Evolution method and device of marine environment situation, electronic equipment and program product

Abstract

The invention discloses a marine environment situation evolution method and device, electronic equipment and a program product, and relates to the technical field of marine environment situation evolution.The evolution method comprises the steps that a current marine entity information set is determined, a preset prediction output model is adopted to process the current marine entity information set, and a feature scene graph is obtained; and receiving a target ocean task scene, determining all task entity information involved in the target ocean task scene, determining a sub-graph of each piece of task entity information based on the feature scene graph, obtaining a situation rule table of the target ocean task scene, and evolving the target ocean task scene based on the situation rule table. The technical problem that the situation of the ocean task scene cannot be dynamically evolved accurately in the prior art is solved.

Description

Technical Field

[0001] The present invention relates to the technical field of marine environment situation evolution, and in particular to a marine environment situation evolution method and apparatus, electronic equipment, and program product thereof. Background Art

[0002] According to cognitive psychology, visual graphics are far easier to understand than digital charts. Therefore, constructing a "situation map" that reflects changes in the ocean environment within a local mission area can maximize the user's understanding of maritime conditions. The ocean environment situation map is a situational system that provides information support for mission operations in the ocean. It primarily describes ocean area detection information from ocean sensors, including marine environmental elements and phenomena. Based on this information, it provides information on the relationship between the ocean environment and the equipment performing the mission, as well as perception information and the equipment's impact on the environment. This supports the collection and three-dimensional visualization of large-scale ocean scene information.

[0003] Marine environment situation representation based on marine platform software is currently the mainstream tool for studying marine situation. Three-dimensional visualization technology allows users to navigate and analyze marine areas on the platform, providing a wealth of marine environmental data and situational information compared to traditional graphical methods. However, in practical applications, these technologies, which require specific operational planning and marine work based on these platforms, often face problems such as excessive invalid information, insufficient expertise, an inability to cope with dynamic environmental changes, and an inability to comprehensively analyze the impact of multiple coupled factors. Three-dimensional visualization technology platforms often lack a focus on representing marine environmental elements and phenomena. Without relevant marine knowledge, users struggle to effectively utilize various element views to comprehensively plan tasks and mitigate risks. Furthermore, 3D visualization of the marine environment is limited by the real-time nature and volume of marine environmental data, making it difficult to conduct dynamic evolution analysis based on a specific scenario.

[0004] To address the above-mentioned problems, no effective solutions have been proposed so far. Summary of the Invention

[0005] The embodiments of the present invention provide a method for evolving the marine environment situation and its device, electronic equipment, and program product, so as to at least solve the technical problem in related technologies that the situation of marine mission scenarios cannot be accurately and dynamically evolved.

[0006] According to one aspect of an embodiment of the present invention, a method for evolving an ocean environment situation is provided, comprising: determining a current ocean entity information set, wherein the current ocean entity information set includes ocean environment element entity information, ocean environment phenomenon entity information, and ocean equipment entity information; processing the current ocean entity information set using a preset prediction output model to obtain a feature scene graph; receiving a target ocean mission scene, and determining all task entity information involved in the target ocean mission scene; based on the feature scene graph, determining a subgraph of each task entity information, obtaining a situation rule table for the target ocean mission scene, and evolving the target ocean mission scene based on the situation rule table.

[0007] Furthermore, before adopting the preset prediction output model to process the current ocean entity information set, it also includes: determining the initial prediction output model; constructing an ocean directed graph and ocean labels based on the ocean knowledge graph and the ocean equipment rule library; determining the ocean prediction labels based on the ocean directed graph and the initial prediction output model; training the initial prediction output model based on the ocean labels and the ocean prediction labels to obtain the preset prediction output model.

[0008] Furthermore, based on the ocean knowledge graph and the ocean equipment rule base, the steps of constructing an ocean directed graph and an ocean label include: extracting ocean entity information and the ocean association relationship between every two ocean entity information from the ocean knowledge graph, wherein the ocean entity information includes ocean environment element entity information and ocean environment phenomenon entity information; constructing a first directed graph based on the ocean entity information and the ocean association relationship; extracting ocean equipment entity information and equipment association relationship from the ocean equipment rule base, wherein the equipment association relationship is the association relationship between ocean equipment entity information and another ocean equipment entity information, ocean environment element entity information or ocean environment phenomenon entity information; constructing a second directed graph based on the ocean equipment entity information and the equipment association relationship; based on the first directed graph and the second directed graph, overlapping all repeated nodes to obtain an ocean directed graph, and extracting all edge relationships in the ocean directed graph to obtain an ocean label, wherein the node is used to represent ocean environment element entity information, ocean environment phenomenon entity information or ocean equipment entity information.

[0009] Furthermore, based on the ocean directed graph and the initial prediction output model, the step of determining the ocean prediction label includes: using the initial prediction output model to extract the node static features of each node on the ocean directed graph; calculating the similarity between the static features of each two nodes; determining that the two nodes indicated by the similarity greater than a preset threshold have an edge relationship, and obtaining the ocean prediction label based on all edge relationships.

[0010] Furthermore, a preset prediction output model is used to process the current ocean entity information set to obtain a feature scene graph, including: using the preset prediction output model to extract the entity static features of each entity information in the current ocean entity information set, and based on the similarity between each two entity static features, determining the edge relationship between each two entity static features; constructing the current directed graph based on all entity information and edge relationships; extracting the edge relationship context dynamic features and node context dynamic features of the current directed graph based on the preset prediction output model and the long short-term memory network model; and constructing the feature scene graph based on the edge relationship context dynamic features and the node context dynamic features.

[0011] Furthermore, the steps of extracting the edge relationship context dynamic features and node context dynamic features of the current directed graph based on the preset prediction output model and the long short-term memory network model include: using the preset prediction output model to extract the node static features of the current directed graph, and based on the node static features, determining the edge relationship features between entity information; based on the ocean knowledge graph, determining the dynamic attributes of each ocean environmental phenomenon entity information; using the long short-term memory network model to encode the dynamic attributes into the node static features to obtain the node dynamic features; determining the neighborhood of the node represented by each entity information; based on the edge relationship features, node dynamic features and neighborhood, incorporating time-varying information into the edge relationship, and encoding the context information into the node and edge relationship to obtain the edge relationship context dynamic features and the node context dynamic features.

[0012] Furthermore, based on the feature scene graph, the step of determining the subgraph of each task entity information includes: determining the neighborhood of the node represented by each task entity information; for each task entity information, obtaining the node context dynamic features of the node represented by the task entity information, and obtaining the edge relationship context dynamic features between the node represented by the task entity information and each node in the neighborhood; determining the sum of the moduli of the edge relationship context dynamic features of the node represented by the task entity information in the neighborhood, wherein the sum of the moduli of the edge relationship context dynamic features is determined based on the moduli of all edge relationship context dynamic features; for each node in the neighborhood, determining the connection probability between the node represented by the task entity information and the node in the neighborhood based on the moduli of the edge relationship context dynamic features between the node represented by the task entity information and the node in the neighborhood and the sum of the moduli of the edge relationship context dynamic features; based on all connection probabilities, determining the subgraph of the task entity information.

[0013] Furthermore, the step of evolving the target ocean mission scenario based on the situation rule table includes: animating each subgraph in the situation rule table based on time in the situation window, and generating a situation report for each subgraph.

[0014] According to another aspect of an embodiment of the present invention, an evolution device for an ocean environment situation is also provided, including: a first determination unit for determining a current ocean entity information set; a processing unit for processing the current ocean entity information set using a preset prediction output model to obtain a feature scene graph; a second determination unit for receiving a target ocean mission scene and determining all task entity information involved in the target ocean mission scene; a third determination unit for determining a subgraph of each task entity information based on the feature scene graph, obtaining a situation rule table for the target ocean mission scene, and evolving the target ocean mission scene based on the situation rule table.

[0015] Furthermore, the evolution device also includes: a first determination module, used to determine the initial prediction output model before using the preset prediction output model to process the current ocean entity information set; a first construction module, used to construct an ocean directed graph and ocean labels based on the ocean knowledge graph and the ocean equipment rule base; a second determination module, used to determine the ocean prediction label based on the ocean directed graph and the initial prediction output model; a first training module, used to train the initial prediction output model based on the ocean label and the ocean prediction label to obtain the preset prediction output model.

[0016] Furthermore, the first construction module includes: a first extraction submodule, used to extract ocean entity information and the ocean association relationship between every two ocean entity information from the ocean knowledge graph, wherein the ocean entity information includes ocean environment element entity information and ocean environment phenomenon entity information; a first construction submodule, used to construct a first directed graph based on the ocean entity information and the ocean association relationship; a second extraction submodule, used to extract ocean equipment entity information and equipment association relationship from the ocean equipment rule base, wherein the equipment association relationship is the association relationship between ocean equipment entity information and another ocean equipment entity information, ocean environment element entity information or ocean environment phenomenon entity information; a second construction submodule, used to construct a second directed graph based on the ocean equipment entity information and the equipment association relationship; a first overlapping submodule, used to overlap all repeated nodes based on the first directed graph and the second directed graph to obtain an ocean directed graph, and extract all edge relationships in the ocean directed graph to obtain ocean labels, wherein the nodes are used to represent ocean environment element entity information, ocean environment phenomenon entity information or ocean equipment entity information.

[0017] Furthermore, the second determination module includes: a third extraction submodule, which is used to extract the node static features of each node on the ocean directed graph using the initial prediction output model; a first calculation submodule, which is used to calculate the similarity between the static features of each two nodes; and a first determination submodule, which is used to determine that there is an edge relationship between two nodes indicated by a similarity greater than a preset threshold, and obtain an ocean prediction label based on all edge relationships.

[0018] Furthermore, the processing unit includes: a first extraction module, which is used to extract the entity static features of each entity information in the current ocean entity information set using a preset prediction output model, and determine the edge relationship between each two entity static features based on the similarity between each two entity static features; a second construction module, which is used to construct the current directed graph based on all entity information and edge relationships; a second extraction module, which is used to extract the edge relationship context dynamic features and node context dynamic features of the current directed graph based on the preset prediction output model and the long short-term memory network model; a third construction module, which is used to construct a feature scene graph based on the edge relationship context dynamic features and the node context dynamic features.

[0019] Furthermore, the second extraction module includes: a second determination submodule, which is used to extract the static features of the nodes of the current directed graph using a preset prediction output model, and determine the edge relationship features between the entity information based on the node static features; a third determination submodule, which is used to determine the dynamic attributes of each marine environmental phenomenon entity information based on the marine knowledge graph; a first encoding submodule, which is used to encode the dynamic attributes into the node static features using a long short-term memory network model to obtain the node dynamic features; a fourth determination submodule, which is used to determine the neighborhood of the node represented by each entity information; a second encoding submodule, which is used to encode time-varying information into the edge relationship based on the edge relationship features, the node dynamic features and the neighborhood, and encode the context information into the nodes and edge relationships to obtain the edge relationship context dynamic features and the node context dynamic features.

[0020] Furthermore, the third determination unit includes: a third determination module, used to determine the neighborhood of the node represented by each task entity information; a first acquisition module, used to obtain the node context dynamic features of the node represented by the task entity information for each task entity information, and obtain the edge relationship context dynamic features between the node represented by the task entity information and each node in the neighborhood; a fourth determination module, used to determine the sum of the moduli of the edge relationship context dynamic features of the node represented by the task entity information in the neighborhood, wherein the sum of the moduli of the edge relationship context dynamic features is determined based on the moduli of all edge relationship context dynamic features; a fifth determination module, used to determine, for each node in the neighborhood, the connection probability between the node represented by the task entity information and the node in the neighborhood based on the moduli of the edge relationship context dynamic features between the node represented by the task entity information and the node in the neighborhood and the sum of the moduli of the edge relationship context dynamic features; a sixth determination module, used to determine the subgraph of the task entity information based on all connection probabilities.

[0021] Furthermore, the third determining unit further includes: a first demonstration module, configured to perform animation demonstration on each subgraph in the situation rule table based on time in the situation window, and generate a situation report for each subgraph.

[0022] According to another aspect of an embodiment of the present invention, a computer program product is also provided, including a non-volatile computer-readable storage medium, the non-volatile computer-readable storage medium storing a computer program, and when the computer program is executed by a processor, implementing any of the above-mentioned methods for evolving the marine environment situation.

[0023] According to another aspect of an embodiment of the present invention, an electronic device is also provided, comprising one or more processors and a memory, wherein the memory is used to store one or more programs, wherein when the one or more programs are executed by one or more processors, the one or more processors implement any one of the above-mentioned methods for evolving the marine environment situation.

[0024] In the present invention, the current ocean entity information set is determined, and a preset prediction output model is used to process the current ocean entity information set to obtain a characteristic scene graph, the target ocean mission scene is received, and all task entity information involved in the target ocean mission scene is determined. Based on the characteristic scene graph, a subgraph of each task entity information is determined, and a situation rule table of the target ocean mission scene is obtained. The target ocean mission scene is evolved based on the situation rule table, thereby solving the technical problem in related technologies that the situation of the ocean mission scene cannot be accurately and dynamically evolved.

[0025] In the present invention, the preset prediction output model obtained in advance is used to process the current ocean entity information set, so that the ocean element-phenomenon-equipment scene graph can be obtained. Then, by further processing the ocean element-phenomenon-equipment scene graph, a feature scene graph can be obtained. Using the feature scene graph, a subgraph of each entity in the ocean mission scene can be accurately and efficiently obtained, so that the ocean mission scene can be evolved according to the subgraph, and the ocean environment situation of concern to the ocean mission scene can be effectively predicted and analyzed, thereby achieving the technical effect of accurately and dynamically evolving the ocean environment situation. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

[0027] Figure 1 is a flow chart of an optional method for evolving ocean environment situation according to an embodiment of the present invention;

[0028] Figure 2 is a schematic diagram of an optional directed graph prediction model training and directed graph automatic expansion according to an embodiment of the present invention;

[0029] Figure 3This is a schematic diagram of an optional method of introducing dynamic time information into node features based on a long short-term memory network (LSTM) according to an embodiment of the present invention;

[0030] Figure 4 is a schematic diagram of an optional directed graph neighborhood aggregation and computation graph according to an embodiment of the present invention;

[0031] Figure 5 is a schematic diagram of generating an optional ocean environment feature scene graph according to an embodiment of the present invention;

[0032] Figure 6 is a schematic diagram of generating an optional ocean environment feature scene graph according to an embodiment of the present invention;

[0033] Figure 7 is a schematic diagram of a dynamic demonstration effect of an optional dynamic ontology relationship according to an embodiment of the present invention;

[0034] Figure 8 is a schematic diagram of generating an optional ocean environment feature scene graph according to an embodiment of the present invention;

[0035] Figure 9 is a schematic diagram of a dynamic demonstration effect of an optional sub-graph according to an embodiment of the present invention;

[0036] Figure 10 is a schematic diagram of an optional ocean environment situation map representation and evolution analysis based on a scene graph according to an embodiment of the present invention;

[0037] Figure 11 is a schematic diagram of an optional device for evolving ocean environment situation according to an embodiment of the present invention;

[0038] Figure 12 This is a hardware structure block diagram of an electronic device (or mobile device) for an evolution method of ocean environment situation according to an embodiment of the present invention. DETAILED DESCRIPTION

[0039] In order to enable those skilled in the art to better understand the solutions of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of the present invention.

[0040] It should be noted that the terms "first", "second", etc. in the description and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that the numbers used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0041] It should be noted that the relevant information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) collected and involved in the present invention are all information and data authorized by the user or fully authorized by all parties, and the collection, storage, use, processing, transmission, provision, disclosure and application of the relevant data comply with the relevant laws, regulations and standards of the relevant regions, take necessary confidentiality measures, do not violate public order and good morals, and provide corresponding operation entrances for users to choose to authorize or refuse. For example, an interface is set up between this system and the relevant users or institutions. Before obtaining relevant information, it is necessary to send an acquisition request to the aforementioned user or institution through the interface, and obtain relevant information after receiving the consent information fed back by the aforementioned user or institution.

[0042] The present invention proposes a knowledge-driven dynamic construction technology for marine environment scene graphs, which can dynamically construct marine environment scene graphs based on knowledge-driven, and automatically realize the adaptive construction and dynamic expansion of new nodes for new data and new knowledge through set rules. It not only provides interpretability and reliability for the scene graph in a knowledge-driven form, but also maintains a certain robustness in the face of insufficient relevant data intelligence or complex situation conditions in specific tasks. It helps to construct a marine environment scene graph with dynamic expansion and adaptive update of edge relationships for situation display based on marine tasks, and provides a basis for dynamic evolution analysis for the display of scene situation, thereby providing professional marine knowledge and internal logic for situation display and task decision-making.

[0043] In addition, the present invention also proposes a feature extraction-based ocean environment feature scene graph generation technology, which obtains a feature scene graph with the ability to perceive context nodes and time through feature extraction, relationship context coding and long-term context coding. The unstructured ocean environment scene graph is converted into a quantitative, structured and computable feature scene graph through feature representation, providing scene situation quantification, storage and review capabilities for situation display, helping to provide a calculation-based decision basis for decision-making applications related to ocean missions, supporting the dynamic evolution analysis of ocean environment situations and ocean environment situation interpretation based on time series relationships, thereby providing quantitative computational support for specific applications of ocean missions.

[0044] In addition, the present invention also proposes a situation information display technology based on supporting dynamic evolution analysis. Based on the feature scene graph nodes and edge relationships with dynamic weights, according to specific application requirements, it supports a certain type of ocean phenomenon or a certain equipment that performs a specific task as the ontology, constructs ontology dynamic relationships from the feature scene graph, and then realizes adaptive updates of ontology relationships based on the time-varying characteristics of the feature scene graph. At the same time, it also supports the dynamic display of the sea area task situation map and its evolution analysis results in the visualization interface through subgraph extraction based on the relevant concept nodes involved in a certain type of specific task scenario, providing adaptive updates, time-series-based dynamic evolution analysis, situation information pros and cons judgment and other services for specific ocean-oriented business applications, helping to provide dynamic and easy-to-read situation visualization graphics for the display of marine environmental situation information, supporting marine environmental situation-related decisions for marine applications, and thus providing visual information guidance and browsing for specific marine task applications.

[0045] The present invention will be described in detail below with reference to various embodiments.

[0046] Example 1

[0047] According to an embodiment of the present invention, an embodiment of a method for evolving marine environmental conditions is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0048] Figure 1 is a flow chart of an optional method for evolving the marine environment situation according to an embodiment of the present invention, such as Figure 1 As shown, the method includes the following steps:

[0049] Step S101 : determining a current ocean entity information set, wherein the current ocean entity information set includes: ocean environment element entity information, ocean environment phenomenon entity information, and ocean equipment entity information.

[0050] In an embodiment of the present invention, the ocean environment is dynamic, and marine equipment is constantly updated over time. Therefore, the ocean knowledge graph and the marine equipment rule base can be updated every period of time (for example, one year) to obtain the current ocean entity information set. The current ocean entity information set refers to the collection of all identifiable entity information in the ocean within a specific time period, including marine environment element entity information, marine environment phenomenon entity information, and marine equipment entity information. This entity information covers various aspects of the ocean environment, such as temperature, salinity, flow rate (ocean environment elements), waves, vortices (ocean environment phenomena), deep-sea detectors, marine platforms (ocean equipment), etc.

[0051] In this embodiment, marine environmental element entity information relates to fixed, measurable elements in the marine environment, such as temperature, pressure, and salinity. Marine environmental phenomenon entity information relates to dynamic phenomena occurring in the ocean, such as ocean currents, storms, and waves. Marine equipment entity information relates to equipment operating in the marine environment, such as ships, submarines, and ocean observation buoys.

[0052] Step S102: Using a preset prediction output model to process the current ocean entity information set to obtain a feature scene graph.

[0053] In an embodiment of the present invention, a preset prediction output model is constructed based on an ocean knowledge graph and a marine equipment rule base. It can automatically identify and construct association relationships, perception relationships, and influence relationships between entities (marine environmental elements, marine environmental phenomena, and marine mission equipment), and represent these relationships in the form of a directed graph. The model can be trained and predicted using deep learning technologies such as multi-level perceptron (MLP) networks and long short-term memory networks (LSTM).

[0054] In an embodiment of the present invention, feature extraction can be performed on each entity information in the current ocean entity information set through a preset prediction output model to calculate the similarity between two entities through features, thereby constructing a directed graph, and then forming a feature scene graph through the edge relationship context dynamic features and node context dynamic features of the directed graph.

[0055] In this embodiment, the feature scene graph is a directed graph that characterizes the relationships between marine environmental elements, phenomena, and marine equipment. It combines static and dynamic time-varying features, reflecting the dynamic changes and interactions of marine entities at a specific moment or time period, and serves as the foundation for subsequent evolutionary analysis.

[0056] Step S103: receiving a target ocean mission scenario and determining all mission entity information involved in the target ocean mission scenario.

[0057] In the embodiment of the present invention, the target marine mission scenario refers to a specific marine activity or mission scenario to be analyzed, such as deep-sea scientific research, marine resource exploitation, marine rescue, etc. Each mission scenario has its own specific entity information requirements and analysis focus.

[0058] In an embodiment of the present invention, the target ocean mission scenario can be decomposed and all entities involved (i.e., mission entity information) can be listed to form a mission entity table. Here, the mission entity information is the ocean entity information directly related to the mission, including but not limited to: ocean equipment, ocean environmental elements, and specific ocean environmental phenomena.

[0059] Step S104: Based on the feature scene graph, determine the subgraph of each task entity information, obtain the situation rule table of the target ocean task scene, and evolve the target ocean task scene based on the situation rule table.

[0060] In this embodiment of the present invention, a dynamic ontological relationship based on graph representation learning can be constructed for each entity with dynamic attributes based on the feature scene graph to obtain its subgraph. This subgraph is a local graph extracted from the feature scene graph, centered on a specific task entity, and contains all direct and indirect related information related to the entity. The extraction of the subgraph helps to focus the analysis and display of information related to the specific task entity.

[0061] In an embodiment of the present invention, a situation rule table of the target ocean mission scenario can be obtained based on a subgraph of all task entity information involved in the target ocean mission scenario, and then the target ocean mission scenario can be dynamically evolved based on time in the situation window based on the situation rule table.

[0062] In this embodiment, the dynamic evolution of the target ocean mission scenario includes tracking and predicting the dynamic changes of equipment, environmental elements and phenomena, and evaluating how these changes affect the mission progress.

[0063] In summary, the pre-trained preset prediction output model can be used to process the current set of ocean entity information, and the ocean element-phenomenon-equipment scene graph can be obtained. Then, by further processing the ocean element-phenomenon-equipment scene graph, a feature scene graph can be obtained. Using the feature scene graph, a subgraph of each entity in the ocean mission scene can be accurately and efficiently obtained, so that the ocean mission scene can be evolved according to the subgraph, and the ocean environment situation of concern to the ocean mission scene can be effectively predicted and analyzed, thereby achieving the technical effect of accurately and dynamically evolving the ocean environment situation, and thus solving the technical problem in related technologies that the situation of the ocean mission scene cannot be accurately and dynamically evolved.

[0064] In order to improve the accuracy of determining the preset prediction output model, in the evolution method of the marine environment situation provided in Example 1 of the present application, before using the preset prediction output model to process the current marine entity information set, an initial prediction output model is determined; based on the marine knowledge graph and the marine equipment rule library, an ocean directed graph and ocean labels are constructed; based on the ocean directed graph and the initial prediction output model, the ocean prediction labels are determined; based on the ocean labels and the ocean prediction labels, the initial prediction output model is trained to obtain the preset prediction output model.

[0065] In the embodiment of the present invention, it can be in the form of knowledge drive (including ocean knowledge graph K G , Marine Equipment Rule Library K R ), build the prediction output model F G , so as to provide the ability to automatically construct the association, perception relationship and impact information of marine environmental elements, marine environmental phenomena and marine mission equipment through this model.

[0066] In an embodiment of the present invention, an initial prediction output model can be determined first. The initial prediction output model can be designed as a multi-layer perceptron (MLP) network, which can extract entity features from input marine entities (marine environmental elements, marine environmental phenomena, and marine equipment).

[0067] In an embodiment of the present invention, entities (marine environment element entities and marine environment phenomenon entities) and their static attributes can be extracted from the marine knowledge graph, a triple relationship (subject-predicate-object) can be established, and a directed graph can be generated based on this. In a directed graph, entities serve as nodes, and the relationships between entities serve as directed edges. At the same time, equipment entities and their association rules are extracted from the marine equipment rule library and added to the directed graph to form a more complete association network between the marine environment and equipment (i.e., a marine directed graph). Afterwards, marine labels can be generated based on the edge relationships of the marine directed graph. The marine labels reflect the actual association relationship and degree of influence between entities, and are used as supervisory information for subsequent model training. Specifically, for each pair of entity nodes in the marine directed graph, by analyzing the triple relationship and equipment rules between the entities, it is determined whether there is an association relationship, and the relationship is labeled, such as "exists" or "does not exist."

[0068] In this embodiment of the present invention, entities in the ocean directed graph can be input into the initial prediction output model to predict the association or perception relationship between entities based on the features output by the model, thereby generating ocean prediction labels. These prediction labels are used for model training to guide the model in learning how to more accurately predict the relationship between entities.

[0069] In an embodiment of the present invention, during the model training phase, ocean labels can be used as supervisory information and compared with predicted ocean labels. The prediction error is calculated using a loss function (e.g., cross-entropy loss), and the model parameters are then adjusted using a backpropagation algorithm to minimize the prediction error. After multiple rounds of iterative training, the model will gradually learn the inherent laws of inter-entity relationships and form a preset prediction output model that can accurately predict inter-entity associations.

[0070] For example, by the formula Calculate the prediction error, where represents the prediction error, Y represents the true ocean label, Indicates the predicted ocean forecast label.

[0071] In this example, the automated prediction and construction of complex relationships between marine entities is achieved through the construction of an initial prediction model, directed graphs, labeling, and model training. The pre-set prediction output model automatically identifies and infers relationships between entities based on the marine knowledge graph and equipment rule base, improving the efficiency and accuracy of constructing marine environmental situation maps.

[0072] In order to improve the accuracy of constructing ocean directed graphs and ocean labels, in the evolution method of the ocean environment situation provided in Example 1 of the present application, ocean entity information and the ocean association relationship between every two ocean entity information are extracted from the ocean knowledge graph, wherein the ocean entity information includes ocean environment element entity information and ocean environment phenomenon entity information; based on the ocean entity information and the ocean association relationship, a first directed graph is constructed; ocean equipment entity information and equipment association relationship are extracted from the ocean equipment rule library, wherein the equipment association relationship is the association relationship between ocean equipment entity information and another ocean equipment entity information, ocean environment element entity information or ocean environment phenomenon entity information; based on the ocean equipment entity information and the equipment association relationship, a second directed graph is constructed; based on the first directed graph and the second directed graph, all repeated nodes are overlapped to obtain an ocean directed graph, and all edge relationships in the ocean directed graph are extracted to obtain ocean labels, wherein the nodes are used to represent ocean environment element entity information, ocean environment phenomenon entity information or ocean equipment entity information.

[0073] In the embodiment of the present invention, all the involved marine environment elements V can be screened from the marine knowledge graph. E 、Phenomenon Entity V P and its static attributes and other ocean entity information, including but not limited to temperature, salinity, current velocity, waves, vortices, etc. Then, all triple relationships between entities (i.e., a combination of entity-relationship-entity) are extracted to form ocean association relationships. The collected ocean entity information is then used as nodes, and the association relationship between each pair of entities is used as a directed edge to construct the first directed graph. In the first directed graph, nodes represent specific ocean entities, and edges represent static associations between entities, such as "influence", "association", etc. Among them, the formula for generating the first directed graph G = (V, E) is as follows:

[0074]

[0075] G=(V,E)=(V,RDF(Phrase));

[0076] Among them, V represents the ocean entity, V E and V P From the Ocean Knowledge Graph V G The marine environment element entity and marine environment phenomenon entity found in the G China-Israel V E and V P For all triples of subject and object, v s and v oBoth belong to the ocean entity V, representing the subject entity and the object entity respectively. Phrase is the relationship between ocean entities, that is, the predicate. The directed graph G is constructed based on triples. The node V is the subject and object of the triple, and the edge E of the directed graph is the predicate Phras of the triple. e .

[0077] In the embodiment of the present invention, in the marine equipment rule library K R Filter out all marine equipment entity information V A , and their equipment association relationships with other equipment, marine environmental elements or phenomena (E A , v o )(where E A Indicates the current equipment entity V A With other entities v o The relationship between v o Represents the marine environment element entity V E 、Marine Environmental Phenomenon Entity V P Or marine equipment entity V A ). Equipment association relationships may include "detection", "impact", "use", etc., reflecting the interaction between marine equipment and the environment when performing tasks.

[0078] In the embodiment of the present invention, the collected marine equipment entity information can be used as nodes and the equipment association relationship as directed edges to form a second directed graph. Unlike the first directed graph, the second directed graph focuses on the dynamic perception relationship and influence between equipment entities and environmental factors and phenomena, as well as the collaborative or competitive relationship between equipment. A It can be expressed as: Among them, V A , E A Respectively represent the marine equipment rule base K R Marine equipment entities and equipment association relationships found in.

[0079] In an embodiment of the present invention, the first directed graph and the second directed graph are overlapped, that is, all nodes in the two graphs are compared to find repeated nodes, and then overlap is performed based on the repeated nodes to obtain an ocean directed graph.

[0080] Afterwards, each edge in the ocean directed graph can be analyzed to determine its actual relationship, label the edge relationship, and generate ocean labels Y. These labels are used as supervision information in subsequent model training to guide the model to learn how to accurately predict the relationship between entities. The formula of ocean label Y is as follows: Y = Edge (G, G A ), Edge is the edge of the directed graph, that is, for the directed graph G and the directed graph G A, superimpose all repeated nodes to form all node-edge relationships of the final directed graph, which is the label Y.

[0081] In this embodiment, the comprehensive application of marine environment information and equipment rules is realized, a directed graph that can reflect the complex relationships of the marine environment is constructed, and guidance is provided for subsequent model training through the generation of marine labels.

[0082] In order to improve the accuracy of determining ocean prediction labels, in the evolution method of the ocean environment situation provided in Example 1 of the present application, an initial prediction output model is used to extract the node static features of each node on the ocean directed graph; the similarity between the static features of each two nodes is calculated; it is determined that there is an edge relationship between two nodes indicated by a similarity greater than a preset threshold, and based on all edge relationships, an ocean prediction label is obtained.

[0083] In the embodiment of the present invention, the entity nodes V of the directed graph (ie, the ocean directed graph) of the marine environment elements, phenomena and marine domain equipment are E 、V P and V A Perform classification and use the initial prediction output model (such as MLP network) to extract node static features v for entity nodes E 、v P and v A , according to v E 、v P and v A The similarity between them and the preset threshold α are used to predict the edge relationship of the ocean directed graph to generate the predicted output of the ocean directed graph (i.e., ocean prediction labels), and then the existing ocean labels Y can be combined for loss calculation and deep learning training. The specific formula is as follows:

[0084] F1(V E , V P , V A )=v E , v P , v A ;

[0085]

[0086] Among them, F1 is an MLP network. The first few layers are encoding layers, and the last fully connected layer is a decoding layer. E Phenomenon V P and equipment V A Each entity node is encoded as a feature vector (i.e. v E , v P , v A), and the modulus of the feature vector is 1; s(u, v) is the similarity between two nodes, u, v belongs to {E, P, A} (that is, element V E Phenomenon V P and equipment V A When the similarity is higher than the set threshold α, the prediction model F1 considers that there is an edge relationship between the two entity nodes u and v. When the similarity is lower than the set threshold, the edge relationship is considered to be non-existent. The predicted edge relationship and entity nodes are combined to form the predicted output of the directed graph. Then the directed graph label Y is combined for loss calculation and back propagation.

[0087] Specifically, each node (entity information) in the ocean directed graph can be input into the initial prediction output model (a pre-designed multi-layer perceptron network). The model encodes the node information through its internal multi-layer neural network, ultimately outputting a static feature vector representing the node's characteristics. These feature vectors not only include the node's static attributes, such as the type of marine environmental element and the name of marine equipment, but may also include knowledge graph information related to the node, such as the node's category and other entities associated with it.

[0088] Then, we can use methods such as cosine similarity or Euclidean distance to calculate the similarity between each pair of generated static feature vectors. Cosine similarity measures the degree of similarity between two vectors in multidimensional space by calculating the cosine of the angle between them. If the feature vectors of two nodes point in similar directions in multidimensional space, their cosine similarity will be high, and vice versa.

[0089] Next, a threshold α is set. If the calculated similarity between two nodes is greater than α, an edge relationship is considered to exist between the two nodes, and a directed edge is added between them in the ocean directed graph. This process is completed by traversing all node pairs. Then, based on these edge relationships, a set of ocean prediction labels is generated, where each label indicates whether an edge relationship exists between a specific node pair.

[0090] In this embodiment, the static features of the ocean directed graph nodes are extracted by using the initial prediction output model. Then, by calculating the similarity between nodes and setting a threshold, it is automatically determined whether there is a relationship between entities. Finally, an ocean prediction label is formed. It not only automatically identifies the association between ocean entities, but also provides supervision information for subsequent model training, which helps the model learn a more sophisticated and complex relationship network between entities in the ocean environment.

[0091] In some optional embodiments, the nodes of the ocean directed graph can be adaptively updated, and the newly added nodes can be automatically associated with the entity edge by combining the trained model (i.e., the preset prediction output model) to obtain an ocean element-phenomenon-equipment scenario graph (i.e., an ocean directed graph) with adaptive expansion capabilities. The adaptive update formula is as follows:

[0092] F1(V A )=v n ;

[0093]

[0094] Among them, F1 is the trained MLP network, and the newly added node V n Encoded as feature vector v n , then calculate v n and all other nodes V in the directed graph k The eigenvector v k The similarity between them, k belongs to {E, P, A} (i.e., element V E Phenomenon V P and equipment V A For any entity in the entity set composed of n, we judge whether there is an edge relationship based on the threshold α, and construct an edge E where the condition is met (i.e. s(n, k) ≥ α). n , so that the directed graph nodes can be updated adaptively to obtain the ocean element-phenomenon-equipment scenario graph with adaptive expansion capabilities.

[0095] Figure 2 is a schematic diagram of an optional directed graph prediction model training and directed graph automatic expansion according to an embodiment of the present invention, such as Figure 2 As shown, it can be found in the existing ocean knowledge graph K G Marine environment elements V were found in E 、Phenomenon Entity V P and its static attributes, and extract the relation triple RDF to generate a directed graph G = (V, E) based on the relation triple. R Marine Equipment Entity V was discovered A And equipment association relationship Phrase, then splice the relationship triple with the equipment association relationship to obtain the edge relationship label Y. In addition, the node classification can be performed according to the directed graph to construct the set V E , V P , V A , then extract the feature of the element node v E , phenomenon node feature extraction v P , equipment node feature extraction v A , then, according to v E 、vP and v A Perform model prediction and similarity calculation F1 to obtain edge relationship prediction output After that, the edge relationship label Y and edge relationship prediction output Perform loss calculation to train the model.

[0096] And, as Figure 2 As shown, when there is a new node V n , we can use the model to predict F1 and extract the features of the newly added nodes v n , and perform edge relationship prediction and output Y n (i.e. newly added node V n and the edge relationship between any node in the current directed graph) to obtain the expanded directed graph G n .

[0097] In order to improve the accuracy of determining the feature scene graph, in the evolution method of the marine environment situation provided in Example 1 of the present application, a preset prediction output model is used to extract the entity static features of each entity information in the current marine entity information set, and based on the similarity between each two entity static features, the edge relationship between each two entity static features is determined; based on all entity information and edge relationships, the current directed graph is constructed; based on the preset prediction output model and the long short-term memory network model, the edge relationship context dynamic features and the node context dynamic features of the current directed graph are extracted; based on the edge relationship context dynamic features and the node context dynamic features, a feature scene graph is constructed.

[0098] In an embodiment of the present invention, each entity information in the current marine entity information set (including marine environmental element entity information, marine environmental phenomenon entity information, and marine equipment entity information) is input into a preset prediction output model to extract the entity static features of each entity information. Then, the cosine similarity or other similarity calculation methods can be used to compare the extracted static feature vectors of each two entities. If the similarity exceeds the preset threshold, it is considered that the two entities are associated, and a directed edge should be added to the directed graph to indicate that entity A has a certain relationship with entity B, such as "influence", "perception", etc. In this way, the association relationship between entities is preliminarily identified and reflected in the directed graph, forming a preliminary directed graph structure, which provides a framework for subsequent dynamic feature analysis. Afterwards, the current directed graph is constructed by combining the acquired entity static features and the determined edge relationships. Each node in the directed graph represents an entity, and the directed edges between nodes represent the association relationship and influence direction between entities.

[0099] In the embodiment of the present invention, the directed graph edge relationship features e are extracted based on the trained MLP directed graph prediction network F1 (i.e., the preset prediction output model) and the LSTM network F2 (i.e., the long short-term memory network model).G and node context dynamic features v G (t), then, based on the form of neighborhood aggregation, the relationship context encoding is performed to propagate the dynamic time-varying information of the node to the edge relationship feature, and the context information is encoded into the node and edge relationship to obtain the edge relationship context dynamic feature e G (t, l) and context-aggregated node context dynamic features v G (t, l), thereby obtaining the feature scene graph G S =(v G (t, l), e G (t, l)).

[0100] Specifically, a pre-defined prediction output model (MLP) can be used to extract features from edge relationships within the constructed directed graph, obtaining static features of these relationships. Simultaneously, an LSTM model is used to process node time series data, capturing dynamic changes in node attributes and generating dynamic node context features. LSTM models can memorize long-term dependencies and are therefore well-suited for processing time series data, such as the temporal trends of marine environmental factors and the status changes of marine equipment. This process extends the static directed graph into a dynamic domain, enabling the feature scene graph to reflect the dynamic interactions between entities and the time-varying characteristics of the entities themselves. The extracted dynamic edge and node context features are then integrated into the directed graph, enhancing the representation of edge relationships and nodes through dynamic features. Using a graph neighborhood aggregation algorithm, the dynamic features of nodes are propagated to their neighboring nodes and edge relationships. This also feeds back dynamic information about edge relationships to nodes, forming mutually influencing relationships. Ultimately, through this series of feature fusion and dynamic propagation, the resulting feature scene graph not only captures static entity relationships but also reflects dynamic interactions and time-varying characteristics between entities, providing richer information for situation analysis and evolution prediction.

[0101] In this example, a pre-defined predictive output model (MLP) and a long short-term memory network (LSTM) model were used to construct a feature scene graph that reflects the dynamic changes in marine entities and their interrelationships. This process not only identifies static relationships but, more importantly, introduces dynamic feature processing, enabling the feature scene graph to more accurately simulate the actual conditions of the marine environment, providing a solid foundation for subsequent evolutionary analysis and mission scenario situation representation.

[0102] In order to improve the accuracy of determining the dynamic features of edge relationship context and the dynamic features of node context, in the evolution method of marine environmental situation provided in Example 1 of the present application, a preset prediction output model is used to extract the static features of the nodes of the current directed graph, and based on the static features of the nodes, the edge relationship features between the entity information are determined; based on the marine knowledge graph, the dynamic attributes of each marine environmental phenomenon entity information are determined; the long short-term memory network model is used to encode the dynamic attributes into the static features of the nodes to obtain the dynamic features of the nodes; the neighborhood of the node represented by each entity information is determined; based on the edge relationship features, the node dynamic features and the neighborhood, the time-varying information is incorporated into the edge relationship, and the context information is encoded into the nodes and edge relationships to obtain the edge relationship context dynamic features and the node context dynamic features.

[0103] In the embodiment of the present invention, based on the constructed ocean element-phenomenon-equipment scene graph G (i.e., the current directed graph), the trained MLP network F1 (i.e., the preset prediction output model) is used to extract the node static features v G , and according to the node static characteristics v G , determine the adjacent edge relationship characteristics e between entities G And, the formula is as follows:

[0104] F1(V G )=v G ;

[0105]

[0106] Among them, V G It is an entity node of the directed graph, and the feature vector v of the node is obtained through the MLP network F1 G ;e G (n→m) represents the edge relationship feature from node n to node m. Because it is a directed graph, the direction will affect the edge relationship feature e G Direction, specifically: Calculate the eigenvector v of node n n and the feature vector v of node m m The similarity between And the similarity Compare with the threshold α to get the comparison result The comparison result is then processed by ReLU (Rectified Linear Unit), and then the processed comparison result is compared with the feature vector v n and the eigenvector v m Multiply the difference between them to get the edge relationship feature e G(n→m). Here, ReLU is a commonly used activation function, whose mathematical expression is f(x)=max(0,x). When the input x is a positive number, the function output is x; when the input x is a negative number, the function output is 0.

[0107] In addition, the dynamic attributes of entities are found in the ocean knowledge graph, mainly the ocean environment phenomenon entity V P Dynamic properties of V P (t), and then based on the long short-term memory network (LSTM) model F2 (i.e., the long short-term memory network model), the dynamic attribute V P (t) Encoding the static features of the entry node v G , forming the node dynamic feature v G (t), the formula is as follows:

[0108] v G (t)=σ(W o ·[h t-1 , V P (t)]+b o );

[0109] h t =v G (t)*tanh(C t );

[0110]

[0111] f t =σ(W f ·[h t-1 , V P (t)]+b f );

[0112] i t =σ(W i ·[h t-1 , V P (t)]+b i );

[0113]

[0114] Among them, tanh is an activation function, σ represents the Sigmoid function (i.e., an activation function); W o and W f is the weight parameter of the output gate and forget gate in the LSTM network, W i and W c is the weight parameter of the input gate in the LSTM network, b(b o 、b f 、b i) is the bias. At time t, the input of the network cell is the representation of the entity's dynamic attributes at the current moment, denoted by V P (t), C t Represents the cell state, which is the characteristics of all dynamic nodes before time t The long-term memory information is decoded by the output gate to obtain the node dynamic feature v at time t G (t);h t-1 represents the parameter weight of the parameter node of the t-1th layer calculation graph (i.e., the parameterized representation of all neighboring nodes of the t-1th layer directed graph node), h t represents the parameter weight of the parameter node of the t-th layer calculation graph; f t It is the feature of the forget gate output in the LSTM network. It is the result after the forget gate calculation. It represents a weight. It is the correlation degree between the state at the last moment t-1 and the state at the moment t in the final output of LSTM. t The larger the value, the more relevant the cell state at time t is to the cell state at the previous time t-1. t It is the output of the output gate, which indicates the characteristics of the input gate node and the temporary state caused by the input at this moment t. The degree of relevance to the final state of the LSTM, that is, the contribution; It is the temporary state at time t, without considering the influence of history, and the impact of the input at the current time t on the entire cell state; i t and The product of Indicates the state generated by the activation of the input received by the cell at the current moment to the final state C t The degree of correlation, that is, the contribution.

[0115] Figure 3 is a schematic diagram of an optional method of introducing dynamic time information based on a long short-term memory network LSTM into node features according to an embodiment of the present invention, such as Figure 3 As shown, the cell state C of the node at the previous moment can be input t-1 , the output state h at the previous moment t-1 And the dynamic attribute V at the current moment P (t), then the LSTM network uses the forget gate, input gate and output gate (i.e., f t 、i t 、o t , respectively represent the outputs of the forget gate, input gate, and output gate after being processed by the Sigmoid function σ at time t) and (i.e., the candidate cell state at time t, through the output i of the input gate t and the current input V P(t) is obtained after processing with a tanh function and is used to update and supplement the cell state C t ), combined with the current input and the state of the previous moment, the cell state and output state are updated. Specifically, the forget gate determines which memory information at the previous time point should be retained, the input gate determines which new information at the current time point should be included in the cell state, and the output gate controls how the cell state affects the output at the current time point. After that, the updated cell state C is output. t and output state h t , these two states will be used as input at the next moment, and also represent the dynamic characteristics of the node at the current time point.

[0116] Then, the relational context encoding module F3 based on neighborhood aggregation and computation graph is introduced to G (t) and edge relationship features e G , weighted on the computational graph, the time-varying information is encoded into the edge relationship, and the context information is encoded into the nodes and edges to obtain the edge relationship context dynamic feature e G (t, l) and node context dynamic features v G (t, l), thus obtaining the ocean element-phenomenon-equipment dynamic feature scene graph, the formula is as follows:

[0117]

[0118] in, is the node V of the context encoding of the layer 0 computation graph G Dynamic Features G (t); is the node V of the k-th layer computation graph context encoding G Dynamic features; σ represents the Sigmoid function; The u in it represents all points in the neighborhood of a node, and k-1 represents the k-1th layer of the computational graph. Represents the neighborhood starting from the node Each node u generates a computational graph feature h k-1 Since different nodes generate different computational graph features, the subscript u is used to represent all the k-1th layer computational graph features proposed for the nodes in the neighborhood; W k is the learnable weight, B k is the bias, is node V G In a neighborhood of a directed graph, u belongs to any node in the neighborhood. For node V G The lth layer of the computation graph features; l is the node V GThe number of neighborhood hops for finding the center of the computation graph is calculated. Since the number of neighbors increases exponentially with the increase of l, l can be set to a small value, such as 3. Edge relationship context dynamic feature e G (t, l) also contains the static edge relationship e G and dynamic edge relations e G : t(t, l), λ is the set empirical parameter, such as 0.3, n, m are the nodes in the directed graph G, v n (t, l) and v m (t, l) are the node context dynamic features of nodes n and m respectively.

[0119] Figure 4 is a schematic diagram of an optional directed graph neighborhood aggregation and calculation graph according to an embodiment of the present invention, such as Figure 4 As shown, assuming there is a directed graph containing nodes A, B, C, D, E, and F, where A has an edge relationship with B, C, D, and E; B has an edge relationship with A, C, E, and F; C has an edge relationship with A and B; D has an edge relationship with A and E; E has an edge relationship with A, B, D, and F; and F has an edge relationship with B and E. It can be determined that the neighborhood of the target node C is A and B (i.e., a one-hop neighborhood). Furthermore, a computational graph can be obtained based on the directed graph. From the computational graph, we can see the association relationship between nodes and the neighborhood of the target node C with different hop counts, for example, the neighborhood of the target node C when the neighborhood hop count is l = 0, l = 1, l = 2, and l = 3.

[0120] For example, we can start from a directed graph node, the neighborhood of this node is the n-1th layer of the computational graph, the neighborhood of all neighboring nodes together is the n-2th layer of the computational graph, and so on to the 1st layer, and finally reason n times. This n is pre-set. The larger n is, the deeper the computational graph is, and the greater the mutual connection between nodes is.

[0121] Figure 5 is a schematic diagram of generating an optional ocean environment feature scene graph according to an embodiment of the present invention, such as Figure 5 As shown, based on the existing directed graph G, the model prediction F1 is used to extract the node static features v G , and calculate the edge relationship characteristics To obtain the static edge relationship feature e G In addition, the dynamic attributes of existing nodes V can be extracted from the ocean knowledge graph. P (t), and through the long short-term memory network LSTM, according to the dynamic attribute V P (t) and node static characteristics v G , get the dynamic node feature v G (t), and combined with the dynamic node feature v G (t) and static edge relationship characteristics eG , calculate the dynamic context node feature v by computing the graph and neighborhood aggregation G (t, l) and dynamic context-varying relation features e G (t, l), and finally obtain the feature scene graph G S .

[0122] In this example, by combining a pre-set prediction output model with an LSTM model, the goal of extracting and fusing static and dynamic information from a directed graph is achieved, ultimately constructing a feature scene graph that incorporates dynamic contextual features of edge relationships. The extraction and encoding of this comprehensive feature not only reflects the static attributes of entities but also captures their dynamic changes over time, providing a more comprehensive data representation for in-depth understanding and predictive analysis of marine environmental conditions.

[0123] In order to improve the accuracy of determining the subgraph of task entity information, in the evolution method of the marine environment situation provided in Example 1 of the present application, the neighborhood of the node represented by each task entity information is determined; for each task entity information, the node context dynamic features of the node represented by the task entity information are obtained, and the edge relationship context dynamic features between the node represented by the task entity information and each node in the neighborhood are obtained; the sum of the moduli of the edge relationship context dynamic features of the node represented by the task entity information in the neighborhood is determined, wherein the sum of the moduli of the edge relationship context dynamic features is determined based on the moduli of all edge relationship context dynamic features; for each node in the neighborhood, the connection probability between the node represented by the task entity information and the node in the neighborhood is determined based on the moduli of the edge relationship context dynamic features between the node represented by the task entity information and the node in the neighborhood and the sum of the moduli of the edge relationship context dynamic features; based on all connection probabilities, the subgraph of the task entity information is determined.

[0124] In the embodiment of the present invention, based on the constructed ocean element-phenomenon-equipment feature scene graph, with a certain / certain type of phenomenon or equipment (node represented by task entity information) as the center, the node V can be determined first. G Neighborhood And get the node V G The node time-varying feature v G (t, 1) (i.e., node context dynamic features) and the time-varying features e of all nodes and edge relationships within one hop (Hop) (l=1) from the node G (t, 1) (i.e., edge relationship context dynamic features), the formula is as follows:

[0125]

[0126] e G (t, 1) = e G +e G：t(t, 1);

[0127] in, Indicates that the directed graph contains node V G and node V G The set of all neighborhood points, σ represents the Sigmoid function, W1 is the learnable weight, B1 is the bias, and n is the number of neighbors. Any node within v n (t) represents the node dynamic characteristics of node n, V u Belongs to Any node within v u (t) represents the node V u The node dynamic characteristics, e G Represents the static edge relationship characteristics between nodes in the set, e G：t (t, 1) represents the dynamic edge relationship characteristics between nodes in the set.

[0128] Then, from the collection Get the static latitude and longitude coordinate attributes of all nodes, and then calculate the slave node V G The edge relationship characteristics e between other nodes in the neighborhood that change over time G (t, 1), the edge relationship feature e is transformed through the last fully connected layer G (t, 1) is represented as the probability P of connection G (t), the formula is as follows:

[0129]

[0130] Among them, e G→n Represents node V G Edge relationship features between other nodes n in the neighborhood, e G→n：t (t, 1) represents node V G Dynamic characteristics of edge relationships between other nodes n in the neighborhood, e G→n (t, 1) represents node V G Dynamic features of edge relationships between nodes n in the neighborhood; v G Represents node V G The static characteristics of the node, v n represents the static characteristics of node n, v n (t, l) and v G (t, 1) are nodes n, V G The node context dynamic features, λ is the set empirical parameter; Node V representing task entity information G In the neighborhood The sum of the modulus lengths of the edge relationship context dynamic features within (i.e., determined based on the modulus lengths of all edge relationship context dynamic features); ||e g→n (t, 1)|| represents the node V represented by the task entity information for each node n in the neighborhood G The modulus of the dynamic feature of the edge relationship between node n and the neighborhood.

[0131] Figure 6 is a schematic diagram of generating an optional ocean environment feature scene graph according to an embodiment of the present invention, such as Figure 6 As shown, in the existing feature scene graph G S On the top, get node V G And get the neighborhood of node 1 Hop Then get the dynamic context node feature v G (t, 1) and dynamic context edge relation features e G (t, 1), then calculate the ratio of the edge relationship characteristic modulus to the total, and get the edge relationship establishment probability P G (t).

[0132] In some optional embodiments, the concept nodes V related to a certain type of specific ocean phenomenon or equipment are G , from the feature scene graph G in the form of graph representation learning S =(v G (l, t), e G (l, t)) to obtain all dynamic ontology attribute nodes v that describe a certain type / certain ocean phenomenon or equipment G The dynamic time-varying characteristics of (l, t) and ocean phenomena G (l, t), automatically build the dynamic ontology relationship (i.e., the subgraph of each node), and then display the evolution of the dynamic ontology trajectory in the situation window with time as the scale. The dynamic demonstration effect is as follows Figure 7 shown.

[0133] Figure 7 is a schematic diagram of a dynamic demonstration effect of an optional dynamic ontology relationship according to an embodiment of the present invention, such as Figure 7 As shown, the trajectory of a certain type of ocean phenomenon or equipment (recorded as the entity) evolving over time is dynamically displayed. This display method can intuitively reflect the changes in the entity's state over time, including the dynamic changes in position, state, and association relationships, allowing analysts to clearly track and understand its evolution process. For example, the dynamic trajectory of a vortex is dynamically displayed in the situation window, including: the position of the vortex (static latitude and longitude coordinates), the dynamic relationship between the vortex and temperature (phenomenon-element association relationship), the flow rate of the vortex (dynamic attribute update), the vortex angle of the vortex (dynamic attribute update), etc.

[0134] In some optional embodiments, the relevant concept nodes V involved in a certain type of specific task scenario can be G , obtain a certain type / equipment V from the feature scene graph by extracting subgraphs A Based on the scene graph centered on the situation, the situation association rule table is automatically generated based on the algorithm, and the dynamic evolution analysis results are visually displayed.

[0135] Specifically, the specific task scenario can be decomposed and all entities involved can be listed to form a task entity table K A For example, as shown in Table 1:

[0136] Table 1

[0137]

[0138]

[0139] Then, according to the task entity table K A Extracting subgraphs from feature scene graphs The formula is as follows;

[0140]

[0141] in, Represents K A Dynamic node and dynamic edge relationships in .

[0142] Afterwards, all entities in the task entity table are traversed, and an ontology dynamic relationship based on graph representation learning is constructed for each entity with dynamic attributes. Included in the situation rules table.

[0143] Figure 8 is a schematic diagram of generating an optional ocean environment feature scene graph according to an embodiment of the present invention, such as Figure 8 As shown, in the existing feature scene graph G S Get the equipment node V A and all child nodes Get child node V G And get the neighborhood of child node-Hop Then get the dynamic context node feature v G (t, 1) and dynamic context edge relation features e G (t, 1), then calculate the ratio of the edge relationship characteristic modulus to the total, and get the edge relationship establishment probability P G (t), determine whether the traversal is completed, if not, then from all child nodes Get the next child node, if yes, then end.

[0144] In order to accurately evolve the target ocean mission scenario, in the ocean environment situation evolution method provided in Example 1 of the present application, each sub-graph in the situation rule table is animated based on time in the situation window, and a situation report for each sub-graph is generated.

[0145] In the embodiment of the present invention, the task scene demonstration process and all entities involved in each process are customized according to the situation rule table. A subgraph is extracted from the feature scene graph with a single equipment (the task scene may contain multiple equipment) as the subgraph center, and its child nodes are other types of entities associated with the equipment, including various types of marine environmental elements and marine environmental phenomena. Then, according to each child node entity, the ontology dynamic relationship based on graph representation learning is indexed from the situation rule table. Then calculate the dynamic relationship x between the equipment node and the child node A (t), and finally, the sub-node sea mission situation map (i.e., sub-map) is obtained. This is dynamically displayed in the situation window based on time, and analysis results are generated. Furthermore, the dynamic demonstration effects of all equipment sub-maps in a single process can be displayed and animated. A situation report is generated for the sea mission situation map generated by each equipment in a single process and saved in a separate project folder.

[0146] Figure 9 is a schematic diagram of a dynamic demonstration effect of an optional sub-graph according to an embodiment of the present invention, such as Figure 9 As shown, for a certain task equipment in a certain task scenario, the entity association level of the task equipment can be determined, including: equipment, elements, phenomena, and element-phenomenon association, element-equipment association, phenomenon-equipment association, and then the attribute association level of the task equipment can be determined, including: equipment goals, tasks, parameters, etc., element range, resolution, value, etc., phenomenon location, type, coding, etc., and the knowledge association level of the task equipment can be further carried out based on marine professional knowledge.

[0147] The following describes in detail another optional specific implementation.

[0148] In an embodiment of the present invention, a method for representing and analyzing the evolution of an ocean environment situation map based on a scene graph is proposed. Figure 10 is a schematic diagram of an optional ocean environment situation map representation and evolution analysis based on a scene graph according to an embodiment of the present invention, such as Figure 10 As shown, the following steps are included:

[0149] S1001: Build a directed graph prediction output model in a knowledge-driven manner (including existing general marine knowledge graphs and existing marine equipment rule base information), providing the ability to automatically construct the association, perception, and impact information of marine environmental elements, marine environmental phenomena, and marine mission equipment;

[0150] S1002: Extract the directed graph edge relationship features and node dynamic time-varying features to form a feature scene graph. Then, in the form of relationship context encoding, propagate the node dynamic time-varying information into the edge relationship features, and simultaneously encode the context information contained in the edge relationship features into the nodes.

[0151] S1003: Based on the relevant concept nodes involved in a specific type of ocean phenomenon, the dynamic time-varying features of all dynamic ontology attribute nodes and ocean phenomenon edges describing a certain type / specific ocean phenomenon or equipment are obtained from the feature scene graph using graph representation learning, and dynamic ontology relationships are automatically constructed.

[0152] S1004, based on the relevant concept nodes involved in a certain type of specific task scenario, obtain a scene graph centered on a certain type / equipment from the feature scene graph by extracting subgraphs, automatically generate a situation association rule table based on the algorithm, and visualize the dynamic evolution analysis results.

[0153] In an embodiment of the present invention, a knowledge-driven approach is used to provide interpretability and reliability for the scene graph, and it maintains a certain degree of robustness in the face of insufficient relevant data intelligence or complex situational situations in specific tasks. This helps to construct a marine environment scene graph with dynamic expansion and adaptive updating of edge relationships for situation display based on marine tasks, and provides a basis for dynamic evolution analysis for the display of scene situations, thereby providing professional marine knowledge and internal logic for situation display and task decision-making. In addition, the unstructured marine environment scene graph is converted into a quantitative, structured, and computable feature scene graph through feature representation, providing the ability to quantify, store, and review the scene situation for situation display, helping to provide a decision-making basis based on calculation for decision-making applications related to marine tasks, supporting the dynamic evolution analysis and interpretation of marine environment situations based on time series relationships, thereby providing quantitative computing support for specific applications of marine tasks. In addition, according to specific application requirements, it supports a certain type of marine phenomenon or a certain equipment that performs a specific task as the ontology, constructing ontology dynamic relationships from the feature scene graph, and then realizing adaptive updates of ontology relationships based on the time-varying features of the feature scene graph. At the same time, it also supports the dynamic display of the sea mission situation map and its evolution analysis results in the visualization interface through sub-graph extraction based on the relevant concept nodes involved in a certain type of specific mission scenario, and provides services such as adaptive update of thematic maps, dynamic evolution analysis based on time series, and judgment of the pros and cons of situation information for specific ocean-oriented business applications. It helps to provide dynamic and easy-to-read situation visualization graphics for the display of marine environmental situation information, support marine environmental situation-related decision-making for sea-area applications, and thus provide visual information guidance and browsing for specific applications of marine missions.

[0154] The following describes it in detail with reference to another embodiment.

[0155] Example 2

[0156] The marine environment situation evolution device provided in this embodiment includes multiple implementation units, each implementation unit corresponding to each implementation step in the above-mentioned embodiment 1.

[0157] Figure 11 is a schematic diagram of an optional ocean environment situation evolution device according to an embodiment of the present invention, such as Figure 11 As shown, the evolution device may include: a first determination unit 1100 , a processing unit 1101 , a second determination unit 1102 , and a third determination unit 1103 .

[0158] The first determining unit 1100 is configured to determine a current ocean entity information set, wherein the current ocean entity information set includes: ocean environment element entity information, ocean environment phenomenon entity information, and ocean equipment entity information;

[0159] The processing unit 1101 is configured to process the current ocean entity information set using a preset prediction output model to obtain a feature scene graph;

[0160] The second determining unit 1102 is configured to receive the target ocean mission scenario and determine all mission entity information involved in the target ocean mission scenario;

[0161] The third determining unit 1103 is configured to determine a subgraph of each task entity information based on the feature scene graph, obtain a situation rule table of the target ocean task scene, and evolve the target ocean task scene based on the situation rule table.

[0162] The above-mentioned evolution device can use the preset prediction output model obtained in advance to process the current set of ocean entity information, and can obtain the ocean element-phenomenon-equipment scene graph. Then, by further processing the ocean element-phenomenon-equipment scene graph, a feature scene graph can be obtained. Using the feature scene graph, a subgraph of each entity in the ocean mission scene can be accurately and efficiently obtained, so that the ocean mission scene can be evolved according to the subgraph. It can effectively predict and analyze the ocean environment situation that the ocean mission scene is concerned about, thereby achieving the technical effect of accurately and dynamically evolving the ocean environment situation, and thus solving the technical problem in related technologies that the situation of the ocean mission scene cannot be accurately and dynamically evolved.

[0163] Optionally, the evolution device also includes: a first determination module, used to determine the initial prediction output model before using the preset prediction output model to process the current ocean entity information set; a first construction module, used to construct an ocean directed graph and ocean labels based on the ocean knowledge graph and the ocean equipment rule base; a second determination module, used to determine the ocean prediction label based on the ocean directed graph and the initial prediction output model; a first training module, used to train the initial prediction output model based on the ocean label and the ocean prediction label to obtain the preset prediction output model.

[0164] Optionally, the first construction module includes: a first extraction submodule, used to extract ocean entity information and the ocean association relationship between every two ocean entity information from the ocean knowledge graph, wherein the ocean entity information includes ocean environment element entity information and ocean environment phenomenon entity information; a first construction submodule, used to construct a first directed graph based on the ocean entity information and the ocean association relationship; a second extraction submodule, used to extract ocean equipment entity information and equipment association relationship from the ocean equipment rule base, wherein the equipment association relationship is the association relationship between ocean equipment entity information and another ocean equipment entity information, ocean environment element entity information or ocean environment phenomenon entity information; a second construction submodule, used to construct a second directed graph based on the ocean equipment entity information and the equipment association relationship; a first overlapping submodule, used to overlap all repeated nodes based on the first directed graph and the second directed graph to obtain an ocean directed graph, and extract all edge relationships in the ocean directed graph to obtain ocean labels, wherein the nodes are used to represent ocean environment element entity information, ocean environment phenomenon entity information or ocean equipment entity information.

[0165] Optionally, the second determination module includes: a third extraction submodule, used to extract the node static features of each node on the ocean directed graph using the initial prediction output model; a first calculation submodule, used to calculate the similarity between the static features of each two nodes; a first determination submodule, used to determine that there is an edge relationship between two nodes indicated by a similarity greater than a preset threshold, and obtain an ocean prediction label based on all edge relationships.

[0166] Optionally, the processing unit includes: a first extraction module, which is used to extract the entity static features of each entity information in the current ocean entity information set using a preset prediction output model, and determine the edge relationship between each two entity static features based on the similarity between each two entity static features; a second construction module, which is used to construct the current directed graph based on all entity information and edge relationships; a second extraction module, which is used to extract the edge relationship context dynamic features and node context dynamic features of the current directed graph based on the preset prediction output model and the long short-term memory network model; a third construction module, which is used to construct a feature scene graph based on the edge relationship context dynamic features and the node context dynamic features.

[0167] Optionally, the second extraction module includes: a second determination submodule, which is used to extract the static features of the nodes of the current directed graph using a preset prediction output model, and determine the edge relationship features between the entity information based on the node static features; a third determination submodule, which is used to determine the dynamic attributes of each marine environmental phenomenon entity information based on the marine knowledge graph; a first encoding submodule, which is used to encode the dynamic attributes into the node static features using a long short-term memory network model to obtain the node dynamic features; a fourth determination submodule, which is used to determine the neighborhood of the node represented by each entity information; a second encoding submodule, which is used to encode time-varying information into the edge relationship based on the edge relationship features, the node dynamic features and the neighborhood, and encode the context information into the nodes and edge relationships to obtain the edge relationship context dynamic features and the node context dynamic features.

[0168] Optionally, the third determination unit includes: a third determination module, used to determine the neighborhood of the node represented by each task entity information; a first acquisition module, used to obtain the node context dynamic features of the node represented by the task entity information for each task entity information, and obtain the edge relationship context dynamic features between the node represented by the task entity information and each node in the neighborhood; a fourth determination module, used to determine the sum of the moduli of the edge relationship context dynamic features of the node represented by the task entity information in the neighborhood, wherein the sum of the moduli of the edge relationship context dynamic features is determined based on the moduli of all edge relationship context dynamic features; a fifth determination module, used to determine, for each node in the neighborhood, the connection probability between the node represented by the task entity information and the node in the neighborhood based on the moduli of the edge relationship context dynamic features between the node represented by the task entity information and the node in the neighborhood and the sum of the moduli of the edge relationship context dynamic features; a sixth determination module, used to determine the subgraph of the task entity information based on all connection probabilities.

[0169] Optionally, the third determining unit further includes: a first demonstration module, configured to perform animation demonstration on each subgraph in the situation rule table based on time in the situation window, and generate a situation report for each subgraph.

[0170] The above-mentioned evolution device may also include a processor and a memory. The above-mentioned first determination unit 1100, processing unit 1101, second determination unit 1102, third determination unit 1103, etc. are all stored in the memory as program units, and the processor executes the above-mentioned program units stored in the memory to realize corresponding functions.

[0171] The processor includes a core that retrieves the corresponding program unit from the memory. One or more cores can be configured, and the target ocean mission scenario can be evolved based on the situation rule table by adjusting the core parameters.

[0172] The above-mentioned memory may include non-permanent memory in a computer-readable medium, random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM, and the memory includes at least one memory chip.

[0173] The present invention also provides a computer program product, which, when executed on a data processing device, is suitable for executing an initialization program having the following method steps: determining a current ocean entity information set, processing the current ocean entity information set using a preset prediction output model, obtaining a characteristic scene graph, receiving a target ocean task scene, and determining all task entity information involved in the target ocean task scene, determining a subgraph of each task entity information based on the characteristic scene graph, obtaining a situation rule table of the target ocean task scene, and evolving the target ocean task scene based on the situation rule table.

[0174] According to another aspect of an embodiment of the present invention, a computer program product is also provided, including a non-volatile computer-readable storage medium, the non-volatile computer-readable storage medium storing a computer program, and when the computer program is executed by a processor, implementing any of the above-mentioned methods for evolving the marine environment situation.

[0175] According to another aspect of an embodiment of the present invention, an electronic device is also provided, comprising one or more processors and a memory, wherein the memory is used to store one or more programs, wherein when the one or more programs are executed by one or more processors, the one or more processors implement the above-mentioned method for evolving the marine environment situation.

[0176] Figure 12 FIG. 1 is a hardware structure diagram of an electronic device (or mobile device) for an evolution method of ocean environment situation according to an embodiment of the present invention. Figure 12 As shown, the electronic device may include one or more processors (e.g., Figure 12 The processors 1202a, 1202b, ..., 1202n, etc., which may include but are not limited to processing devices such as microprocessors (MCUs) or programmable logic devices (FPGAs), and a memory 1204 for storing data. In addition, the processors 1202a, 1202b, ..., 1202n, etc., may also include: a display, an input / output interface (I / 0 interface), a universal serial bus (USB) port (which may be included as one of the ports of the I / 0 interface), a network interface, a keyboard, a power supply, and / or a camera. It will be understood by those skilled in the art that Figure 12 The structure shown is only for illustration and does not limit the structure of the above electronic device. Figure 12 More or fewer components than shown, or with Figure 12 Different configurations shown.

[0177] The serial numbers of the above embodiments of the present invention are for description only and do not represent the advantages or disadvantages of the embodiments.

[0178] The embodiments or examples of the present disclosure are not exhaustive, but are merely illustrations of some embodiments or examples, and are not intended to be specific limitations on the scope of protection of the present disclosure. In the absence of contradiction, each step in a certain embodiment or example can be implemented as an independent example, and the steps can be arbitrarily combined. For example, a solution after removing some steps in a certain embodiment or example can also be implemented as an independent example, and the order of the steps in a certain embodiment or example can be arbitrarily exchanged. In addition, the optional methods or optional examples in a certain embodiment or example can be arbitrarily combined; in addition, the various embodiments or examples can be arbitrarily combined. For example, some or all steps of different embodiments or examples can be arbitrarily combined, and a certain embodiment or example can be arbitrarily combined with the optional methods or optional examples of other embodiments or examples.

[0179] In the above embodiments of the present invention, the description of each embodiment has its own focus. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0180] In the several embodiments provided by the present invention, it should be understood that the disclosed technical content can be implemented in other ways. Among them, the device embodiments described above are merely illustrative. For example, the division of the units can be a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of units or modules, and can be electrical or other forms.

[0181] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple units. Some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0182] In addition, the functional units in the various embodiments of the present invention may be integrated into a single processing unit, each unit may exist physically separately, or two or more units may be integrated into a single unit. The aforementioned integrated units may be implemented in the form of hardware or software functional units.

[0183] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: a USB flash drive, a read-only memory (ROM, Read-Only Memory), a memory card, ... y Memory), random access memory (RAM), mobile hard disk, magnetic disk or CD-ROM and other media that can store program code.

[0184] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be regarded as within the scope of protection of the present invention.

Claims

1. A method for evolving marine environmental conditions, characterized in that: include: Determine a current ocean entity information set, wherein the current ocean entity information set includes: ocean environment element entity information, ocean environment phenomenon entity information, and ocean equipment entity information; Using a preset prediction output model to process the current ocean entity information set to obtain a feature scene graph; Receive a target ocean mission scenario and determine all mission entity information involved in the target ocean mission scenario; Based on the characteristic scene graph, a subgraph of each task entity information is determined, a situation rule table of the target ocean task scene is obtained, and the target ocean task scene is evolved based on the situation rule table.

2. The evolution method according to claim 1, characterized in that: Before using the preset prediction output model to process the current ocean entity information set, the method further includes: Determine the initial prediction output model; Based on the ocean knowledge graph and ocean equipment rule library, build ocean directed graph and ocean labels; Determining an ocean prediction label based on the ocean directed graph and the initial prediction output model; Based on the ocean label and the ocean prediction label, the initial prediction output model is trained to obtain the preset prediction output model.

3. The evolution method according to claim 2, characterized in that: Based on the ocean knowledge graph and ocean equipment rule base, the steps to construct the ocean directed graph and ocean labels include: Extracting ocean entity information and ocean association relationships between every two pieces of ocean entity information from the ocean knowledge graph, wherein the ocean entity information includes: ocean environment element entity information and ocean environment phenomenon entity information; Constructing a first directed graph based on the ocean entity information and the ocean association relationship; Extracting the marine equipment entity information and equipment association relationship from the marine equipment rule library, wherein the equipment association relationship is an association relationship between the marine equipment entity information and another marine equipment entity information, the marine environment element entity information, or the marine environment phenomenon entity information; Constructing a second directed graph based on the marine equipment entity information and the equipment association relationship; Based on the first directed graph and the second directed graph, all repeated nodes are overlapped to obtain the ocean directed graph, and all edge relationships in the ocean directed graph are extracted to obtain the ocean label, wherein the node is used to represent the ocean environment element entity information, the ocean environment phenomenon entity information or the ocean equipment entity information.

4. The evolution method according to claim 2, characterized in that The step of determining an ocean prediction label based on the ocean directed graph and the initial prediction output model comprises: Extracting node static features of each node on the ocean directed graph using the initial prediction output model; Calculating the similarity between the static features of each two nodes; It is determined that two nodes indicated by the similarity greater than a preset threshold have an edge relationship, and the ocean prediction label is obtained based on all the edge relationships.

5. The evolution method according to claim 1, characterized in that: The step of using a preset prediction output model to process the current ocean entity information set to obtain a feature scene graph includes: Extracting entity static features of each entity information in the current ocean entity information set using the preset prediction output model, and determining an edge relationship between each two entity static features based on the similarity between each two entity static features; Constructing a current directed graph based on all the entity information and the edge relationships; Extracting edge relationship context dynamic features and node context dynamic features of the current directed graph based on the preset prediction output model and the long short-term memory network model; The feature scene graph is constructed based on the edge relationship context dynamic features and the node context dynamic features.

6. The evolution method according to claim 5, characterized in that: The step of extracting edge relationship context dynamic features and node context dynamic features of the current directed graph based on the preset prediction output model and the long short-term memory network model includes: Extracting static features of nodes of the current directed graph using the preset prediction output model, and determining edge relationship features between the entity information based on the static features of the nodes; Determine the dynamic attributes of each of the marine environmental phenomenon entity information based on the marine knowledge graph; The long short-term memory network model is used to encode the dynamic attributes into the node static features to obtain the node dynamic features; Determining a neighborhood of each node represented by the entity information; Based on the edge relationship features, the node dynamic features and the neighborhood, time-varying information is incorporated into the edge relationship, and context information is encoded into the node and the edge relationship to obtain the edge relationship context dynamic features and the node context dynamic features.

7. The evolution method according to claim 1, characterized in that: The step of determining a subgraph of each task entity information based on the feature scene graph includes: Determining the neighborhood of each node represented by the task entity information; For each task entity information, obtaining node context dynamic features of the node represented by the task entity information, and obtaining edge relationship context dynamic features between the node represented by the task entity information and each node in the neighborhood; Determine the sum of the modulus lengths of the edge relationship context dynamic features of the node represented by the task entity information within the neighborhood, wherein the sum of the modulus lengths of the edge relationship context dynamic features is determined based on the modulus lengths of all the edge relationship context dynamic features; For each node in the neighborhood, determining a connection probability between the node represented by the task entity information and the node in the neighborhood based on the modulus length of the edge relationship context dynamic feature between the node represented by the task entity information and the node in the neighborhood and the sum of the modulus lengths of the edge relationship context dynamic features; Based on all the connection probabilities, the subgraph of the task entity information is determined.

8. The evolution method according to claim 1, characterized in that: The step of evolving the target ocean mission scenario based on the situation rule table includes: In the situation window, animation demonstration is performed on each subgraph in the situation rule table based on time, and a situation report of each subgraph is generated.

9. An evolution device for ocean environment situation, characterized in that: include: A first determining unit is configured to determine a current ocean entity information set, wherein the current ocean entity information set includes: ocean environment element entity information, ocean environment phenomenon entity information, and ocean equipment entity information; a processing unit, configured to process the current ocean entity information set using a preset prediction output model to obtain a feature scene graph; A second determining unit is configured to receive a target ocean mission scenario and determine all mission entity information involved in the target ocean mission scenario; The third determining unit is used to determine the subgraph of each task entity information based on the feature scene graph, obtain the situation rule table of the target ocean task scene, and evolve the target ocean task scene based on the situation rule table.

10. An electronic device, characterized in that: It includes one or more processors and a memory, wherein the memory is used to store one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the method for evolving the marine environment situation as described in any one of claims 1 to 8.

Images