A method for constructing a knowledge graph of ship activities

Abstract

The present invention belongs to the technical field of ship activity analysis, and specifically relates to a method for constructing a ship activity knowledge graph. Based on the core idea of ​​"process-event-behavior", the present invention adds two entities, process and behavior, on the basis of the SEM model, and designs a ship activity knowledge graph ontology model on this basis, so as to more completely express the ship's activity behavior at sea from ship behavior to navigation process; then, combined with the marine ship activity data, the marine ship data is semantically analyzed and processed, the entities of the marine ship data are extracted, and the association relationship between the entities is determined to complete the instance filling. The method of the present invention can support the knowledge representation of routine ship activities and emergencies, and can achieve the semantic enhancement effect of spatiotemporal activity query and backtracking, and further improve the efficiency of maritime traffic monitoring and management, and has certain application value.

Description

Technical Field

[0001] The present invention belongs to the technical field of ship activity analysis, and specifically relates to a method for constructing a ship activity knowledge graph. Background Art

[0002] With the deepening of global economic integration, the number of ships has continued to grow in recent years, maritime ship activities have become increasingly frequent, and maritime ship accidents have occurred from time to time. In order to ensure the safe and efficient operation of maritime traffic, maritime authorities and coast guard departments of various countries have jointly established a global ship automatic identification and maritime traffic monitoring system (Automatic Identification System, AIS) to enhance users' comprehensive situational awareness of global maritime traffic. In the 30 years of operation, the AIS System has accumulated a huge amount of ship positioning data. At the same time, there are also public data sources related to maritime traffic incidents around the world. These data sources provide a huge space for data-driven technology innovation for maritime traffic management.

[0003] Many researchers have tried to apply big data analysis technology to ship trajectory analysis to help maritime traffic monitoring and management develop towards intelligence. Traditional methods mainly use ship positioning data to mine the routine activities of ships, without combining other maritime multi-source event data to deeply analyze the emergencies and accidents of ships, and lack in-depth knowledge mining. Therefore, how to use cutting-edge technologies such as artificial intelligence to efficiently improve maritime traffic monitoring and management has put forward new demands and challenges for the mining of ship trajectory data and maritime multi-source event data. In existing research, the main focus is on mining semantic information based on ship trajectory data to analyze ship activity behavior. In addition, some scholars start from ship event data to study the laws of ship activities based on event analysis. However, the model describing ship events uses events as the basic unit and cannot express a certain navigation activity process and the basic behavior of the ship. However, in order to deeply analyze the causes of ship events and more completely express ship activities, it is often necessary to combine all event sets in the ship's navigation process and the ship's behavior information before and after the event, such as whether the ship has deceleration behavior before and after the collision; whether there is a collision before the sinking. Therefore, it is very necessary to construct a conceptual model that can describe the ship's navigation process and behavior pattern.

[0004] The trajectory data of marine ship activities has spatio-temporal characteristics. Modeling based on trajectory spatio-temporal data can express the spatio-temporal activity rules of ship navigation. The existing spatio-temporal data models (STDM) can be mainly divided into the following four types: object-oriented, event-based, process-oriented and graph-based spatio-temporal data models. Object-oriented spatio-temporal data models can characterize the historical changes of spatio-temporal objects, but it is difficult to express the relationship between spatio-temporal objects in the process of change; event-based spatio-temporal data models can express the spatio-temporal change process more clearly, but do not better reflect the driving force of events on the spatio-temporal change of objects; process-oriented spatio-temporal data models mainly express the sequential relationship and logical relationship of spatio-temporal process changes, and need to be improved in expressing the causal relationship of changes; graph-based spatio-temporal data models enhance the topological semantics and causal relationship expression of spatio-temporal changes, which is more suitable for discrete spatio-temporal change processes, but the expression of continuous changes needs further research. Due to the unfixed track of marine ship spatio-temporal activities and the characteristics of ship events such as suddenness, continuity and correlation, the above spatio-temporal data models are difficult to fully describe the entire event and process of ship spatio-temporal activities.

[0005] In addition, the modeling of ship activity events can be based on general event models. At present, the academic community uses concept-based event models and attribute-based event models to describe event models. Among them, the concept-based event models mainly include SEM (Simple Event Model) model, EO (Event Ontology Model) model and CIDOC-CRM (CIDOC Conceptual Reference Model) model. The SEM ontology model proposed by Hage represents and reasoned about events by defining classes, attributes and constraints. Its core classes include events (Event), actors (Actor), places (Place) and time (Time), aiming to improve the versatility of the model. Other events can be expanded on this basis. EO is a lightweight small ontology based on concepts. The design of EO is very simple, consisting of 4 classes (Event and 3 implicit classes Agent, Factor and Product) and 17 attributes. EO defines the minimum event and relies on an external vocabulary to refine the knowledge expressed. Similar to SEM, it is modularly designed, so it is more flexible. However, EO does not have explicit Actor and Place classes. CIDOC-CRM is a large concept-based ontology with no formal restrictions. It contains 140 classes and 144 attributes. A subset of them can be used to represent events. SEM generalizes the CIDOC-CRM model and adds View representation. The attribute-based event model includes the LODE model (Linking Open Descriptions of Events). LODE is a lightweight attribute-based ontology that contains one class (Event) and six attributes (atTime, circa, inSpace, atPlace, involved, and involvedAgent). Like EO, LODE does not have explicit Actor and Place classes. Compared with the LODE model, SEM also provides lightweight description elements for events, but does not introduce strong semantic definitions that are prone to inconsistencies. In addition, SEM uses types, constraints, and authorities to better access external data. Hage then modeled the ship trajectory based on the SEM model to identify the ship's events, realizing the transformation from ship trajectory data to ship event semantic information. However, events extracted based solely on the SEM model can only describe routine events of the ship, and still cannot express the relationship between the ship's activity process and behavior. Summary of the invention

[0006] The purpose of the present invention is to provide a method for constructing a ship activity knowledge graph to solve the problem that the ship activities cannot be fully expressed using the SEM model.

[0007] In order to solve the above technical problems, the technical solutions provided by the present invention and the corresponding beneficial effects of the technical solutions are as follows:

[0008] The present invention provides a method for constructing a ship activity knowledge graph, comprising the following steps:

[0009] 1) Based on the six entities of process, event, participant, place, time and behavior, a ship activity knowledge graph ontology model is constructed; wherein the process represents a sea voyage process of a ship from the port trigger to the next port arrival, the participant represents the ship, the place represents an entity with a spatial location, the time represents an entity with a time feature, and the behavior represents the basic behavior of the ship activity;

[0010] 2) Acquire maritime ship data, perform semantic analysis on the acquired maritime ship data, extract the entities of the maritime ship data, and determine the relationship between the entities; according to the entities of the maritime ship data and the relationship between the entities, fill in the constructed ship activity knowledge graph ontology model with instances to complete the construction of the ship knowledge activity graph.

[0011] The beneficial effects of the above technical solution are as follows: in the face of the problem that the SEM model cannot fully express the process and behavior of the ship, the present invention is based on the core idea of ​​"process-event-behavior". On the basis of the SEM model, two entities, process and behavior, are added. On this basis, a ship activity knowledge graph ontology model is designed to more completely express the ship's activities at sea from ship behavior to navigation process; then, combined with the marine ship activity data, the marine ship data is semantically analyzed and processed, the entities of the marine ship data are extracted, and the association relationship between the entities is determined to complete the instance filling. The method of the present invention can support the knowledge representation of routine ship activities and emergencies, and can achieve the semantic enhancement effect of spatiotemporal activity query and backtracking, and further improve the efficiency of marine traffic monitoring and management, and has certain application value.

[0012] Further, in step 2), the association relationship between the entities includes: an inclusion relationship between events and behaviors, an inclusion relationship between processes and events, a causal relationship between events, and a sequential relationship between behaviors; the inclusion relationship between events and behaviors indicates that an event includes at least one behavior; the inclusion relationship between processes and events indicates that a process includes at least one event; the causal relationship between events indicates that there is a causal relationship between one event and another event; the sequential relationship between behaviors indicates the chronological relationship between two behaviors.

[0013] The beneficial effects of the above technical solution are: clearly defining the association relationship between entities can assist in extracting the temporal relationship and spatial correlation of ship activity entities, and help assist in the extraction of ship activity entities.

[0014] Furthermore, before the semantic analysis of the acquired marine ship data in step 2), the step of converting the spatiotemporal point sequence of the trajectory in the acquired marine ship data into Stop / Move semantic information obtained by using the Stop / Move model is included, and the obtained Stop / Move semantic information is represented by a trajectory sequence point set T, and the trajectory sequence point set T includes a trajectory point set S for stop points and a trajectory segment set M for movement:

[0015] S={p i ,p i+1 ,p i+2 ,…,p i+m-1 ,p i+m}

[0016] M={<p start ,p end >}

[0017] Among them, p i Indicates the trajectory point whose spatial position basically does not change within a certain period of time; p start and p end Indicates the two dividing points of staying and moving.

[0018] The beneficial effects of the above technical solution are: using the Stop / Move model to obtain Stop / Move semantic information can mine the information and knowledge hidden in the spatiotemporal point series data of marine ship data; and after obtaining the Stop / Move semantic information, it is associated with geographic spatial elements to mine higher-order semantic knowledge.

[0019] Furthermore, the semantic analysis process in step 2) includes: directly extracting each entity according to the data field information describing the marine ship data according to the field information; and mapping the association relationship between the entities according to the association between the fields.

[0020] The beneficial effect of the above technical solution is that entities and the relationships between entities can be directly extracted based on the data field information describing the marine ship data, so that entities and the relationships between entities can be quickly and accurately determined.

[0021] Furthermore, the semantic analysis processing in step 2) includes: a) using the following AE algorithm to extract behaviors in the subject from the Stop / Move semantic information, wherein the behaviors include anchoring behaviors, anchoring behaviors and moving behaviors, and the moving behaviors include acceleration behaviors, deceleration behaviors and uniform speed behaviors. The extraction and determination process includes: calculating the distance between each trajectory point in the stop point trajectory point set S and the port location, identifying the trajectory points below the set distance threshold as stop points, and determining the behaviors at the stop points based on the stop points, wherein the behaviors at the stop points include anchoring behaviors and anchoring behaviors; calculating the acceleration of each trajectory segment in the moving trajectory segment set M, and determining the movement of the ship according to the positive and negative values ​​of the acceleration. b) according to the behaviors in the subject extracted in step a), the following EE algorithm is used to extract events in the subject, the events include ship departure events and ship arrival events: anchoring behavior and acceleration behavior with a time interval lower than a set time threshold are determined as a ship departure event, and deceleration behavior and anchoring behavior with a time interval lower than a set time threshold are determined as a ship arrival event; c) according to the events in the subject extracted in step b), the following PE algorithm is used to extract processes in the subject, the processes include navigation processes: backward matching the ship departure event with a ship arrival event with the shortest time interval and with the same participants, and determining the ship departure event and the ship arrival event as a navigation process.

[0022] The beneficial effects of the above technical solution are: using the AE algorithm to extract the behaviors and events in the subject from the Stop / Move semantic information, the behaviors and events in the subject can be accurately extracted, the transformation from trajectory point sequence data to activity knowledge graph is completed, and the task of extracting high-order knowledge from low-order data information is completed.

[0023] Furthermore, the semantic analysis processing in step 2) includes: using a preset neural network model to extract events in the subject from the text describing the maritime ship data; wherein the input of the neural network model includes word embedding, position embedding and event type embedding, and the output of the neural network model is the score of the screened words.

[0024] The beneficial effects of the above technical solution are: using a preset neural network model to extract events in the subject from the text describing maritime ship data, it is possible to accurately extract behaviors and events in the subject, complete the transformation from trajectory point sequence data to activity knowledge graph, and complete the task of extracting high-order knowledge from low-order data information.

[0025] Furthermore, the neural network model is a DMCNN model.

[0026] The beneficial effect of the above technical solution is that the use of the DMCNN model can ensure the uniqueness of the sentence-level semantic features finally obtained.

[0027] Furthermore, the events include routine events and emergencies. The routine events include ship departure events, ship arrival events, ship berthing events and ship docking events. The emergencies include natural disaster events, maritime accidents and pirate events.

[0028] The beneficial effect of the above technical solution is that the method of the present invention can not only analyze routine events of ships, but also be used to analyze sudden events of ships, more completely express ship activities, and facilitate in-depth analysis of the causes of ship events. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 is a flow chart of the method for constructing a ship activity knowledge graph of the present invention;

[0030] Figure 2 It is a schematic diagram of the components of the ship activity of the present invention;

[0031] Figure 3 is a schematic diagram of a ship activity model (SAM) of the present invention;

[0032] Figure 4 is a sample diagram of attribute constraints in the SAM model of the present invention;

[0033] Figure 5-1 It is a schematic diagram of the ship's port of call in the geospatial topological relationship between the ship's trajectory and geographical elements;

[0034] Figure 5-2 It is a schematic diagram of the sea area where the ship is stranded in the geospatial topological relationship between the ship's trajectory and geographical elements;

[0035] Figure 5-3 It is a schematic diagram of the ship passing through the port in the geospatial topological relationship between the ship's trajectory and geographical elements;

[0036] Figure 5-4 It is a schematic diagram of the ship's entry into the sea area in the geospatial topological relationship between the ship's trajectory and geographical elements;

[0037] Figure 5-5 It is a schematic diagram of the ship's departure from the sea area in the geospatial topological relationship between the ship's trajectory and geographical elements;

[0038] Figure 6 It is a schematic diagram of filling an example of a ship process based on the method of the present invention;

[0039] Figure 7 A schematic diagram of filling ship events and ship behavior examples based on the method of the present invention;

[0040] Figure 8 It is a flow chart of the ship activity knowledge graph application system development of the present invention;

[0041] Fig. 9 It is an overview diagram of the ship activity knowledge graph application system of the present invention;

[0042] Fig.10 It is a conventional ship activity query result diagram of the present invention;

[0043] Fig.11 It is a sudden ship incident query result diagram of the present invention;

[0044] Fig.12 It is a spatiotemporal query result diagram of ship activities of the present invention;

[0045] Fig.13 It is a time-space retrospective diagram of a single ship activity of the present invention;

[0046] Fig.14 It is a time-space retrospective diagram of multiple ship activities of the present invention. DETAILED DESCRIPTION

[0047] In order to monitor and analyze the activities of ships at sea, the present invention comprehensively utilizes multi-source data, and constructs a ship activity knowledge graph based on the extraction of the semantic information of the trajectory, providing an effective way to transform the low-knowledge-density trajectory spatiotemporal point sequence into high-order semantic knowledge. The specific process includes: firstly, analyzing the characteristics and components of ship activities, and designing the ship activity knowledge graph model layer with "process-event-behavior" as the core idea; then using the Stop / Move model and other methods to extract the trajectory semantic information, extract ship emergencies, and complete the instance layer filling.

[0048] A method for constructing a ship activity knowledge graph of the present invention is described in detail below in conjunction with the accompanying drawings and embodiments.

[0049] Method Example:

[0050] An embodiment of a method for constructing a ship activity knowledge graph of the present invention, the overall process is as follows Figure 1 As shown, the process is as follows:

[0051] Step 1: Analyze the characteristics and components of ship activities, take "process-event-behavior" as the core idea, and design a ship activity knowledge graph ontology model based on six entities: "process, event, participant, location, event and behavior". The specific process is as follows:

[0052] 1. Analyze the characteristics, classification and components of ship activities.

[0053] Compared with other types of activities, maritime ship activities have more distinct characteristics, such as dispersion, periodicity, predictability and proximity. By analyzing the composition of maritime activity elements and constructing the activity elements into a ship activity knowledge graph, the low-order feature information of ship activities can be converted into high-order semantic knowledge, which is helpful for ship maritime traffic monitoring, ship maritime incident analysis, etc.

[0054] (1) Characteristics of ship activities. Activities are the sum of behaviors that are united by a common purpose and complete certain social functions. Activities are composed of purpose, motivation, behavior and commonality, and have a complete structural system. The present invention defines ship activities as the sum of ship behaviors for a common purpose during the navigation process at sea, which is composed of activity purpose, ship events, ship behaviors and commonality. Ship activities have the following characteristics: ① Spatial dispersion: Since ship activities are not restricted by the transportation network, the scope and trajectory of activities will be freer than on land, so ship activities have spatial dispersion characteristics. ② Temporal periodicity: The time of ship activities is closely related to the marine environment. The periodicity of the marine climate will affect the time of ship activities and also have periodicity. ③ Behavior predictability: The size of the ship is large, so the ship activities will not jump, suddenly rotate, etc., and the ship activities usually have certain route restrictions, so the trajectory of the ship activities has a certain predictability. ④ Activity proximity: Due to the complex marine environment, reefs, sea ice, etc. are prone to cause marine accidents, so ships will choose safe routes for navigation, and the activities of ships on the same route will have proximity.

[0055] (2) Classification of ship activities. According to different classification methods, ship activities can be divided into different types. ① According to the nature of ship activities, ship activities can be divided into two categories: military ship activities and civil ship activities. Since this article mainly studies civil ship activities, the following will only classify civil ship activities again. ② According to the purpose of ship activities, ship activities can be divided into transportation activities, offshore engineering activities, offshore work activities, fishing activities, rescue activities, etc., among which transportation activities can be further divided into passenger activities and cargo activities; offshore engineering activities mainly refer to dredging, salvaging, lifting and resource mining activities; offshore work activities mainly refer to navigation, towing, clearing the way, supply and investigation activities. ③ According to the different sea areas of ship activities, ship activities can be divided into inland sea activities, territorial sea activities and high seas activities. Among them, inland sea activities refer to the activities of ships in the sea area within the demarcated territorial sea width baseline, territorial sea activities refer to the activities of ships in the sea area extending a certain width outward from the baseline, and high seas activities refer to the activities of ships in the sea area between the outer edge of one country's territorial sea and the outer edge of other countries' territorial sea. ④ According to the different activities of ships, ship activities can be divided into navigation activities, mooring activities, operating activities and accident avoidance activities. Navigation activities refer to the directional drifting movement of ships at sea, mooring activities refer to the behavior of ships when they are moored at a dock, anchored or tied to a buoy, operating activities refer to the behavior of ships when they are carrying out engineering operations and working at sea, and accident avoidance activities refer to the temporary and proactive behavior of ships after an unexpected accident occurs.

[0056] (3) Ship activity components. Based on the above-mentioned ship activity characteristics, general activity components cannot meet the needs of describing ship activities. Taking into account the needs of expressing the ship's navigation process and behavior patterns, the present invention proposes to express ship activities in three levels of progressive granularity based on the "process-event-behavior" concept: ① Process elements (Process), describing a ship's sea navigation process from the port to the next port, including the ship's departure place (Place of Departure), destination (Destination), departure time (Departure Time), arrival time (Arrival Time) and ship subject (Actor) and other information. ② Event elements (Event), describing routine events and emergencies that occur during the ship's activities, including information such as the subject (Actor) of the event, the start time (Time of Start) of the event, the end time (Time of End) of the event, the start location (Place of Start) of the event and the end location (Place of End) of the event. ③ Behavior elements (Action), describing the basic behaviors (such as movement and anchoring) during the ship's activities, including information such as the subject (Actor), time (Time) and place (Place) of the behavior.

[0057] The navigation process of a ship includes a series of ship behaviors, and the events that occur during the navigation process also include a series of ship behaviors. There is a causal relationship between ship events, and a sequential relationship between ship behaviors. Figure 2 shown.

[0058] 2. The knowledge graph ontology model obtained based on six entities: "process, event, participant, place, time and behavior".

[0059] The design of the knowledge graph model layer relies on the design of the ontology model. The SEM model is a general event ontology model, which is composed of core classes, type systems and property constraints. The core classes include: sem:Event (events that occurred), sem:Actor (participants), sem:Place (where), sem:Time (when). Each core class corresponds to its related type sem:Type, and is usually introduced from other dictionaries. For example, sem:PlaceType "capital" of sem:Place "Beijing" can be introduced from the Thesaurus of Geographical Names (TGN). Since the SEM model cannot fully express the ship process and behavior, it needs to be further expanded in combination with the ship activity elements. The present invention utilizes the type system and property constraint rules of the SEM model, adopts the ontology description language (Web ontology language, OWL) to supplement the concept system, entity class and entity relationship, and designs a ship activity ontology model (Ship Activity Model, SAM) including process-event-behavior.

[0060] (1) Define the SAM ontology concept system.

[0061] For maritime traffic monitoring, analyzing and mining different events in the same voyage process is of great significance for the correlation analysis and traceability analysis between events. The existing SEM model can associate ship events with the same ship based on the participants sem:Actor of the event, but it cannot determine whether they are events in the same trip. The new process entity class sam:Process in the present invention can meet this requirement.

[0062] The SAM model includes 6 core concepts: ①Process: refers to the process of a ship sailing from a port to the next port. ②Event: refers to the cause of the change in the state of a ship at sea, including natural disasters, maritime accidents and other accidents. ③Actor: refers to the subject participating in the event, that is, the ship. ④Place: refers to the place, which refers to an entity with a spatial location, such as a specific place name or longitude and latitude. ⑤Time: refers to time, which refers to an entity with time characteristics, such as a time point or a time period. ⑥Action: refers to the behavior, which refers to the more basic behavior of ship activities, such as anchoring, moving (uniform speed, acceleration and deceleration) and anchoring.

[0063] like Figure 3 As shown in the figure, the hollow arrows represent the rdfs:subClassOf (parent class) attribute, the dotted arrows represent the rdf:type (class) attribute, and the solid arrows represent other attributes. The types of maritime navigation processes in this paper include transportation (passenger and freight), fishing, cruising, escort, etc.; according to different methods of ship event extraction, this paper extracts ship events according to different ways of ship event extraction. Regular events are extracted based on AIS semantic information, and emergencies are extracted based on description text. Ship events are divided into regular events and emergencies. Regular events include departure, arrival, berthing and docking. Emergencies are further divided into natural disaster events, maritime accidents and pirate events. Natural disaster events include tsunamis, typhoons, sea ice, storm surges, waves and other maritime disaster events; maritime accidents include fires, shipwrecks, collisions and casualties. Ship behaviors are divided into moving (uniform speed, acceleration and deceleration), anchoring and anchoring according to the navigation status of the ship.

[0064] (2) Define the association relationship of the SAM ontology model.

[0065] Through the departure time and arrival time of the ship's voyage, the temporal relationship between all ship events and ship behaviors within this time period can be obtained. However, the simple temporal relationship often has no practical significance for the analysis of ship activities, so further semantic relationship expression is needed. The entity semantic relationship in the COVID-19 case activity knowledge graph includes composition, causality, follow-up and concurrent relationships. Drawing on the ideas of the above semantic relationships, this paper describes the entity association relationship of ship activities in three main types: one is the composition relationship between processes, events and behaviors; the second is the causal relationship between events; and the third is the sequential relationship between behaviors.

[0066] ①contain: containment relationship. sem:Event (ship event) contains a series of sam:Action (ship behavior), for example, the ship event ex:departure (departure) contains the ship behaviors ex:weighAnchor (weigh anchor) and ex:accelerate (accelerate), and this relationship is represented by sam:hasAction in this article; sam:Process (navigation process) contains a series of sem:Event (ship events), for example, a ship navigation process ex:transport (transport) contains the ship events ex:departure (departure) and ex:arrival (arrival), and this relationship is represented by sam:hasEvent in this article. ②cause: causal relationship, that is, a certain ship event leads to the occurrence of another ship event. For example, there is a causal relationship between the ship events ex:collision (collision) and ex:wreck (sinking) of the Titanic colliding with the iceberg, and this relationship is represented by sam:cause in this embodiment. ③followed: sequential relationship, that is, there is a sequence relationship between two ship behaviors that cannot be reversed. For example, the ship behavior ex:weighAnchor (weighing anchor) is completed before ex:accelerate (accelerating), which is represented by sam:followed. The above relationship is shown in Table 1.

[0067] Table 1 Semantic associations of ship activities

[0068]

[0069] (3) Define the attribute constraints of the SAM ontology model.

[0070] SAM model attributes include three types: core entity attributes, type attributes, and some other attributes. The core entity attributes include sam:processProperty, sem:eventProperty, and sam:actionProperty; the type attribute includes sem:typeProperty; and other attributes include the sub-attributes of sem:accordingTo and sem:hasTimeStemp. The core entity attributes relate the core entities to each other, and the type attributes relate each core entity to sem:type, and are represented by corresponding sub-attributes for easy query, such as sem:eventType. And sem:accordingTo is used to relate sem:View and sem:Authority to express opinions. sem:hasTimeStemp has 7 sub-properties: one represents the time point sem:hasTimeStemp, two represent the time interval sem:hasBeginTimeStemp and sem:hasEndTimeStemp, and four represent uncertain time intervals sem:hasEarliestBeginTimeStamp, sem:hasLatestBeginTimeStemp, sem:hasEarliestEndTimeStamp and sem:hasLatestEndTimeStamp.

[0071] The attribute constraints of the SAM model inherit most of the rules of the SEM model. The role of the entity is defined through the role sem:Role, such as Actor ("pirate ship") (pirate ship). sem:Temporary defines types with external dependencies. For example, the capital ship in the escort formation is defined only when there is an escort formation. The definition of the role sem:Role does not depend on external conditions, but depends on the relationship between entities. sem:View defines the viewpoint and is associated with sem:Authority through sem:accordingTo, such as Figure 4 Figure 2 shows an example of the above attribute constraints.

[0072] Step 2: Acquire maritime ship data, perform semantic analysis on the acquired maritime ship data, extract the entities of the maritime ship data, and determine the association relationship between the entities; based on the entities of the maritime ship data and the association relationship between the entities, fill in the constructed ship activity knowledge graph ontology model with instances to complete the construction of the ship knowledge activity graph.

[0073] Specific:

[0074] 1. Obtain marine vessel data.

[0075] The data sources are the ship automatic identification system (Automatic Identification System, AIS) data, the event data of the "Boiling Ring Trade" information system, and the global port information data. The present invention uses the trajectory semantic extraction algorithm and the spatial calculation method to extract entities and relationships (which will be introduced in the following paragraphs).

[0076] 2. Analyze the characteristics of maritime ship data and determine the data field information describing the maritime ship data.

[0077] AIS ship trajectory data uses the unique identification code MMSI to encode the ship, including the static data information of the ship and the dynamic trajectory information of the ship, a total of 30 fields, as shown in Table 2.

[0078] Table 2 AIS ship track data

[0079]

[0080] The event data of the "Boiling Huanmao" information system is the structured data of the maritime accident list and the piracy event list on the "Boiling Huanmao" website. The maritime accident list includes 10 fields, as shown in Table 3: accident, type, time, location, result, SOLAS status, involved ships, number of deaths / missing, number of injuries, accident details. The piracy event list includes 10 fields, as shown in Table 4: date, ship name, IMO, latitude, longitude, region, country, action, result, details.

[0081] Table 3 Marine accident data

[0082]

[0083] Table 4 Pirate incident data

[0084]

[0085] The global port information data includes 9 fields, as shown in Table 5: port code, port name, country, route, longitude, latitude, and introduction.

[0086] Table 5 Port data

[0087]

[0088] The above data have the following characteristics: (1) Time series. AIS trajectory data and event data with date attributes have obvious time series characteristics, and the time attributes and time series relationships of the corresponding entities can be extracted. (2) Spatial correlation. AIS trajectory data, event data and port data all have clear longitude and latitude information, and their spatial correlation can be extracted through spatial calculation. Therefore, the time series of the above data can assist in extracting the time series relationship of ship activity entities, and spatial correlation can assist in extracting ship activity entities.

[0089] 3. Based on the acquired maritime ship data, semantic extraction of ship activity trajectories is performed.

[0090] Since AIS trajectory data is organized in the form of a spatiotemporal point sequence, its semantic information is relatively weak. In order to mine the information and knowledge hidden in these spatiotemporal point sequence data, Spaccapietra et al. proposed a Stop / Move model. Using migratory bird trajectory data as the data source, the Stop / Move semantic information is extracted from the spatiotemporal point sequence data. The trajectory sequence point set is represented by T, the stop (stay point) is represented by the trajectory point set S, and the move (movement) is represented by the trajectory segment set M.

[0091] S={p i ,p i+1 ,p i+2 ,…,p i+m-1 ,p i+m} (1)

[0092] M={<p start ,p end >} (2)

[0093] In the formula, p i It indicates that the trajectory point whose spatial position does not change in a certain period of time. Only one trajectory point stays at the same position, so p i The trajectory points represented are discrete, S is a subset of T, and p start and p end Indicates the two dividing points of stop and move.

[0094] After converting the spatiotemporal point sequence data of the trajectory into the above-mentioned Stop / Move semantic information, and then associating it with geographic spatial elements, higher-order semantic knowledge can be mined. This paper uses the spatial topological relationship between the location point data of global ports and the surface data of sea area division and the Stop / Move trajectory point / segment set to extract the geographic correlation relationship between the two, that is, the extraction of the above-mentioned geographic correlation relationship can be extracted using the topological relationship between points, lines and surfaces, such as Figure 5-1 to Figure 5-5 shown.

[0095] 4. Extract ship activity entities and the relationships between entities.

[0096] The core entities defined in the ship activity pattern layer are mainly Process, Event, Action, Place, Time and Actor (participant, i.e. ship). The relationships between entities are mainly three semantic relationships: contain, cause and followed. The extraction of entities and their associations is mainly carried out through three methods: first, directly mapping the fields of the structured data table to entities and attributes, and mapping the relationship between entities according to the association between the fields; second, extracting using the semantic knowledge of the trajectory, combining the Stop / Move model, port location data and the speed value of the trajectory point, extracting the ship behavior entity, and calculating the time, location, etc. of the ship behavior; third, extracting event elements and associations from unstructured text data. The following is a detailed introduction.

[0097] (1) Direct extraction based on the fields of the data table. sem:Event (ship event) includes regular events and emergencies. The entity extraction of emergencies relies on the mapping of the accident field of the maritime accident data and the action field of the pirate data. sem:Time (time) is mapped from the time (time) and date (date) fields of the above two data. sem:Place (place) is mapped from the place (place) of the maritime accident data and the lat (latitude) and lon (longitude) of the pirate data. sem:Actor (participant) is mapped from the ship_name (ship name) field of the pirate data. However, the Actor of the maritime accident data needs to be entity extracted from the accident description text in the third method.

[0098] (2) Extract using the semantic knowledge of the trajectory. ① Design the AE (action extract) algorithm to extract the sam:Action (behavior) entity: First, calculate the distance between the stop set (S) and the port location (P). The stop point (s1) below the distance threshold (y1) can be considered as a stop point. There are two ship behaviors involved here, one is the anchoring behavior before docking at the port (A1), and the other is the anchoring behavior after docking (W1). The time of A1 inherits the start time of s1, and the time of W1 inherits the end time of s1. In addition, calculate the acceleration (a) of each point in the move segment, and judge the acceleration, deceleration and uniform speed sailing behavior of the ship (collectively referred to as moving behavior) according to the positive and negative values ​​of the acceleration. ② Design EE (event extract) algorithm to extract regular events in sem:Event: the anchoring behavior (W2) and acceleration behavior (Ac2) with a time interval lower than a certain time threshold (y2) together constitute a ship departure event (D2), the departure time inherits the start time of the anchoring behavior, and the departure place inherits the location of the anchoring behavior, that is, D2 = EE (W2, Ac2); similarly, the deceleration behavior (De2) and anchoring behavior (A2) with a time interval lower than a certain time threshold (y2) together constitute a ship arrival event (Ar2), the arrival time inherits the end time of the anchoring behavior, and the arrival place inherits the location of the anchoring behavior, that is, Ar2 = EE (De2, A2). ③ Design PE (process extract) algorithm to extract sam:Process (process) entity: for each ship departure event (D3), match it backward with a ship arrival event (Ar3) with the same Actor and the shortest time interval, then they together constitute a navigation process (Pr3). The departure time and place inherit the departure time and place of the departure event, and the arrival time and place inherit the end time and place of the arrival event.

[0099] (3) The DMCNN model is used to extract event elements and associations from the text describing maritime ship data. The input features of the DMCNN model are CWF (word embedding), PF (position embedding) and EF (event type embedding). The three features together express a word feature. The position embedding feature refers to the position feature of the word relative to the trigger word and the candidate argument, and the event type feature actually refers to the type of the trigger word. In this way, each word in the text has different features in different events. The convolution layer of the DMCNN model is consistent with the ordinary CNN. The difference is that when pooling, DMCNN performs segmented pooling, which uses the trigger word and the candidate argument as the segmentation point to ensure the uniqueness of the sentence-level semantic features finally obtained. The classification layer of the DMCNN model inputs the above word-level features and sentence-level features and outputs the score of the corresponding role. The DMCNN model is used to extract maritime emergencies from the detailed text of the maritime accident data. The result indicators are shown in Table 6.

[0100] Table 6 Event extraction indicators

[0101]

[0102]

[0103] The three entity semantic association relationships defined in the pattern layer are mainly extracted through the time attributes and description texts of the entities. ① Extraction of contain (containment relationship). The containment relationship between behavior and event and the containment relationship between event and process are identified through time and space attributes. For the same ship, if the start time and end time of the ship behavior are within the start time and end time range of a certain event of the ship, then it is considered that there is a containment relationship between the ship event and the ship behavior. The containment relationship between the ship process and the ship event is also determined by the containment relationship between the start time and the end time. ② Extraction of cause (causal relationship). The causal relationship mainly refers to the relationship between two sudden ship events. For ship sudden events belonging to the same ship process, the trigger words such as "cause", "because", and "caused by" are identified by natural language processing of the text describing the event, so as to establish the causal relationship between the sudden ship events. For example, "collision with iceberg leads to shipwreck" can establish a causal relationship between the ship event sam:collision and the ship event sam:wreck. ③ Extraction of followed (followed relationship). The sequential relationship mainly refers to the irreversible order relationship between two ship behaviors. The sequential relationship of ship behaviors belonging to the same ship event is used in combination with common sense of ship navigation to extract the sequential relationship of ship behaviors. For example, the ship behavior ex:weighAnchor (weighing anchor) is completed before ex:accelerate (accelerating), and there is a sequential relationship between the two.

[0104] 5. Based on the entities extracted in step 4 and the relationships between them, fill in the ship activity knowledge graph instances.

[0105] The example of ship activity knowledge graph is filled as follows: Figure 6 and Figure 7 shown. Figure 6 An example of a graph showing the entity association of a ship process (sam:Process): The ship transport process ex:transport-x1 of CHANGCHENG9HAO (Great Wall No. 9) departed from Yangpu Port at 15:49:01 on 2021-08-10, which includes ship events such as departure event ex:departure-x and arrival event ex:arrival-x.

[0106] The Great Wall No. 9 ship event mentioned above also includes some ship behaviors and event attributes. For example, ex:departure-x (departure event) includes two behaviors: accelerate ex:accelerate-x and weigh anchor ex:weigh anchor-x. There is a sequential relationship between these two behaviors, because only after weighing anchor can the acceleration behavior be performed. Figure 7 shown.

[0107] The method introduced above is experimentally verified below.

[0108] Using AIS data from 2020-01 to 2021-09 and "Boiling Ring Trade" maritime event data as ship activity data, and global port location data and global sea area division data as basic data, the semantic information of ship activities is mined, and entities such as ship behavior, ship events and ship processes and the relationships between entities are extracted to construct a ship activity knowledge graph, which is applied to the query and visualization of ship activities. This embodiment uses Protégé to build the ontology layer. The ship activity knowledge graph application development process is as follows: Figure 8 As shown in the figure, the postgresql database is used to manage the AIS data, python is used to mine the semantic information of the trajectory, and the entities and entity relationships are extracted, and the knowledge graph is stored in the Neo4j database. The react framework is used to develop the B / S front-end and back-end systems, which are mainly used for the overview, query and visualization applications of the graph.

[0109] The ship activity knowledge graph system developed based on the above environment describes the general navigation activities and emergency processes of ships with entities and relationships such as ship behavior, ship events, and ship processes. It can perform trajectory analysis, event process analysis, and ship navigation correlation analysis on ship activities. Fig. 9As shown: (a) is the query input box of the graph entity; (b) is the condition filter of the graph query; (c) displays the relevant graph relationship of the current query entity; (d) is the visualization of the trajectory of the ship entity on the map under the current query condition; (e) displays the attributes of the current query entity and the associated events.

[0110] The query requirements of the ship activity knowledge graph generally include queries on routine ship activities and queries on emergencies.

[0111] (1) Query of conventional ship activities. Conventional ship activity queries can be mainly queried by ship name, activity time, activity location and other conditions. This article uses Cypher statements to query in the backend. For example, querying a ship named CHANGCHENG9HAO (Great Wall 9) and its activity map in the South China Sea in the past month (2021-08-2021-09) involves a compound query of the three conditions of Actor, Time and Place. The query results are as follows: Fig.10 As shown, the 11 nodes in the figure are Fig.10 The straight line with three nodes in the middle column (the straight line is towards Fig.10 The left side of the vertical direction is tilted) as the symmetry axis, which is approximately symmetrical. The three nodes represent Actor, Process, and Place respectively. Among the remaining eight nodes, the top four nodes represent Action, the bottom two nodes represent Time, and the remaining two nodes represent Event. It can be seen from the graph that the Great Wall No. 10 ship departed from Yangpu Port at 2021-08-10 15:49:01. The departure event includes two behaviors: weigh anchor-10 and accelerate-10. It returned to Yangpu Port at 2021-08-10 15:49:01.

[0112] (2) Query of sudden ship incidents. Sudden ship incidents generally refer to natural disasters, maritime accidents and pirate incidents. For example, on July 25, 2020, a Panamanian-flagged bulk carrier Wakashio carrying 4,000 tons of fuel oil ran aground on a reef near Mauritius while sailing from China to Brazil. The bulk carrier was damaged by cracks in the hull. This was a ship grounding accident. When querying its activity map, the ship name Wakashio and the time of the incident 2020-07-25 were used as query conditions. The query results are as follows: Fig.11As shown in the figure, the 1 node in the first row represents Actor, the 2 nodes in the second row both represent Event, and the 3 nodes in the third row represent Time, Place, and Time respectively. As can be seen from the figure, Wakashio experienced event stranded-1 (stranded) near PORT LOUIS in Mauritius on 2020-07-25, and event crack-1 (broken) at 2020-08-15 16:30:00.

[0113] (3) Spatiotemporal query of ship activities. The first two queries are to query the activities of ships when the ship object is known. However, in reality, there are cases where the ship object is unknown and the ship activities are queried based on time and space information. For example: On 2021-09-12, near 1°24.72′N, 104°55.072′E, an oil tanker "Athenia" encountered pirates. Now it is necessary to query the ship activities in the surrounding waters during this period. In this case, the query conditions of the graph are time and space. The query results are as follows: Fig.12 As shown, the innermost node in the graph represents Time, the 9 nodes in the middle circle all represent Event, and the 9 nodes in the outermost circle all represent Actor. As can be seen from the graph, on 2021-09-12, near 1°23.996N, 104°55.978E, the query returns the first 10 ship entities, including "MING SING", "SUNLY 11", "DSI1081", "MARINA 2211", "PULAU MUNGGING", "PUTERI ZAMRUD", "AVENIR ADVANTAGE", "FERRAGAMO DOLPHIN", "DDKG2", and "ASL ABADI 1". The associated ship events include berthed (berthed-x) and the ship processes include transport (transports-x).

[0114] The present invention mainly relies on the visualization of the map query results and the corresponding trajectories to complete the spatiotemporal backtracking of ship activities. The visualization of the trajectory is mainly achieved by displaying the spatial distribution of the trajectory points on the map and connecting them in chronological order with arrows, so that users can view the distribution of the original trajectory to confirm the accuracy of the trajectory semantics expressed by the map. The system interface uses the line drawing method of trajectory points to draw the trajectories involved in the map query return results on the map, which is used to intuitively express the original real data.

[0115] (1) Temporal and spatial retrospection of single ship activities. Fig.13As shown: Query the activities of Great Wall 9 near Yangpu Port. It departed from Yangpu Port at 15:49:01 on 2021-08-10 and returned to Yangpu Port at 01:10:57 on 2021-09-06. The atlas matches the trajectory on the map.

[0116] (2) Spatiotemporal tracing of multiple ship activities. As introduced in the paragraphs about ship activity query, the results returned by the spatiotemporal query involve the trajectories of multiple ships, which are visualized on the map. The ship activities within 45 nautical miles around 1°24.72′N, 104°55.072′E on 2021-09-12 are queried, and the activity trajectories of the first 10 ships in the results returned by the map are plotted on the map, as shown in the figure below. Fig.14 This visualization method directly provides an intuitive and visual display of the query results, which not only verifies the accuracy of the graph, but also facilitates subsequent spatial analysis.

[0117] In summary, under the background of increasingly frequent marine ship activities and frequent marine accidents, constructing a ship activity knowledge graph is of great significance for the management and analysis of marine ship activities. This invention analyzes the components of ship activities, proposes a SAM ship activity ontology model based on the SEM model and the idea of ​​"process-event-behavior", and expresses semantic relationships such as inclusion, causality and sequence, thereby expressing the routine activities and emergencies of ships; using AIS data, event data of the "Boiling Ring Trade" information system, and global port information data, using trajectory semantic extraction algorithms and spatial calculation methods to extract entities and relationships, and using the spatiotemporal attribute characteristics of entities to extract semantic associations, and then constructing the data layer of the ship activity knowledge graph; finally, based on Neo4j, node.js, and react.js, a ship activity knowledge graph application system is developed, and the query of the graph is realized using Cypher statements, and the query results and trajectories are visualized, and used for the spatiotemporal backtracking of ship activities. The main contributions of this invention are as follows: (1) Based on the idea of ​​"process-event-behavior", two entities, process and action, are added to the SEM model, and the SAM ship activity ontology model is proposed, which more completely expresses the ship's activities at sea from ship behavior to navigation process. (2) By constructing a ship activity knowledge graph, the transformation from trajectory point sequence data to activity knowledge graph is realized, completing the task of extracting high-order knowledge from low-order data information.

Claims

1. A method for constructing a ship activity knowledge graph, characterized in that: The steps include: 1) Based on the six entities of process, event, participant, place, time and behavior, a ship activity knowledge graph ontology model is constructed; wherein the process represents a sea voyage process of a ship from the port trigger to the next port, the participant represents the ship, the place represents an entity with a spatial location, the time represents an entity with a time feature, the behavior represents the basic behavior of the ship activity, and the event represents the reason for the change of the state of the ship at sea; 2) Obtain maritime ship data, convert the spatiotemporal point sequence of the trajectory in the acquired maritime ship data into Stop / Move semantic information obtained by using the Stop / Move model, and then perform semantic analysis and processing to extract the entities of the maritime ship data and determine the association relationship between the entities; according to the entities of the maritime ship data and the association relationship between the entities, fill the constructed ship activity knowledge graph ontology model with instances to complete the construction of the ship knowledge activity graph; The association relationships between the entities include: an inclusion relationship between events and behaviors, an inclusion relationship between processes and events, a causal relationship between events, and a sequential relationship between behaviors; the inclusion relationship between events and behaviors indicates that an event contains at least one behavior; the inclusion relationship between processes and events indicates that a process contains at least one event; the causal relationship between events indicates that there is a causal relationship between one event and another event; the sequential relationship between behaviors indicates the chronological relationship between two behaviors.

2. The method for constructing a ship activity knowledge graph according to claim 1, characterized in that: The obtained Stop / Move semantic information is represented by a trajectory sequence point set T, which includes a trajectory point set S for stop points and a trajectory segment set M for movement: S={p i ,p i+1 ,p i+2 ,…,p i+m-1 ,p i+m } M={<p start ,p end >} Among them, p i Indicates the trajectory point whose spatial position does not change substantially within a certain period of time; p start and p end Indicates the two dividing points of staying and moving.

3. The method for constructing a ship activity knowledge graph according to claim 1, characterized in that: The semantic analysis process in step 2) includes: According to the data field information describing the marine ship data, each entity is directly extracted according to the field information; Map the associations between entities based on the associations between fields.

4. The method for constructing a ship activity knowledge graph according to claim 2, characterized in that: The semantic analysis process in step 2) includes: a) The following AE algorithm is used to extract the behaviors in the subject from the Stop / Move semantic information, wherein the behaviors include anchoring behavior, anchoring behavior and moving behavior, and the moving behavior includes acceleration behavior, deceleration behavior and uniform speed behavior: the distance between each track point in the stop point track point set S and the port location is calculated, and the track point below the set distance threshold is identified as the docking point, and the behavior at the docking point is determined according to the docking point, and the behavior at the docking point includes anchoring behavior and anchoring behavior; the acceleration of each track segment in the moving track segment set M is calculated, and the moving behavior of the ship is determined according to the positive and negative values ​​of the acceleration; b) according to the behaviors in the subject extracted in step a), extracting events in the subject using the following EE algorithm, wherein the events include ship departure events and ship arrival events: determining an anchoring behavior and an acceleration behavior whose time interval is less than a set time threshold as a ship departure event, and determining a deceleration behavior and an anchoring behavior whose time interval is less than a set time threshold as a ship arrival event; c) According to the events in the subject extracted in step b), the following PE algorithm is used to extract the process in the subject, wherein the process includes a navigation process: backward matching the ship departure event with a ship arrival event with the shortest time interval and the same participants, and determining the ship departure event and the ship arrival event as a navigation process.

5. The method for constructing a ship activity knowledge graph according to claim 1, characterized in that: The semantic analysis process in step 2) includes: A preset neural network model is used to extract events in the subject from text describing maritime ship data; wherein the input of the neural network model includes word embedding, position embedding and event type embedding, and the output of the neural network model is the score of the screened words.

6. The method for constructing a ship activity knowledge graph according to claim 5, characterized in that: The neural network model is a DMCNN model.

7. The method for constructing a ship activity knowledge graph according to claim 1, characterized in that: The events include routine events and emergencies. The routine events include ship departure events, ship arrival events, ship berthing events and ship docking events. The emergencies include natural disaster events, maritime accidents and pirate events.

Images