterized in that A computer program / instruction is stored thereon, and when the computer program / instruction is executed by a processor, the knowledge-driven ship track prediction method described in any one of claims 1 to 4 is implemented.