Construction method of ship name labeling model and outlier detection method based on ship name

Abstract

The present invention provides a method for constructing a ship name labeling model and a method for detecting outliers based on ship names, and relates to the technical field of navigation and geographic information systems. The method for constructing a ship name labeling model includes: obtaining multiple ship names; preprocessing multiple ship names, screening out ship names with specified features as first ship names, the specified features including the ship type code and the ship city code; constructing a ship name entity set based on the specified features, the ship name entity set including entity data corresponding to the specified features; labeling the first ship name according to a preset labeling method based on the entity data, generating a ship name data set; inputting the ship name data set into a first neural network structure for training, and obtaining a ship name labeling model. The ship type is obtained through the ship name labeling model, and then the ship's activity range is determined; the ship's trajectory information is obtained; and the trajectory information that exceeds the ship's activity range is marked as an outlier.

Description

Technical Field

[0001] The present invention relates to the technical field of navigation and geographic information systems, and in particular to a method for constructing a ship name annotation model and a method for detecting outliers based on ship names. Background Art

[0002] Existing ship outlier detection technologies have been able to filter out most of the outlier trajectory points by cleaning garbled data, processing based on heading angles, setting unreachable distance and speed thresholds, etc. However, these are all based on the normal update cycle of ship trajectory points. When a large number of trajectories are missing, the speed and distance of several adjacent points can no longer be used as a basis for distinguishing normal trajectory points from outlier trajectory points, and the relevant technology is still difficult to handle.

[0003] Therefore, it is particularly important to develop a method that can improve the ship name labeling and outlier detection based on ship names when a large number of ship trajectories are missing. Summary of the invention

[0004] In order to solve the above technical problems, the present application proposes a method for constructing a ship name annotation model and an outlier detection method based on ship names.

[0005] A first aspect of the present invention provides a method for constructing a ship name labeling model.

[0006] The second aspect of the present invention proposes an outlier detection method based on ship names.

[0007] A third aspect of the present invention proposes an outlier detection system based on ship names.

[0008] A fourth aspect of the present invention provides a system for constructing a ship name labeling model.

[0009] The fifth aspect of the present invention proposes another outlier detection system based on ship names.

[0010] A sixth aspect of the present invention provides a readable storage medium.

[0011] In view of this, a first aspect of the present invention proposes a method for constructing a ship name labeling model, comprising: obtaining multiple ship names; preprocessing the multiple ship names, screening out ship names with specified features as first ship names, the specified features including the ship type code and the ship city code; constructing a ship name entity set based on the specified features, the ship name entity set including entity data corresponding to the specified features; labeling the first ship name according to a preset labeling method based on the entity data to generate a ship name data set; inputting the ship name data set into a first neural network structure for training to obtain a ship name labeling model.

[0012] The ship name data set includes a training data set, a verification data set and a test data set; the first neural network structure includes: a BiLSTM structure (Bidlrectional Long Short-Term, bidirectional long short-term memory network structure) and a CRF structure (Conditional Random Fidlds, conditional random field, a discriminant probabilistic undirected graph structure); the ship name data set is input into the first neural network structure for training, and the steps of obtaining the ship name labeling model include: inputting the training data set into the BiLSTM structure for training to generate a first ship name labeling model; inputting the verification data set and the first ship name labeling model into the CRF structure to generate evaluation parameters; optimizing the first ship name labeling model based on the evaluation parameters to generate a second ship name labeling model; inputting the test data set into the second ship name labeling model for testing to generate test results; and determining the second ship name labeling model whose test results meet the preset requirements as the ship name labeling model.

[0013] In this technical solution, a method for constructing a ship name labeling model includes: obtaining multiple ship names; preprocessing the multiple ship names, screening out a ship name with specified features as a first ship name, the specified features including a ship type code and a ship city code; constructing a ship name entity set based on the specified features, the ship name entity set including entity data corresponding to the specified features; labeling the first ship name according to a preset labeling method based on the entity data to generate a ship name data set; inputting the ship name data set into a first neural network structure for training to obtain a ship name labeling model. By constructing a ship name annotation model, the name of the ship can be annotated, and then the actual meaning of the ship name can be obtained. Specifically, first, a large number of ship names are obtained, and the ship names are preprocessed, and the ship names without specified features are deleted, and the ship names with specified features are screened out as the first ship name. The first ship name has the specified features, and the specified features include but are not limited to the type code of the ship and the city code of the ship. The first ship name with the specified features can extract the corresponding ship type and the city to which it belongs. Then, a ship name entity set is constructed based on the specified features, and the ship name entity set includes entity data corresponding to the specified features. Further, it is necessary to annotate the first ship name according to a preset annotation method based on the entity data to generate a ship name data set, and by matching the annotations, a corresponding relationship can be formed between each word in the first ship name and the entity data, so as to achieve sentence segmentation of the first ship name, and then extract the fields in the first ship name. Any first ship name includes multiple fields, each field includes at least one character, so it is necessary to divide the first ship name into multiple fields instead of multiple characters by annotation, and then segment the first ship name by annotation, and form a ship name data set with the first ship name and its corresponding entity features, and input the ship name data set into the first neural network structure for training to obtain a ship name annotation model, which can recognize the name of the ship, thereby obtaining the entity data in each ship name. Entity recognition of the ship name is realized. Through the ship name annotation model established in this application, the ship name can be quickly marked, so that the computer can recognize the entity expressed by the ship name, without the need to manually mark the ship name.

[0014] At the same time, the ship name data set includes a training data set, a verification data set and a test data set; the first neural network structure includes: a BiLSTM structure and a CRF structure; the ship name data set is input into the first neural network structure for training, and the steps of obtaining the ship name labeling model include: inputting the training data set into the BiLSTM structure for training to generate a first ship name labeling model; inputting the verification data set and the first ship name labeling model into the CRF structure to generate evaluation parameters; optimizing the first ship name labeling model based on the evaluation parameters to generate a second ship name labeling model; inputting the test data set into the second ship name labeling model for testing to generate test results; determining the second ship name labeling model whose test results meet the preset requirements as the ship name labeling model, training the training data set through the BiLSTM structure, the BiLSTM structure can take into account the information before and after the input information, so it can better capture the bidirectional semantic dependency and improve the performance of the model, and optimizing the first ship name labeling model through the CRF structure and the verification data set can make the labeling results more reasonable. Improve the success rate of model labeling. Further testing through the test data set ensures that the obtained model has good test results and improves the performance of the model.

[0015] Optionally, the preset notation method includes a BIO (B stands for Begin, I stands for Inside, and O stands for Outside) notation method.

[0016] In this technical solution, the preset annotation method includes the BIO annotation method, by which the name of the ship is annotated. Even if there are entities represented by multiple characters in the ship name, they can still be identified and treated as one entity.

[0017] The second aspect of the present invention provides an outlier detection method based on ship names: for ships, the outlier detection method includes: determining the ship name; labeling the ship name through a ship name labeling model to determine the labeling result, the labeling result includes the type code of the ship, and the ship name labeling model is constructed using the ship name labeling model construction method recorded in the first aspect; obtaining a correspondence library of entity information, the correspondence library includes: a corresponding information set of type code and ship type; determining the type of ship according to the ship type code and the corresponding information set; determining the activity range of the ship based on the type of the ship; obtaining the trajectory information of the ship; marking the trajectory information beyond the activity range of the ship as a trajectory outlier.

[0018] The outlier detection method based on the ship name provided by the present invention includes: determining the name information of the ship; determining the type of the ship based on the name information of the ship; determining the activity range of the ship based on the type of the ship; obtaining the trajectory information corresponding to the ship; comparing the trajectory information with the activity range of the ship, and marking the trajectory information beyond the activity range of the ship as a trajectory outlier. In this embodiment, the type of the ship can be determined by the name information of the ship. Different types of ships have different activity rules. Therefore, the activity rules of the ship can be determined by the type of the ship, and the activity range of the ship can be determined according to the activity rules. After that, the trajectory information of the ship is determined, and the trajectory information of the ship is compared with the activity range of the ship, and the points in the trajectory information that exceed the activity range are marked as ship trajectory outliers. Compared with the scheme of obtaining outliers by calculating the speed, direction, time and other data of the ship trajectory in the related art, the method of determining the outliers of the ship trajectory in this application does not require too much data, and can also determine the information of the outliers. In this application, the characteristics that inland and offshore ships will not leave their fixed areas to work are fully utilized. If the trajectory of a ship that moves across regions is found, it means that the ship trajectory is unreasonable, so it is used as an outlier.

[0019] Optionally, determining the activity range of the ship based on the type of the ship includes: determining a navigation pattern of the ship based on the type of the ship, and determining the activity range of the ship based on the navigation pattern.

[0020] In the technical solution, determining the activity range of a ship based on the type of the ship includes: determining the navigation law of the ship based on the type of the ship, and determining the activity range of the ship based on the navigation law. Different types of ships have their own fixed navigation laws and will not break their navigation laws to navigate, so the activity range of the ship can be determined according to the navigation laws. That is to say, after determining the type of the ship, the activity range of the ship can be determined, and the ship will not easily leave the determined activity range of the ship.

[0021] Optionally, based on the type of ship, the navigation pattern of the ship is determined, and the step of determining the activity range of the ship based on the navigation pattern includes: when the type of ship is a cargo ship, determining the city to which the cargo ship belongs according to the marking result and the corresponding relationship library, the marking result also includes the city code of the ship, and the corresponding relationship library also includes: the city code and the corresponding information set of the city; based on the city to which the cargo ship belongs, determining the river basin where the cargo ship is located; based on the river basin where the cargo ship is located, determining the activity range of the cargo ship.

[0022] In this technical solution, based on the type of ship, the steps of determining the activity range of the ship include: when the ship is a cargo ship, determining the city to which the cargo ship belongs; based on the city to which the cargo ship belongs, determining the river basin where the cargo ship is located; based on the river basin where the cargo ship is located, determining the activity range of the cargo ship. When the type of ship is a cargo ship, it means that the ship needs to move between multiple ports, and inland or offshore ships tend to move between multiple designated ports and will not leave these ranges. Therefore, the river basin where the ship is sailing can be determined by the city to which the ship belongs, and then the activity range can be determined. The cargo ship will not leave the activity range easily. Therefore, once the trajectory information of a cargo ship that is not within the activity range appears, it means that the trajectory information belongs to an outlier. There may be situations where "one ship has multiple codes" and "one code has multiple ships".

[0023] Optionally, based on the type of ship, the navigation pattern of the ship is determined, and the step of determining the activity range of the ship based on the navigation pattern includes: when the type of ship is an engineering ship, obtaining the navigation track of the engineering ship within a set time period, and determining the activity center according to the navigation track; determining the first distance based on the activity radius of the engineering ship; determining a circular range with the activity center as the center and the first distance as the radius; and determining the activity range of the engineering ship according to the circular range.

[0024] In the technical scheme, based on the type of ship, the step of determining the activity range of the ship includes: when the ship is an engineering ship, obtaining the navigation track of the engineering ship within a set time period, and determining the activity center according to the navigation track; determining the first distance based on the activity radius of the engineering ship; determining a circular range with the activity center as the center and the first distance as the radius; and determining the activity range of the engineering ship according to the circular range. Engineering ships generally operate around a sea area, so the working range of the engineering ship can be determined by analyzing the navigation track of the project within a set time period. By averaging the longitude and latitude of the navigation track within the set time period, the center of a working range can be obtained as the activity center, and a circular range is circled with the first distance as the radius, and the activity range of the engineering ship is determined according to the circular range. The set time period can be determined according to different types of engineering ships. For example, for engineering ships working in the same sea area for a long time, the set time period can be set to 3 months, while for engineering ships working in the same sea area for a short time, the set time period can be shortened accordingly. At the same time, according to different types of engineering ships, the first distance can be lengthened accordingly. The first distance of conventional engineering ships is 10 nautical miles. When the activity radius of the engineering ship is large and too many outliers are detected, the first distance can be appropriately expanded to make the detected outlier data more accurate.

[0025] Optionally, based on the type of ship, the navigation pattern of the ship is determined, and the step of determining the activity range of the ship based on the navigation pattern includes: when the type of ship is a tugboat, obtaining the navigation track of the tugboat, and determining the port according to the navigation track of the tugboat; determining the second distance based on the activity radius of the tugboat; determining a square range with the port as the center point and the second distance as the side length; and determining the activity range of the tugboat according to the square range.

[0026] In this technical solution, based on the type of ship, the steps of determining the activity range of the ship include: when the ship is a tugboat, obtaining the navigation track of the tugboat, and determining the port according to the navigation track of the tugboat; determining the second distance based on the activity radius of the tugboat; determining a square range with the port as the center point and the second distance as the side length; determining the activity range of the tugboat according to the square range. When the ship is a tugboat, it means that the ship generally moves according to the port. The tugboat in a port will not easily leave the port and go to another port. Therefore, when the ship is a tugboat, the port to which the tugboat belongs can be determined according to the navigation track of the tugboat, and a square range can be established with the port as the center point and the second distance as the side length. The activity range of the tugboat is the square range. The second distance can be specifically set according to different ports. Generally, the watershed that the tugboat is responsible for will also change according to the size of the port and the density of the port. Therefore, the second distance can be set specifically according to the port. Thereby improving the accuracy of the outlier data.

[0027] The third aspect of the present invention provides an outlier detection system based on ship names, including: a name acquisition module, used to determine the name of the ship; a ship type determination module, used to label the ship name through a ship name labeling model, and determine the labeling result, the labeling result includes the type code of the ship, and the ship name labeling model is constructed using the construction method of the ship name labeling model as recorded in the first aspect; a correspondence library acquisition module, used to obtain a correspondence library of entity information, the correspondence library includes: a corresponding information set of type code and ship type; a ship type determination module, used to determine the type of ship according to the ship type code and the corresponding information set; an activity range determination module, used to determine the activity range of the ship based on the type of the ship; a trajectory acquisition module, used to obtain the trajectory information of the ship; an output module, used to mark the trajectory information that exceeds the activity range of the ship as a trajectory outlier.

[0028] The fourth aspect of the present invention provides a system for constructing a ship name labeling model, including a memory and a processor, the memory storing programs or instructions that can be run on the processor, and the programs or instructions, when executed by the processor, implement the steps of the method for constructing a ship name labeling model provided by the first aspect of the present invention.

[0029] The fifth aspect of the present invention provides an outlier detection system based on ship names, including a memory and a processor, the memory storing programs or instructions that can be executed on the processor, and the programs or instructions, when executed by the processor, implement the steps of the outlier detection method based on ship names provided in the second aspect of the present invention.

[0030] The sixth aspect of the present invention provides a readable storage medium having programs and / or instructions stored thereon. When the programs and / or instructions are executed by a processor, the steps of the method for constructing a ship name labeling model provided in the first aspect and / or the steps of the method for detecting outliers based on ship names provided in the second aspect are implemented.

[0031] The existing ship outlier detection technology can filter out most of the outlier track points by cleaning garbled data, processing based on heading angle, setting the distance speed unreachable threshold, etc., but it is all based on the normal update cycle of the ship track point. When a large number of tracks are missing, the speed and distance of the adjacent points before and after can no longer be used as the basis for distinguishing normal track points from outlier track points, and the relevant technology is still difficult to handle. At the same time, the existing related technologies are all based on some pure mathematical methods, and the ship's own attributes are rarely mined. The present invention combines the information attributes after the Chinese ship name recognition classification to define the ship's geographical activity range, detect outliers that do not belong to the ship's activity range and filter them. At the same time, when a large number of ship tracks are missing, the present invention can also detect ship outliers well, solve the phenomenon of ship track missing, multi-segment track, track jump, etc. caused by human tampering ("one ship with multiple codes", "one code with multiple ships"), equipment failure, signal interference, channel congestion, etc., and facilitate personnel in navigation and related fields to obtain the accurate position and historical track of the ship when querying the ship.

[0032] Additional aspects and advantages of the present invention will become apparent from the following description or may be learned by practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] The above and / or additional aspects and advantages of the present invention will become apparent and easily understood from the description of the embodiments in conjunction with the following drawings, in which:

[0034] Figure 1 It is one of the flow charts of the method for constructing a ship name labeling model according to an embodiment of the present invention;

[0035] Figure 2 is one of the flow charts of the outlier detection method based on ship names according to an embodiment of the present invention;

[0036] Figure 3 is one of the block diagrams of the outlier detection system based on ship name according to an embodiment of the present invention;

[0037] Figure 4 is a block diagram of a system for constructing a ship name labeling model according to an embodiment of the present invention;

[0038] Figure 5 is a second block diagram of an outlier detection system based on ship names according to an embodiment of the present invention;

[0039] Figure 6 is a schematic diagram of a ship name marking process according to an embodiment of the present invention;

[0040] Figure 7 is a schematic diagram of trajectory outliers detected according to an embodiment of the present invention;

[0041] Figure 8 This is a second flow chart of a method for detecting outliers based on ship names according to an embodiment of the present invention. DETAILED DESCRIPTION

[0042] In order to more clearly understand the above-mentioned purposes, features and advantages of the present invention, the present invention is optionally described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be noted that the embodiments of the present application and the features in the embodiments can be combined with each other without conflict.

[0043] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Therefore, the protection scope of the present invention is not limited to the specific embodiments disclosed below.

[0044] A first aspect of the present invention provides a method for constructing a ship name labeling model.

[0045] Refer to the following Figure 1 The invention describes a method for constructing a model based on ship name marking in some embodiments of the invention.

[0046] like Figure 1 As shown, the construction method of the ship name labeling model includes:

[0047] S101: Obtain multiple ship names;

[0048] S103: Preprocessing the multiple ship names, screening out a ship name with a specified feature as a first ship name, where the specified feature includes a ship type code and a ship city code;

[0049] S105: constructing a ship name entity set based on the specified features, where the ship name entity set includes entity data corresponding to the specified features;

[0050] S107: annotating the first ship name according to a preset annotation method based on the entity data to generate a ship name data set;

[0051] S109: input the training data set into the BiLSTM structure for training, and generate a first ship name labeling model;

[0052] S111: Input the verification data set and the first ship name labeling model into the CRF structure to generate evaluation parameters;

[0053] S113: Optimizing the first ship name labeling model based on the evaluation parameters to generate a second ship name labeling model;

[0054] S115: inputting the test data set into the second ship name labeling model for testing, and generating a test result;

[0055] S117: Determine the second ship name labeling model whose test results meet the preset requirements as the ship name labeling model.

[0056] In this embodiment, the method for constructing a ship name labeling model includes: obtaining multiple ship names; preprocessing the multiple ship names, screening out ship names with specified features as the first ship name, the specified features including the ship type code and the ship city code; constructing a ship name entity set based on the specified features, the ship name entity set including entity data corresponding to the specified features; labeling the first ship name according to a preset labeling method based on the entity data to generate a ship name data set; inputting the ship name data set into a first neural network structure for training to obtain a ship name labeling model. The ship name data set includes a training data set, a verification data set and a test data set; the first neural network structure includes: a BiLSTM structure and a CRF structure. Afterwards, the training data set is input into the BiLSTM structure for training to generate a first ship name labeling model; the verification data set and the first ship name labeling model are input into the CRF structure to generate evaluation parameters; the first ship name labeling model is optimized based on the evaluation parameters to generate a second ship name labeling model; the test data set is input into the second ship name labeling model for testing to generate test results; the second ship name labeling model whose test results meet the preset requirements is determined as the ship name labeling model. By constructing a ship name annotation model, the name of the ship can be annotated, and then the actual meaning of the ship name can be obtained. Specifically, first, a large number of ship names are obtained, and the ship names are preprocessed, and the ship names without specified features are deleted, and the ship names with specified features are screened out as the first ship name. The first ship name has the specified features, and the specified features include but are not limited to the type code of the ship and the city code of the ship. The first ship name with the specified features can extract the corresponding ship type and the city to which it belongs. Then, a ship name entity set is constructed based on the specified features. The ship name entity set includes entity data corresponding to the specified features. Further, it is necessary to annotate the first ship name according to a preset annotation method based on the entity data to generate a ship name data set, and by matching the annotations, a corresponding relationship can be formed with each word in the first ship name and the entity data, so as to achieve the first ship name segmentation, and then extract the fields in the first ship name. Any first ship name includes multiple fields, each field includes at least one character, so it is necessary to divide the first ship name into multiple fields rather than multiple characters by annotation, and then segment the first ship name by annotation, and form a ship name data set with the first ship name and its corresponding entity features. The ship name data set includes a training data set, a verification data set and a test data set; the first neural network structure includes: a BiLSTM structure and a CRF structure.Input the training data set into the BiLSTM structure for training to generate the first ship name labeling model; input the verification data set and the first ship name labeling model into the CRF structure to generate evaluation parameters; optimize the first ship name labeling model based on the evaluation parameters to generate the second ship name labeling model; input the test data set into the second ship name labeling model for testing to generate test results; determine the second ship name labeling model whose test results meet the preset requirements as the ship name labeling model. The training data set is trained by the BiLSTM structure. The BiLSTM structure can take into account the information before and after the input information, so it can better capture the bidirectional semantic dependency and improve the performance of the model. The first ship name labeling model is optimized by the CRF structure and the verification data set, which can make the labeling results more reasonable. Improve the success rate of the model's labeling. Further testing with the test data set ensures that the obtained model has good test results and improves the performance of the model. The model can recognize the name of the ship, thereby obtaining the entity data in each ship name. Realize entity recognition of the ship name. Through the ship name labeling model established in the present application, the name of the ship can be quickly labeled, so that the computer can recognize the entity expressed by the ship name without the need for manual labeling of the ship name.

[0057] Optionally, the preset labeling method includes BIO labeling method.

[0058] In this embodiment, the preset annotation method includes a BIO annotation method, and the name of the ship is annotated by the BIO annotation method. Even if there are entities represented by multiple characters in the ship name, they can still be identified and treated as one entity.

[0059] A second aspect of the present invention provides a method for detecting outliers based on ship names: for ships, the method for detecting outliers is as follows: Figure 2 As shown, including:

[0060] S201: Determine the name of the ship;

[0061] S203: labeling the ship name by using the ship name labeling model to determine the labeling result, where the labeling result includes the ship type code, and the ship name labeling model is constructed by using the ship name labeling model construction method recorded in the first aspect;

[0062] S205: Acquire a corresponding relationship library of entity information, the corresponding relationship library including: a corresponding information set of type code and ship type;

[0063] S207: Determine the type of the ship according to the type code of the ship and the corresponding information set;

[0064] S209: Determine the activity range of the ship based on the type of the ship;

[0065] S211: Acquire the track information of the ship; mark the track information beyond the activity range of the ship as a track outlier.

[0066] The outlier detection method based on the ship name provided by the present invention includes: determining the name information of the ship; determining the type of the ship based on the name information of the ship; determining the activity range of the ship based on the type of the ship; obtaining the trajectory information corresponding to the ship; comparing the trajectory information with the activity range of the ship, and marking the trajectory information beyond the activity range of the ship as a trajectory outlier. In this embodiment, the type of the ship can be determined by the name information of the ship. Different types of ships have different activity rules. Therefore, the activity rules of the ship can be determined by the type of the ship, and the activity range of the ship can be determined according to the activity rules. After that, the trajectory information of the ship is determined, and the trajectory information of the ship is compared with the activity range of the ship, and the points in the trajectory information that exceed the activity range are marked as ship trajectory outliers. Compared with the scheme of obtaining outliers by calculating the speed, direction, time and other data of the ship trajectory in the related art, the method of determining the outliers of the ship trajectory in this application does not require too much data, and can also determine the information of the outliers. In this application, the characteristics that inland and offshore ships will not leave their fixed areas to work are fully utilized. If the trajectory of a ship that moves across regions is found, it means that the ship trajectory is unreasonable, so it is used as an outlier.

[0067] Optionally, the ship name consists of the type of ship, the city to which the ship belongs, and the ship number.

[0068] In this embodiment, the name information includes the type of ship, such as a cargo ship, engineering ship, tugboat, tanker, etc., the city to which the ship belongs (such as City A, City B) and the ship number, representing a specific ship. The track information of the ship can be determined by the ship number, and the track information of the ship can be determined by the name information of the ship. The type of ship and the city to which the ship belongs can also be determined, and then the activity range of the ship can be determined. The track outliers of the ship can be determined by the track information and the activity range.

[0069] Optionally, determining the activity range of the ship based on the type of the ship includes: determining a navigation pattern of the ship based on the type of the ship, and determining the activity range of the ship based on the navigation pattern.

[0070] In this embodiment, determining the activity range of the ship based on the type of the ship includes: determining the navigation law of the ship based on the type of the ship, and determining the activity range of the ship based on the navigation law. Different types of ships have their fixed navigation laws, and most ships will not break their navigation laws to navigate, so the activity range of the ship can be determined according to the navigation laws. That is, after determining the type of the ship, the activity range of the ship can be determined, and the ship will not easily leave the determined activity range of the ship.

[0071] Optionally, based on the type of ship, the navigation pattern of the ship is determined, and the step of determining the activity range of the ship based on the navigation pattern includes: when the type of ship is a cargo ship, determining the city to which the cargo ship belongs according to the marking result and the corresponding relationship library, the marking result also includes the city code of the ship, and the corresponding relationship library also includes: the city code and the corresponding information set of the city; based on the city to which the cargo ship belongs, determining the river basin where the cargo ship is located; based on the river basin where the cargo ship is located, determining the activity range of the cargo ship.

[0072] In this embodiment, based on the type of ship, the step of determining the activity range of the ship includes: when the ship is a cargo ship, determining the city to which the cargo ship belongs; based on the city to which the cargo ship belongs, determining the river basin where the cargo ship is located; based on the river basin where the cargo ship is located, determining the activity range of the cargo ship. When the type of ship is a cargo ship, it means that the ship needs to move between multiple ports, and inland or offshore ships tend to move between multiple designated ports and will not leave these ranges. Therefore, the river basin where the ship is sailing can be determined by the city to which the ship belongs, and then the activity range can be determined. The cargo ship will not leave the activity range easily. Therefore, once the trajectory information of a cargo ship that is not within the activity range appears, it means that the trajectory information belongs to an outlier. There may be situations of "one ship with multiple codes" and "one code with multiple ships".

[0073] Optionally, based on the type of ship, the navigation pattern of the ship is determined, and the step of determining the activity range of the ship based on the navigation pattern includes: when the type of ship is an engineering ship, obtaining the navigation track of the engineering ship within a set time period, and determining the activity center according to the navigation track; determining the first distance based on the activity radius of the engineering ship; determining a circular range with the activity center as the center and the first distance as the radius; and determining the activity range of the engineering ship according to the circular range.

[0074] In this embodiment, based on the type of ship, the step of determining the activity range of the ship includes: when the ship is an engineering ship, obtaining the navigation track of the engineering ship within a set time period, and determining the activity center according to the navigation track; determining the first distance based on the activity radius of the engineering ship; determining a circular range with the activity center as the center and the first distance as the radius; and determining the activity range of the engineering ship according to the circular range. Engineering ships generally operate around a sea area, so the working range of the engineering ship can be determined by analyzing the navigation track of the project within a set time period. By averaging the longitude and latitude of the navigation track within the set time period, the center of a working range can be obtained as the activity center, and a circular range is circled with the first distance as the radius, and the activity range of the engineering ship is determined according to the circular range. The set time period can be determined according to different types of engineering ships. For example, for an engineering ship working in the same sea area for a long time, the set time period can be set to 3 months, while for an engineering ship working in the same sea area for a short time, the set time period can be shortened accordingly. At the same time, according to different types of engineering ships, the first distance can be lengthened accordingly. The first distance of conventional engineering ships is 10 nautical miles. When the activity radius of the engineering ship is large and too many outliers are detected, the first distance can be appropriately expanded to make the detected outlier data more accurate.

[0075] Optionally, based on the type of ship, the navigation pattern of the ship is determined, and the step of determining the activity range of the ship based on the navigation pattern includes: when the type of ship is a tugboat, obtaining the navigation track of the tugboat, and determining the port according to the navigation track of the tugboat; determining the second distance based on the activity radius of the tugboat; determining a square range with the port as the center point and the second distance as the side length; and determining the activity range of the tugboat according to the square range.

[0076] In this embodiment, based on the type of ship, the step of determining the activity range of the ship includes: when the ship is a tugboat, obtaining the navigation track of the tugboat, and determining the port according to the navigation track of the tugboat; determining the second distance based on the activity radius of the tugboat; determining a square range with the port as the center point and the second distance as the side length; determining the activity range of the tugboat according to the square range. When the ship is a tugboat, it means that the ship generally moves according to the port, and the tugboat in a port will not easily leave the port and go to another port. Therefore, when the ship is a tugboat, the port to which the tugboat belongs can be determined according to the navigation track of the tugboat, and a square range is established with the port as the center point and the second distance as the side length. The activity range of the tugboat is the square range. The second distance can be specifically set according to different ports. Generally, the watershed that the tugboat is responsible for will also change according to the size of the port and the density of the port. Therefore, the second distance can be specifically set according to the port. Thereby improving the accuracy of the outlier data.

[0077] Furthermore, the outlier detection method based on ship names disclosed in this application, as Figure 8 shown, includes:

[0078] S1: Data preparation: Collect ship certificate materials and construct a ship name dictionary. First, based on the materials collected from past ship certificates and the matching results of AIS (Automatic Identification System) information, prepare an annotated dataset containing ship names, and construct a Chinese ship name dictionary containing approximately 23,000+. It should be noted that for each ship name in the dictionary, there is one or more corresponding ships.

[0079] S2: Data preprocessing: Preprocess the data and filter the ship names to prepare for data annotation. Before training the NER model, the data needs to be preprocessed. Perform part-of-speech tagging and filtering on the ship names. If the ship name in the dictionary only corresponds to one ship and the ship name does not contain any geographical location or ship type characteristics (such as words like "tug", "work", "cargo", "oil", etc.), it will be deleted and not used as the dataset for model calculation.

[0080] S3: Data annotation: Select the BIO annotation method to annotate the ship names. In order to identify and annotate, for example, the ship name "Ac Cargo 12345" as "A" (province), "c" (city), "Cargo" (ship type), "12345" (number), where "A" and "c" are both codes, and in the actual ship name, "A" and "c" can be "text" such as "XX Cargo 12345". The actual ship name is similar to the expression form of a license plate. We select the commonly used BIO annotation method under sequence annotation. This method can handle composite entities (i.e., an entity consists of multiple parts, such as the ship name here consists of multiple parts like province, city, type, and number). The BIO annotation method includes: B-X (representing the beginning of entity X), I-X (representing the middle or end of entity X), O (representing an entity that does not belong to any type), Figure 6 is the result after annotation by this method.

[0081] S4: Model training: Divide the training set, validation set and test set, and set the index parameters to optimize the model. The labeled ship name dataset is divided into training set, validation set and test set, of which the training set accounts for 70%, the validation set accounts for 15%, and the test set accounts for 15%. Convert the text type into a feature vector that the model can process, and select a suitable model architecture. This example uses BiLSTM-CRF (a deep memory model that combines a bidirectional long short-term memory network and a conditional random field) for processing. Generate a prediction result through the model for the input data, and calculate the difference between the prediction result and the true label (such as cross entropy loss). Calculate the gradient based on the loss and update the model parameters. Monitor the loss function, accuracy, recall and other indicators during the training process. Here we set the threshold of the index rate as follows: Precision is higher than 0.8, Recall is higher than 0.8, and F1 value (the harmonic mean of F1Score accuracy and recall) is higher than 0.85. Through this model training, a ship name can be received and labeled and distinguished.

[0082] S5: Outlier detection: Ship analysis is brought into the model trained in S4 to detect outliers: Based on the results of the S4 model training, when analyzing the trajectory points of the ship, the ship name can be used to distinguish the label and determine the city and ship type to which it belongs. Furthermore, the cities are mainly cities along the A River, cities along the B River, etc., and the types include cargo (cargo ships), engineering (engineering ships), tugboats, fishing (fishing boats) and oil (oil tankers). Furthermore, 1) For cargo ships, the scope is restricted by aligning the provinces identified after labeling. For example, if the province is in the A River Basin, its scope is the A River Basin. If the identified province is in the B River Basin, then its activity range is the B River Basin; 2) For engineering ships, its activity range is roughly within 10 nautical miles around, with the latitude and longitude center point represented by its trajectory that appeared in the past 1 to 3 months as the center of the circle and 10 nautical miles as the radius to set its activity range; 3) For tugboats, their activity range is usually within 50 nautical miles of the port area, and the area of ​​50 nautical miles square in the port area is its activity range. When points exceeding the above range appear, they will be detected as outliers and filtered, and will not be displayed when generating ship trajectories. Figure 7 As shown in the figure, the activity range of a certain ship is set in River Basin A, so most of the track points of the ship also appear in the activity area. However, the ship also has track points in River Basin B, which is not within the activity range. Therefore, the track points that appear in River Basin B are marked as track outliers.

[0083] This solution has the following advantages:

[0084] 1) After the related art translates the ship names of inland waterway transport ships in combination with geographic information, no further mining of the Chinese ship names is performed. The present invention constructs a CNR Chinese ship name dataset based on the matched Chinese ship names, mines the ship's place of origin, ship type, and main navigation range of the ship, and provides a reference for subsequent trajectory outlier detection.

[0085] 2) When related technologies detect and process ship outliers, most of them are based on the attributes of the trajectory points themselves. They use pure mathematical methods to judge ship outliers based on the speed difference, distance difference, heading difference, etc. between two adjacent trajectory points, using speed thresholds or acceleration thresholds, ignoring the information that can be provided by the ship's own attributes. The present invention combines the information attributes after Chinese ship name recognition and classification to define the geographical activity range of the ship, detect outliers that do not belong to the ship's activity range and filter them. At the same time, when a large number of ship trajectories are missing, the present invention can also detect ship outliers well.

[0086] In this application, a Chinese ship name recognition data set is constructed by matching the names of a large number of ships in Chinese translation, marking the ship names, setting location tags and ship attribute tags, and defining the activity range of the ship. Different from the traditional pure mathematical outlier detection method, the present invention mines the ship's own attributes and combines the mined information to propose a new ship outlier detection method, which can improve the problem of low accuracy of ship outlier detection when a large number of ship trajectories are missing.

[0087] This solution mainly includes two methods: a labeling method for creating a Chinese ship name NER dataset and a method for detecting ship outliers. The labeling method is: by collecting the user's ship information or translating the AIS English ship name, a labeling dataset containing ships is prepared, and a Chinese ship name dataset is constructed. It should be noted that for each Chinese ship name in the dataset, there is one or more corresponding ships matching it. The Chinese ship name calibration method is set to split the Chinese ship name, such as provincial abbreviation vocabulary + place name pinyin vocabulary + ship type vocabulary + digital number, and identify the province (Province), city (City) and type (TYPE); the outlier detection method is: based on the ship attributes mined in the previous step, combined with geographic information, the ship's activity range is defined. For example, the activity range of the ship Su Yancheng Cargo 1234 is the Yangtze River section. When it appears at a point beyond the activity range such as the Guangxi region and Hainan region, it is detected as an outlier and filtered. This method can perform detection regardless of whether the ship position is continuously updated or a large number of trajectories are missing, which can improve the accuracy of ship outlier detection. In the actual production and operation field, it is convenient for navigation and related personnel to obtain accurate ship position information when querying ships.

[0088] The third aspect of the present invention provides a ship name based outlier detection system 300, such as Figure 3 As shown, it includes: a name acquisition module 301, which is used to determine the name of the ship; a ship type determination module 303, which is used to annotate the ship name through a ship name annotation model to determine the annotation result, the annotation result includes the type code of the ship, and the ship name annotation model is constructed by the construction method of the ship name annotation model as recorded in the first aspect; a corresponding relationship library acquisition module 305, which is used to obtain the corresponding relationship library of entity information, and the corresponding relationship library includes: a corresponding information set of type code and ship type; a ship type determination module 307, which is used to determine the type of the ship according to the type code and the corresponding information set of the ship; an activity range determination module 309, which is used to determine the activity range of the ship based on the type of the ship; a trajectory acquisition module 311, which is used to obtain the trajectory information of the ship; an output module 313, which is used to mark the trajectory information beyond the activity range of the ship as a trajectory outlier. Since it is used to implement the steps of the outlier detection method based on the ship name provided in the second aspect of the present invention, the outlier detection system 300 based on the ship name has all the technical effects of the outlier detection method based on the ship name, which will not be repeated here.

[0089] A fourth aspect of the present invention provides a system 400 for constructing a ship name labeling model, such as Figure 4 As shown, it includes a memory 402 and a processor 404, the memory stores programs or instructions that can be run on the processor, and when the programs or instructions are executed by the processor, the steps of the method for constructing a ship name marking model provided by the first aspect of the present invention are implemented.

[0090] The ship name labeling model construction system 400 provided in accordance with an embodiment of the present invention comprises a memory 402 and a processor 404, and a program stored in the memory 402 and executable on the processor 404, and when the program is executed by the processor 404, the steps defined in the ship name labeling model construction method described in the first aspect of the present invention are implemented. At the same time, since the ship name labeling model construction system 400 of the present application can implement the steps of the ship name labeling model construction method provided in the first aspect of the present invention, the ship name labeling model construction system 400 provided in this embodiment has all the beneficial effects of the ship name labeling model construction method described in the first aspect.

[0091] A fifth aspect of the present invention provides a ship name-based outlier detection system 500, such as Figure 5 As shown, it includes a memory 502 and a processor 504, the memory stores programs or instructions that can be run on the processor, and when the programs or instructions are executed by the processor, the steps of the outlier detection method based on ship names provided by the second aspect of the present invention are implemented.

[0092] The outlier detection system 500 based on ship names provided in an embodiment of the present invention comprises a memory 502, a processor 504, and a program stored in the memory 502 and executable on the processor 504. When the program is executed by the processor 504, the steps defined in the outlier detection method based on ship names as described in the second aspect of the present invention are implemented. At the same time, since the outlier detection system 500 based on ship names of the present application can implement the steps of the outlier detection method based on ship names provided in the second aspect of the present invention, the outlier detection system 500 based on ship names provided in this embodiment has all the beneficial effects of the outlier detection method based on ship names as described in the second aspect above.

[0093] The sixth aspect of the present invention provides a readable storage medium having programs and / or instructions stored thereon. When the programs and / or instructions are executed by a processor, the steps of the method for constructing a ship name labeling model provided in the first aspect and / or the steps of the method for detecting outliers based on ship names provided in the second aspect are implemented.

[0094] The readable storage medium provided according to an embodiment of the present invention can implement the steps of the method for constructing a ship name labeling model provided in the first aspect and / or the steps of the outlier detection method based on ship names provided in the second aspect of the present invention when the program and / or instructions stored thereon are executed by a processor. Therefore, it has all the beneficial technical effects of the above-mentioned method for constructing a ship name labeling model provided in the first aspect and / or the outlier detection method based on ship names provided in the second aspect, which will not be repeated here.

[0095] In this specification, the term "plurality" means two or more than two, unless otherwise clearly defined. The terms "installed", "connected", "connected", "fixed" and the like should be understood in a broad sense. For example, "connected" can be a fixed connection, a detachable connection, or an integral connection; "connected" can be a direct connection or an indirect connection through an intermediate medium. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood according to the specific circumstances.

[0096] In the description of this specification, the description of the terms "one embodiment", "some embodiments", etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.

[0097] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for outlier detection based on ship names, characterized in that: For ships, the outlier detection method includes: Determine the name of the vessel; The ship name is annotated by a ship name annotating model to determine an annotation result, wherein the annotation result includes a type code of the ship, and the ship name annotating model is constructed by a method for constructing a ship name annotating model; Acquire a corresponding relationship library of entity information, the corresponding relationship library comprising: a corresponding information set of type code and ship type; Determining the type of the ship according to the type code of the ship and the corresponding information set; Based on the type of the ship, determining the activity range of the ship; Obtaining track information of the ship; Marking the track information beyond the activity range of the vessel as a track outlier; The method for constructing the ship name marking model includes: Get multiple ship names; Preprocessing the multiple ship names, screening out the ship names with specified features as the first ship name, the specified features including the type code of the ship and the code of the city to which the ship belongs; Building a ship name entity set based on the specified features, the ship name entity set including entity data corresponding to the specified features; Based on the entity data, the first ship name is annotated according to a preset annotation method to generate a ship name data set; Inputting the ship name data set into a first neural network structure for training to obtain a ship name labeling model; The ship name data set includes a training data set, a verification data set and a test data set; The first neural network structure includes: a BiLSTM structure and a CRF structure; The step of inputting the ship name data set into a first neural network structure for training to obtain a ship name labeling model comprises: Inputting the training data set into the BiLSTM structure for training to generate a first ship name labeling model; Inputting the verification data set into the first ship name labeling model, and inputting the obtained results into the CRF structure to generate evaluation parameters; Optimizing the first ship name labeling model based on the evaluation parameters to generate a second ship name labeling model; Inputting the test data set into the second ship name labeling model for testing, and generating a test result; The second ship name labeling model whose test result meets the preset requirements is determined as the ship name labeling model.

2. The outlier detection method based on ship name according to claim 1, characterized in that: The preset labeling method includes the BIO labeling method.

3. The outlier detection method based on ship name according to claim 1 or 2, characterized in that: Based on the type of the vessel, determining the activity range of the vessel includes: Based on the type of the ship, the navigation pattern of the ship is determined, and based on the navigation pattern, the activity range of the ship is determined.

4. The outlier detection method based on ship name according to claim 3, characterized in that: The step of determining the navigation law of the ship based on the type of the ship, and determining the activity range of the ship based on the navigation law comprises: When the type of the ship is a cargo ship, the city to which the cargo ship belongs is determined according to the marking result and the corresponding relationship library, the marking result also includes the city code to which the ship belongs, and the corresponding relationship library also includes: a corresponding information set of the city code and the city; Based on the city to which the cargo ship belongs, determining the river basin where the cargo ship is located; Based on the watershed where the cargo ship is located, the activity range of the cargo ship is determined.

5. The outlier detection method based on ship name according to claim 3, characterized in that: The step of determining the navigation law of the ship based on the type of the ship, and determining the activity range of the ship based on the navigation law comprises: When the type of the ship is an engineering ship, obtaining a navigation track of the engineering ship within a set time period, and determining an activity center according to the navigation track; Determining a first distance based on an activity radius of the engineering vessel; Determine a circular range with the activity center as the center and the first distance as the radius; The activity range of the engineering vessel is determined according to the circular range.

6. The outlier detection method based on ship name according to claim 3, characterized in that: The step of determining the navigation law of the ship based on the type of the ship, and determining the activity range of the ship based on the navigation law comprises: When the type of the ship is a tugboat, obtaining the navigation track of the tugboat, and determining the port according to the navigation track of the tugboat; determining a second distance based on an activity radius of the tugboat; Determine a square range with the port as the center point and the second distance as the side length; The activity range of the tugboat is determined according to the square range.

7. A ship name based outlier detection system, characterized in that: For ships, the outlier detection system comprises: A name acquisition module is used to determine the name of the ship; A ship type determination module, used to mark the ship name through a ship name marking model to determine a marking result, wherein the marking result includes a type code of the ship, and the ship name marking model is constructed using a ship name marking model construction method; A corresponding relationship library acquisition module is used to acquire a corresponding relationship library of entity information, wherein the corresponding relationship library includes: a corresponding information set of type code and ship type; A ship type determination module, used to determine the type of the ship according to the type code of the ship and the corresponding information set; An activity range determination module, used to determine the activity range of the ship based on the type of the ship; A track acquisition module, used to acquire track information of the ship; An output module, used for marking the track information beyond the activity range of the ship as a track outlier; The method for constructing the ship name marking model includes: Get multiple ship names; Preprocessing the multiple ship names, screening out the ship names with specified features as the first ship name, the specified features including the type code of the ship and the code of the city to which the ship belongs; Building a ship name entity set based on the specified features, the ship name entity set including entity data corresponding to the specified features; Based on the entity data, the first ship name is annotated according to a preset annotation method to generate a ship name data set; Inputting the ship name data set into a first neural network structure for training to obtain a ship name labeling model; The ship name data set includes a training data set, a verification data set and a test data set; The first neural network structure includes: a BiLSTM structure and a CRF structure; The step of inputting the ship name data set into a first neural network structure for training to obtain a ship name labeling model comprises: Inputting the training data set into the BiLSTM structure for training to generate a first ship name labeling model; Inputting the verification data set into the first ship name labeling model, and inputting the obtained results into the CRF structure to generate evaluation parameters; Optimizing the first ship name labeling model based on the evaluation parameters to generate a second ship name labeling model; Inputting the test data set into the second ship name labeling model for testing, and generating a test result; The second ship name labeling model whose test result meets the preset requirements is determined as the ship name labeling model.

8. A ship name based outlier detection system, characterized in that: The method comprises a memory and a processor, wherein the memory stores a program or an instruction that can be run on the processor, and when the program or the instruction is executed by the processor, the steps of the outlier detection method based on ship names as described in any one of claims 1 to 6 are implemented.

9. A readable storage medium, characterized in that: Programs and / or instructions are stored thereon, and when the programs and / or the instructions are executed by a processor, the steps of the outlier detection method based on ship names as described in any one of claims 1 to 6 are implemented.

Images