Method and device for constructing knowledge graph based on electromagnetic situation

Abstract

The application relates to an electromagnetic situation-based knowledge graph construction method and device, applied to the field of electromagnetic situation awareness, wherein the method comprises the following steps: constructing initial triples of an initial knowledge graph according to electromagnetic signal instances; the electromagnetic signal instances comprise signal intent classification of electromagnetic signals; inputting the initial triples into a preset knowledge graph embedding model with a path attention mechanism; determining implicit triples with entity relationships in the initial triples according to preset implicit association relationships in the knowledge graph embedding model; the preset implicit association relationships comprise indirect relationships derived based on electromagnetic field knowledge; and expanding the initial knowledge graph based on the implicit triples, thereby solving the problem that current knowledge graphs based on electromagnetic signals are difficult to adapt to dynamic evolution of electromagnetic signal situations.

Description

Technical Field

[0001] This application relates to the field of electromagnetic situational awareness, and in particular to a method and apparatus for constructing a knowledge graph based on electromagnetic situational awareness. Background Technology

[0002] With the explosive growth of communication, radar, electronic warfare, and Internet of Things (IoT) devices, the electromagnetic space has become a highly crowded, dynamic, time-varying, and extremely complex field with distinct multi-source heterogeneous characteristics. The data sources in the electromagnetic space are extremely diverse, including but not limited to I / Q data (quadrature phase modulation data) from reconnaissance receivers and spectrum analyzers, signal data from different systems such as radar, communication, and data links, as well as geographical and environmental data, equipment and platform data, etc.

[0003] To address the challenge of the explosive growth of multi-source heterogeneous data in current electromagnetic space perception and cognition, current methods primarily rely on artificial features and prior rules to build knowledge graphs for this data, aiming to achieve semantic associations and dynamic reasoning between data. However, when processing time-varying signal features, radiation source behavior, and dynamic environments, bottlenecks exist such as weak semantic associations, low reasoning efficiency, and insufficient scalability. Therefore, current knowledge graphs for electromagnetic signals are difficult to adapt to the dynamic evolution of electromagnetic signal situations.

[0004] There is currently no effective solution to the problem that knowledge graphs based on electromagnetic signals cannot adapt to the dynamic evolution of electromagnetic signal situations. Summary of the Invention

[0005] This embodiment provides a method and apparatus for constructing a knowledge graph based on electromagnetic situation, in order to solve the problem that current knowledge graphs based on electromagnetic signals are difficult to adapt to the dynamic evolution of electromagnetic signal situation in related technologies.

[0006] Firstly, this embodiment provides a method for constructing a knowledge graph based on electromagnetic situation, the method comprising:

[0007] Based on electromagnetic signal instances, initial triples are constructed for an initial knowledge graph; the electromagnetic signal instances include the signal intent classification of the electromagnetic signals.

[0008] The initial triples are input into a preset knowledge graph embedding model; in the knowledge graph embedding model, implicit triples with entity relationships are determined from the initial triples according to preset implicit relationships; the preset knowledge graph embedding model has a path attention mechanism to map the graph structure composed of entities and relationships in the initial knowledge graph to a continuous low-dimensional vector space; the preset implicit relationships include indirect relationships derived from knowledge in the electromagnetic domain;

[0009] The initial knowledge graph is expanded based on the hidden triples.

[0010] In some embodiments, constructing the initial triples of the initial knowledge graph based on electromagnetic signal instances includes:

[0011] Based on a preset ontology editing tool, an ontology model of an initial knowledge graph is constructed according to knowledge in the electromagnetic domain; the ontology model of the initial knowledge graph contains the entity and attribute content of the electromagnetic signal;

[0012] Based on the electromagnetic signals containing signal intent classification in the electromagnetic signal examples, a preset sequential deep model is trained to obtain an intent classification neural network.

[0013] The electromagnetic signal instance is input into the intent classification neural network to obtain the signal intent classification of the electromagnetic signal instance;

[0014] Based on the signal intent classification of the electromagnetic signal instance, and the entity and attribute content of the electromagnetic signal, the initial triples of the initial knowledge graph are constructed.

[0015] In some embodiments, the initial triples are input into a preset knowledge graph embedding model; the knowledge graph embedding model determines the implicit triples with entity relationships in the initial triples according to preset implicit associations, including:

[0016] The initial triples are converted into entity embedding vectors and relation embedding vectors, and then input into the preset knowledge graph embedding model.

[0017] Using the preset knowledge graph embedding model and based on the path attention mechanism, the path scores of multiple entity paths between the entity embedding vectors corresponding to the initial triples are calculated; the entity paths contain the preset implicit associations.

[0018] The hidden triples are determined based on the path scores.

[0019] In some embodiments, the step of calculating path scores for multiple entity paths between entity embedding vectors corresponding to the initial triples using the preset knowledge graph embedding model and based on a path attention mechanism includes:

[0020] The paths of the head and tail entities of the entity embedding vector corresponding to the initial triple are marked to obtain the marked entity paths;

[0021] Based on a preset nonlinear activation function, the labeled entity paths, as well as the global importance weights and local importance weights of the entity paths, are processed to obtain the path scores of multiple entity paths between the entity embedding vectors.

[0022] In some embodiments, the initial knowledge graph is expanded based on the hidden triples, including:

[0023] The entities, attributes, and signal intents corresponding to the implicit triples are classified and added to the initial knowledge graph to obtain an extended knowledge graph.

[0024] In some embodiments, the method further includes: storing the extended knowledge graph in a preset graph database, and mining association information rules in the extended knowledge graph through the graph database; the association information rules are determined by association information in the extended knowledge graph that occurs more than a preset frequency;

[0025] Based on the association information rules and the signal intent classification of the electromagnetic signal instances, a new knowledge triplet is established;

[0026] The extended knowledge graph is updated based on the new knowledge triples.

[0027] In some embodiments, mining the association rules in the extended knowledge graph through the graph database includes:

[0028] By using the graph database, we can mine the association rules with dynamic temporal relationships in the extended knowledge graph.

[0029] Secondly, this embodiment provides a knowledge graph construction device based on electromagnetic situation, the device comprising: a construction module, a hidden relationship acquisition module, and an extension module;

[0030] The construction module is used to construct initial triples of the initial knowledge graph based on electromagnetic signal instances; the electromagnetic signal instances include the signal intent classification of the electromagnetic signals.

[0031] The implicit relation acquisition module is used to input the initial triples into a preset knowledge graph embedding model, and determine the implicit triples with entity relations in the initial triples according to the preset implicit relation. The preset knowledge graph embedding model introduces a path attention mechanism to map the graph structure composed of entities and relations in the initial knowledge graph to a continuous low-dimensional vector space. The preset implicit relation includes indirect relations derived based on electromagnetic domain knowledge.

[0032] The extension module is used to extend the initial knowledge graph based on the hidden triples.

[0033] Thirdly, this embodiment provides an electronic device including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the knowledge graph construction method based on electromagnetic situation described in the first aspect.

[0034] Fourthly, this embodiment provides a storage medium storing a computer program that, when executed by a processor, implements the knowledge graph construction method based on electromagnetic situation described in the first aspect above.

[0035] Compared with related technologies, the knowledge graph construction method and apparatus based on electromagnetic situation provided in this embodiment constructs initial triples of the initial knowledge graph by using the inherent entities and attributes in electromagnetic signal instances, as well as additionally determined signal intent classifications, to realize the association between electromagnetic signals and signal intents. Subsequently, the initial triples are input into a preset knowledge graph embedding model that introduces a path attention mechanism. Then, the knowledge graph embedding model determines the implicit triples in the initial triples that include implicit relationships. Based on the implicit triples, the implicit indirect relationship paths in the initial knowledge graph are expanded, so that the knowledge graph has apparent association paths and implicit association paths corresponding to electromagnetic signals. This is beneficial to improving the adaptability of the knowledge graph to dynamically evolving electromagnetic signals, and further improving the accuracy and efficiency of signal intent analysis for electromagnetic signals.

[0036] Details of one or more embodiments of this application are set forth in the following drawings and description to make other features, objects and advantages of this application more readily apparent. Attached Figure Description

[0037] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings:

[0038] Figure 1 This is a hardware structure block diagram of the terminal of the knowledge graph construction method based on electromagnetic situation provided in the embodiments of this application;

[0039] Figure 2 This is a flowchart of the knowledge graph construction method based on electromagnetic situation provided in the embodiments of this application;

[0040] Figure 3 This is a flowchart of the electromagnetic target knowledge graph construction method based on deep learning provided in this specific embodiment;

[0041] Figure 4 This is a schematic diagram of the electromagnetic domain knowledge graph ontology provided in this specific embodiment;

[0042] Figure 5a This is a schematic diagram of the time-frequency domain waveform corresponding to a reconnaissance signal debugging method provided in this specific embodiment;

[0043] Figure 5b This is a time-frequency domain waveform diagram corresponding to another detection signal debugging method provided in this specific embodiment;

[0044] Figure 5c This is a time-frequency domain waveform diagram corresponding to another reconnaissance signal debugging method provided in this specific embodiment;

[0045] Figure 6a This is a time-frequency domain waveform diagram corresponding to an interference signal debugging method provided in this specific embodiment;

[0046] Figure 6b This is a time-frequency domain waveform diagram corresponding to another interference signal debugging method provided in this specific embodiment;

[0047] Figure 6c This is a time-frequency domain waveform diagram corresponding to another interference signal debugging method provided in this specific embodiment;

[0048] Figure 7a This is a schematic diagram of the time-frequency domain waveform corresponding to a detection signal debugging method provided in this specific embodiment;

[0049] Figure 7b This is a time-frequency domain waveform diagram corresponding to another detection signal debugging method provided in this specific embodiment;

[0050] Figure 7c This is a time-frequency domain waveform diagram corresponding to another detection signal debugging method provided in this specific embodiment;

[0051] Figure 8 This is a schematic diagram illustrating the accuracy change of the signal intent classification neural network provided in this preferred embodiment during the training process;

[0052] Figure 9 This is a schematic diagram illustrating the loss change of the signal intent classification neural network provided in this preferred embodiment during the training process;

[0053] Figure 10 This is a schematic diagram of the PConvKB model provided in this specific embodiment;

[0054] Figure 11 This is a schematic diagram of the knowledge graph reasoning path and confidence propagation provided in this specific embodiment;

[0055] Figure 12 This is a structural block diagram of the knowledge graph construction device based on electromagnetic situation provided in the embodiments of this application. Detailed Implementation

[0056] To better understand the purpose, technical solution, and advantages of this application, the application is described and illustrated below in conjunction with the accompanying drawings and embodiments.

[0057] Unless otherwise defined, the technical or scientific terms used in this application shall have the general meaning understood by one of ordinary skill in the art to which this application pertains. Words such as “a,” “an,” “an,” “the,” “the,” and “these” used in this application do not indicate quantitative limitation and may be singular or plural. The terms “comprising,” “including,” “having,” and any variations thereof used in this application are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or device that comprises a series of steps or modules (units) is not limited to the listed steps or modules (units) but may include steps or modules (units) not listed, or may include other steps or modules (units) inherent to these processes, methods, products, or devices. Words such as “connected,” “linked,” and “coupled” used in this application are not limited to physical or mechanical connections but may include electrical connections, whether direct or indirect. “Multiple” used in this application refers to two or more. “And / or” describes the relationship between related objects, indicating that three relationships may exist; for example, “A and / or B” can represent: A alone, A and B simultaneously, and B alone. Normally, the character " / " indicates that the objects before and after it are in an "or" relationship. The terms "first," "second," "third," etc., used in this application are merely to distinguish similar objects and do not represent a specific order of objects.

[0058] The method embodiments provided in this example can be executed on a terminal, computer, or similar computing device. For example, it can run on a terminal. Figure 1 This is a hardware structure block diagram of the terminal for the knowledge graph construction method based on electromagnetic situation provided in this application embodiment. For example... Figure 1 As shown, a terminal may include one or more ( Figure 1 Only one is shown in the diagram. A processor 102 and a memory 104 for storing data are also included. The processor 102 may be, but is not limited to, a microprocessor (MCU) or a programmable logic device (FPGA). The terminal may also include a transmission device 106 for communication functions and an input / output device 108. Those skilled in the art will understand that… Figure 1 The structure shown is for illustrative purposes only and does not limit the structure of the terminal described above. For example, the terminal may also include components that are larger than... Figure 1The more or fewer components shown, or having the same Figure 1 The different configurations shown are illustrated.

[0059] The memory 104 can be used to store computer programs, such as application software programs and modules, like the computer program corresponding to the knowledge graph construction method based on electromagnetic situation in this embodiment. The processor 102 executes various functional applications and data processing by running the computer programs stored in the memory 104, thereby implementing the above-described method. The memory 104 may include high-speed random access memory, and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory 104 may further include memory remotely located relative to the processor 102, and these remote memories can be connected to the terminal via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0060] The transmission device 106 is used to receive or send data via a network. This network includes a wireless network provided by the terminal's communication provider. In one example, the transmission device 106 includes a Network Interface Controller (NIC), which can connect to other network devices via a base station to communicate with the Internet. In another example, the transmission device 106 can be a Radio Frequency (RF) module used for wireless communication with the Internet.

[0061] Electromagnetic situation awareness refers to electronic countermeasures reconnaissance technology that detects and predicts the state and situation formed by the frequency-using equipment, equipment configuration, and electromagnetic activities and changes of various parties in the electromagnetic space within a specific time and space range. It mainly conducts reconnaissance of electromagnetic signals in the electromagnetic space.

[0062] Current electromagnetic space perception and cognition face the challenge of an explosive growth in multi-source heterogeneous data. However, traditional methods for electromagnetic radiation source identification and intent analysis based on expert rules and template matching mainly rely on manual features and prior rules. These methods have poor generalization ability for new systems and non-cooperative electromagnetic signals, making it difficult to effectively achieve semantic association and dynamic reasoning between electromagnetic data when constructing knowledge graphs. In particular, when dealing with time-varying electromagnetic signal characteristics, radiation source behavior, and dynamic environments, bottlenecks such as weak semantic association, low reasoning efficiency, and insufficient scalability exist, leading to a break in the cognitive chain from electromagnetic signals to radiation source behavior, and failing to meet the urgent need for real-time, accurate situational awareness and decision support.

[0063] To avoid the traditional template-based analysis of electromagnetic behavior and intent, existing technologies generally combine artificial intelligence and electromagnetic situational awareness. For example, data-driven methods for identifying and analyzing radiation source signals employ electromagnetic signal processing methods based on deep learning such as CNN, LSTM, and Transformer, which effectively solves the problem of generalization ability. However, deep learning-based electromagnetic signal processing methods only achieve signal-level / target-level classification, lacking domain knowledge guidance and reasoning logic links. As a result, they still cannot associate combat behavior and combat intent during the construction of knowledge graphs, i.e., they cannot complete the reasoning from signal features to combat behavior and intent.

[0064] Furthermore, existing technologies lack large-scale, high-quality labeled electromagnetic datasets for electromagnetic signal intent recognition, which restricts the development of high-order behavior analysis models based on neural networks. Meanwhile, the construction and reasoning processes of knowledge graphs often rely on fixed rules, making it difficult to adapt to the dynamic evolution of situations, resulting in insufficient real-time performance, accuracy, and interpretability of behavior analysis and intent prediction. Therefore, there is an urgent need for a dynamic cognitive framework that can deeply integrate signal features, knowledge graphs, and intelligent reasoning to provide technical support for reasoning and analysis from "signals" to "intents."

[0065] To establish a complete analysis chain from electromagnetic signal parameters to behavioral intent, this embodiment provides a method for constructing a knowledge graph based on electromagnetic situation. Figure 2 This is a flowchart of the knowledge graph construction method based on electromagnetic situation provided in the embodiments of this application, such as... Figure 2 As shown, the process includes the following steps:

[0066] Step S210: Based on the electromagnetic signal instances, construct the initial triples of the initial knowledge graph; the electromagnetic signal instances include the signal intent classification of the electromagnetic signals.

[0067] In constructing a knowledge graph based on electromagnetic situation, it is necessary to first integrate electromagnetic domain knowledge, including industry expert knowledge and open-source intelligence, to determine the entities and attributes in the knowledge graph pattern layer and form the initial structural framework of the knowledge graph. At the same time, it is necessary to obtain the signal intent classification of multiple open-source electromagnetic signal instances, as well as other specific content of the electromagnetic signals, such as the depth and distance of electromagnetic signal detection, and sonar for interference signals.

[0068] Based on the signal intent classification of electromagnetic signals, as well as the entity and attribute content of electromagnetic signals, the initial triples of the initial knowledge graph are constructed.

[0069] Step S220: Input the initial triples into the preset knowledge graph embedding model; in the knowledge graph embedding model, determine the implicit triples with entity relationships in the initial triples according to the preset implicit associations.

[0070] The preset knowledge graph embedding model has a path attention mechanism, which is used to map the graph structure composed of entities and relations in the initial knowledge graph to a continuous low-dimensional vector space; the preset implicit relationships include indirect relationships derived from knowledge in the electromagnetic domain.

[0071] Knowledge graph embedding models utilize supervised learning to learn embeddings and vector representations of nodes and edges. After inputting initial triples into a pre-defined knowledge graph embedding model, the model uses a path attention mechanism to uncover implicit relationships among entities within the initial triples, identifying implicit triples with entity relationships. This approach not only considers the initial triples themselves but also calculates the local and global importance of multi-hop paths between entities within the initial triples through path attention, thereby uncovering deep relationships hidden within the knowledge graph.

[0072] Step S230: Expand the initial knowledge graph based on the hidden triples.

[0073] Specifically, the initial knowledge graph is expanded by using implicit triples with implicit relationships. The entities, attributes, and signal intents corresponding to the implicit triples are classified and added to the initial knowledge graph to obtain an expanded knowledge graph, which can adapt to dynamically evolving electromagnetic signals.

[0074] Through the above steps, initial triples of the initial knowledge graph are constructed using the inherent entities and attributes in electromagnetic signal instances, as well as the additionally determined signal intent classification, thus realizing the association between electromagnetic signals and signal intents. Subsequently, the initial triples are input into a pre-defined knowledge graph embedding model that incorporates a path attention mechanism. The knowledge graph embedding model then determines the implicit triples in the initial triples that include implicit relationships. Based on the implicit triples, the implicit indirect relationship paths in the initial knowledge graph are expanded, so that the knowledge graph contains both apparent and implicit relationship paths corresponding to electromagnetic signals. This improves the adaptability of the knowledge graph to dynamically evolving electromagnetic signals and further enhances the accuracy and efficiency of signal intent analysis for electromagnetic signals.

[0075] In some embodiments, step S210, which constructs initial triples of the initial knowledge graph based on electromagnetic signal instances, includes:

[0076] Step S211: Based on the preset ontology editing tool, construct the ontology model of the initial knowledge graph according to the knowledge of the electromagnetic field; the ontology model of the initial knowledge graph contains entities and attribute content of electromagnetic signals.

[0077] Preferred, through The ontology editing tool integrates expert knowledge in the electromagnetic field and open-source intelligence to define the entity types, data attributes, object attributes, and partial instances of electromagnetic signals in the initial knowledge graph, thereby constructing the ontology model of the initial knowledge graph.

[0078] Step S212: Based on the electromagnetic signals in the electromagnetic signal instances that contain signal intent classification, train the preset sequential deep model to obtain the intent classification neural network.

[0079] Specifically, based on different parameter distributions and modulation methods, a dataset is constructed that includes labels corresponding to different signal intent classifications, forming electromagnetic signal instances containing signal intent classifications. For example, the different signal intent classifications here include three categories: reconnaissance, jamming, and detection.

[0080] Based on the electromagnetic signal, a preset sequential deep model is trained to obtain an intent classification neural network; the preset sequential deep model is a deep learning model with a sequential model structure, and the obtained intent classification neural network is a deep learning model for electromagnetic signal classification.

[0081] Preferably, the deep learning model for electromagnetic signal classification is a Conv1D-LSTM hybrid model. In this model, Conv1D first performs "noise reduction" and "local feature extraction" on the electromagnetic signal to obtain a feature sequence, and then feeds the feature sequence into LSTM to understand its "dynamic process". Through the serial structure of local feature extraction and temporal modeling, this model can efficiently process strong temporal and non-stationary data such as electromagnetic signals.

[0082] Step S213: Input the electromagnetic signal instance into the intent classification neural network to obtain the signal intent classification of the electromagnetic signal instance.

[0083] Specifically, electromagnetic signal instances that have not undergone signal intent classification are input into the trained intent classification neural network to achieve signal-intent classification of electromagnetic signal instances.

[0084] Step S214: Based on the signal intent classification of electromagnetic signal instances, and the entity and attribute content of electromagnetic signals, construct the initial triples of the initial knowledge graph.

[0085] Specifically, the signal intent classification of electromagnetic signal instances obtained through intent classification neural networks is combined with the entity and attribute content in the ontology model of the initial knowledge graph to construct knowledge triples corresponding to the signal intents, i.e., initial triples, thus forming the initial knowledge graph. The initial knowledge graph contains the signal intent classifications corresponding to the existing electromagnetic signal instances.

[0086] By using the above method, electromagnetic signals of different signal intent classifications are processed with different parameter distributions and modulation methods to obtain electromagnetic signal instances containing signal intent classifications. Then, a deep learning model is trained to obtain the signal intent classifications corresponding to the electromagnetic signal instances, so as to construct an electromagnetic domain knowledge graph covering electromagnetic signals, radiation sources, tasks, and intents, thereby improving the coherence and traceability of electromagnetic situation understanding.

[0087] In some embodiments, step S220 involves inputting the initial triples into a preset knowledge graph embedding model; and in the knowledge graph embedding model, determining the implicit triples with entity relationships in the initial triples based on preset implicit associations, including:

[0088] Step S221: Convert the initial triples into entity embedding vectors and relation embedding vectors, and input them into the preset knowledge graph embedding model.

[0089] Specifically, the triples in the initial knowledge graph constructed by S4 are converted into entity and relation embedding vectors and input into the PConvKB model. This model introduces a path attention mechanism, comprehensively considering the local and global importance of multi-hop paths between entities, and uses a scoring function to quantify the true confidence of candidate triples. Based on implicit association patterns in existing knowledge (such as the interference relationship between reconnaissance signals and radar targets), the model infers new implicit triples with high confidence and adds them to the knowledge graph, thereby expanding and improving the knowledge graph.

[0090] Step S222: Using a preset knowledge graph embedding model and based on the path attention mechanism, calculate the path scores of multiple entity paths between the entity embedding vectors corresponding to the initial triples; the entity paths contain preset implicit relationships.

[0091] Specifically, the paths of the head and tail entities of the entity embedding vectors corresponding to the initial triples are marked to obtain the marked entity paths; based on a preset nonlinear activation function, the marked entity paths, as well as the global importance weights and local importance weights of the entity paths, are processed to obtain the path scores of multiple entity paths between entity embedding vectors.

[0092] By introducing a path attention mechanism, the local and global importance of multi-hop paths between entities is comprehensively considered, and the true confidence of candidate triples in the initial triplet is quantified using a scoring function.

[0093] Step S223: Determine the hidden triples based on the path score.

[0094] In the method for determining implicit triples, based on the implicit correlation patterns in existing electromagnetic domain knowledge, such as the interference relationship between reconnaissance signals and radar targets, new implicit triples are inferred with high confidence by introducing a knowledge graph embedding model with path attention mechanism, and then added to the knowledge graph to expand and improve the knowledge graph.

[0095] In the above method, a path-aware convolutional knowledge graph reasoning model, namely a knowledge graph embedding model that introduces a path attention mechanism, is used to reason about the entity relationships of the initial triples in the knowledge graph. The knowledge graph embedding model embeds the entity relationship paths in the knowledge graph into multi-dimensional feature tensors, uses convolutional neural networks to extract the implicit features in the paths, and dynamically assigns weights to the entity paths of different reasoning using the path attention mechanism. This enables knowledge reasoning to make full use of complex relationship patterns, determine implicit triples in a dynamic environment, and then realize the knowledge graph update and evolution based on the implicit triples.

[0096] In some embodiments, the knowledge graph construction method based on electromagnetic situational awareness further includes: storing the extended knowledge graph in a preset graph database; mining association rules in the extended knowledge graph through the graph database; determining association rules by association information in the extended knowledge graph that occurs more than a preset frequency; establishing new knowledge triples based on the association rules and the signal intent classification of electromagnetic signal instances; and updating the extended knowledge graph based on the new knowledge triples.

[0097] The process involves expanding the initial knowledge graph and storing it in a pre-defined graph database, such as the Neo4j graph database. Then, using a query language within the graph database, frequent association pattern mining is performed on the expanded knowledge graph to extract recurring high-frequency structures—that is, association rules determined by associations that appear more frequently than a preset frequency.

[0098] Preferably, graph databases are used to mine and expand the rules of association information with dynamic temporal relationships in the knowledge graph. For example, the dynamic temporal relationship between specific signal features corresponding to electromagnetic signals and subsequent behavioral intentions in the knowledge graph.

[0099] Subsequently, the newly discovered association rules are processed by the intent classification neural network on the real-time updated electromagnetic signals to output new intent classification results. Based on the new association rules and the new intent classification results, new knowledge triples are jointly constructed and fed back to update the extended knowledge graph, enabling the extended knowledge graph to have dynamic evolution capabilities and thus reflect changes in the electromagnetic situation in real time.

[0100] Furthermore, visualization tools from graph databases can be used to visually represent the analysis results of new input electromagnetic signals in the knowledge graph in a graphical form, thus aiding decision-making.

[0101] Structurally, the integrated framework of "ontology modeling - feature extraction - knowledge reasoning - behavior analysis" for the knowledge graph construction method based on electromagnetic situation described in the above embodiments has good scalability. It can easily incorporate new electromagnetic signal data sources, new entities, and new association information rules, making it suitable for continuously evolving and complex adversarial scenarios. Furthermore, it can highlight the propagation channel characteristics at different locations, providing effective data support for subsequent deep learning methods to mine propagation channel characteristics and understand the electromagnetic environment. Simultaneously, the semantic expression and visualization of the knowledge graph make the reasoning process and conclusions highly interpretable, enhancing the credibility of the analysis results.

[0102] The present embodiment will be described and explained below through specific examples.

[0103] This specific embodiment provides a knowledge graph construction method that covers the entire process from knowledge ontology modeling and dataset preparation to intelligent reasoning and decision support. Figure 3 This is a flowchart of the electromagnetic target knowledge graph construction method based on deep learning provided in this specific embodiment. (Reference) Figure 3 The method includes the following steps:

[0104] Step S310: Construct the map ontology library.

[0105] Specifically, use The ontology editing tool, based on the knowledge of military experts, constructs a structured ontology model of the electromagnetic target domain (i.e., the ontology model of the initial knowledge graph in the aforementioned embodiments).

[0106] Figure 4 This is a schematic diagram of the electromagnetic domain knowledge graph ontology provided in this specific embodiment. (Refer to...) Figure 4 First, define the entity type and create first subclasses such as signal characteristics, radiation source type, and behavior pattern, then create several subclasses, as follows: Figure 4 The knowledge graph ontology shown in the diagram covers electromagnetic communication in tactical applications, including communication, meteorological environment, and other related content.

[0107] The main attributes of an object include whether a device / signal interferes with another device / signal, whether a signal belongs to another device, whether the device operates in a certain frequency band, and whether a signal / device is detecting another device / signal. The main attributes of a data object include radar cross-section, frequency, inertial navigation time, operating frequency band, and modulation method.

[0108] After completing the construction of entities and attributes in the initial ontology model of the knowledge graph, several instances are added to the current ontology library. The instances include: 1. An active electronically scanned array (ASEA) radar mounted on a fighter jet. This radar is a long-range, fast-scanning, and multi-functional radar system. It is also an airborne radar capable of inter-pulse frequency conversion and rapid scanning, making it difficult for the enemy to detect and locate. It can also use time-division methods for electronic intelligence gathering, jamming, surveillance, or communication, with a detection range exceeding 220km.

[0109] 2. The main combat equipment of electronic warfare aircraft in the Navy and Marine Corps is the Tactical Jamming System (TJS), which is used to conduct electronic attacks on combat targets. The system can launch targeting jamming, dual-frequency jamming, frequency sweeping jamming and noise jamming, etc.

[0110] 3. The hull-mounted sonar detection system of the naval destroyer is a low-frequency sonar (3.5 kHz). The effective range for direct detection is 15-20 km, the effective range for seabed reflection detection is 20-30 km, and the detection range for the second convergence zone is about 120 km. The active mode operates at a frequency of 3500 Hz and the maximum transmission power is 150 kW.

[0111] Will as Figure 4 The completed ontology library serves as the schema layer benchmark for the knowledge graph, constraining the import and inference of all subsequent electromagnetic signal data.

[0112] Step S320: Construct a dataset with signal intent labels based on entities and attributes.

[0113] Specifically, a high-quality labeled dataset is constructed, and a pre-defined sequential deep learning model is trained to obtain an intent classification neural network.

[0114] By defining the operational intent of three types of signals—reconnaissance, jamming, and detection—and classifying these intents as data labels, 2000 I / Q data signals are generated for each type of label. Through different parameter distributions and modulation methods, the three types of signals exhibit significant differences in the time and frequency domains, facilitating the learning of the classification model.

[0115] Figure 5a This is a time-frequency domain waveform diagram corresponding to a reconnaissance signal debugging method provided in this specific embodiment. Figure 5b This is a time-frequency domain waveform diagram corresponding to another detection signal debugging method provided in this specific embodiment. Figure 5c This is a time-frequency domain waveform diagram corresponding to another reconnaissance signal debugging method provided in this specific embodiment. Specifically, the reconnaissance signal includes three scanning modes: 1. Linear scan: (Refer to...) Figure 5aThe first image shows a time-frequency domain waveform obtained after QPSK (Quadrature Phase Shift Keying) modulation of the reconnaissance signal. The frequency linearly changes from (0.1-0.3MHz) to (0.7-0.9MHz), simulating the linear frequency sweep of traditional radar. 2. Sinusoidal Modulation Scan: Reference Figure 5b This is a schematic diagram of the time-frequency domain waveform obtained after complex baseband FM tuning of the reconnaissance signal. Sinusoidal modulation (random modulation depth) is superimposed on a basic linear frequency sweep to simulate nonlinear frequency disturbances in a complex environment. 3. Random Step Scan: Reference Figure 5c This is a schematic diagram of the time-frequency domain waveform obtained after complex baseband LFM tuning of the reconnaissance signal. Interpolation is performed between multiple random frequency points (e.g., 5-15 steps) to simulate frequency hopping reconnaissance or discontinuous frequency sweeping. Normally distributed phase noise (mean 0, standard deviation 0.1) is added to simulate phase jitter in actual signal transmission. The signal is generated in complex exponential form, naturally containing phase and frequency information.

[0116] Reconnaissance signals typically need to cover a wide frequency band, and the three scanning methods cover actual reconnaissance scenarios such as linear, nonlinear, and discrete frequency hopping.

[0117] Figure 6a This is a time-frequency domain waveform diagram corresponding to an interference signal debugging method provided in this specific embodiment. Figure 6b This is a time-frequency domain waveform diagram corresponding to another interference signal debugging method provided in this specific embodiment. Figure 6c This is a time-frequency domain waveform diagram corresponding to another interference signal debugging method provided in this specific embodiment. Specifically, there are three types of interference signals: 1. Pulse interference: (Refer to...) Figure 6a The first image shows a time-frequency domain waveform obtained after orthogonal noise amplitude modulation of the interference signal. A square wave is used to modulate the carrier (frequency 0.4-0.6MHz) to simulate common radar pulse interference (duty cycle 0.1-0.4, frequency 1kHz-10kHz). 2. Continuous Wave Interference (CW): Refer to... Figure 6b The image shows a time-frequency domain waveform obtained after orthogonal noise frequency modulation of the interference signal. Pure carrier wave with added phase noise simulates simple continuous wave suppression interference (such as narrowband noise interference). 3. Sweep Interference: Refer to... Figure 6c This is a schematic diagram of the time-frequency domain waveform obtained after complex baseband frequency sweeping debugging of the interference signal. The carrier frequency is linearly swept with time (direction random ±1) to simulate broadband frequency sweeping interference and cover the target frequency band.

[0118] The carrier frequency is fixed in the middle frequency band (0.4-0.6MHz), which distinguishes it from the high and low frequency bands of the reconnaissance signal, simulating the targeted suppression of the target frequency band by the jamming signal.

[0119] Figure 7a This is a time-frequency domain waveform diagram corresponding to a detection signal debugging method provided in this specific embodiment. Figure 7b This is a time-frequency domain waveform diagram corresponding to another detection signal debugging method provided in this specific embodiment. Figure 7c This is a time-frequency domain waveform diagram corresponding to another detection signal tuning method provided in this specific embodiment. Specifically, the detection signal is divided into three categories: 1. Linear frequency modulation (Linear): (Refer to...) Figure 7a The first image shows a time-frequency domain waveform obtained after complex baseband LFM modulation of the detection signal. A linear frequency modulated (LFM) signal (start / end frequency 0.1-0.3MHz / 0.7-0.9MHz) was generated to simulate the commonly used LFM (Linear Frequency Modulation) signal in radar. 2. Nonlinear Frequency Modulation: Refer to... Figure 7b This is a schematic diagram of the time-frequency domain waveform obtained after performing complex baseband nonlinear frequency modulation on the detection signal. A sinusoidal frequency modulation is superimposed on the linear signal to simulate a nonlinear chirping signal (such as bird radar or complex modulated communication signals). 3. Phase Code: Refer to... Figure 7c This is a schematic diagram of the time-frequency domain waveform obtained after complex baseband single-pulse tuning of the probe signal, using random phase codes (7 / 13 / 31 bits). Phase), simulating phase-coded signals such as BPSK (e.g., GPS signals or digital communication modulation).

[0120] Step S330: Train the Conv1D-LSTM hybrid model intention classification neural network using the signal intention dataset.

[0121] Specifically, using the dataset with signal intent labels constructed in step S320, a signal intent classification neural network (i.e., the intent classification neural network in the aforementioned embodiment) is trained. The neural network constructed here is a deep learning model for electromagnetic signal classification and adopts a sequential model structure.

[0122] Preferably, the model of the signal intent classification neural network starts with a one-dimensional convolutional layer. First, there is a Conv1D layer with 64 filters and a kernel size of 5, which uses the ReLU activation function to extract signal features. Then, a BatchNormalization layer is followed to accelerate training and enhance stability. Next is a second convolutional layer with 128 filters and a kernel size of 5, followed by a Dropout layer (dropout rate set to 0.3) to prevent overfitting.

[0123] Following this are two LSTM layers: the first has 64 units and returns a sequence, while the second has 32 units. Next is a fully connected dense layer with 64 neurons, using the ReLU activation function, followed by a dropout layer (with a dropout rate of 0.4). Finally, there is an output layer, a dense layer with the number of neurons equal to the number of classes, using the softmax activation function to output the probability of each class, thus achieving the task of classifying electromagnetic signals.

[0124] Figure 8 This is a schematic diagram illustrating the accuracy change of the signal intent classification neural network provided in this preferred embodiment during the training process. Figure 9 This is a schematic diagram illustrating the loss change of the signal intent classification neural network provided in this preferred embodiment during the training process. Figure 8 The x-axis represents the epoch, indicating the process of training the model once using all samples in the entire training dataset, and the y-axis represents the model accuracy. Figure 9 The x-axis represents epochs, and the y-axis represents model accuracy; Reference Figure 8 and Figure 9 As can be seen, during the training process, both the training and validation accuracies rose rapidly and eventually converged to close to 1.0, while the loss continued to decrease, indicating that the model learned effectively and was not overfitted.

[0125] Step S340, based on the classified signal meaning Figure 3 Tuples are combined with ontology to form a preliminary knowledge graph.

[0126] Specifically, the signal processing results are integrated with domain knowledge to form an initial knowledge base. For example, an unknown signal "Sig_001" classified as a "reconnaissance signal" can generate an initial triple (Sig_001, belonging to the category, reconnaissance signal).

[0127] Meanwhile, if "operating frequency = 1.2GHz" is extracted from the signal, a triple (Sig_001, operating frequency, 1.2GHz) can be generated. All the triples constitute the data layer of the initial knowledge graph.

[0128] Step S350: Use the PConvKB model to perform reasoning on the knowledge graph to expand and improve it.

[0129] Specifically, the PConvKB neural network inference model is used to uncover hidden deep relationships in the knowledge graph. All triples in the initial knowledge graph constructed in S4 are converted into entity and relation embedding vectors and input into the PConvKB model. Figure 10 This is a schematic diagram of the PConvKB model provided in this specific embodiment. (Reference) Figure 10This model not only considers the initial triples themselves, but also calculates the local and global importance of multi-hop paths between entities through an attention mechanism. The path between the head entity and the tail entity is labeled as: , among which, entity path It consists of a series of relationships between entities. and Let $\mathbf{i}$ and $\mathbf{i}$ represent the global and local importance weights of the $i$-th path, respectively, where $i$ ranges from $1$ to $N$. For each triple ${C-class, P'-attribute, R-attribute value constraint}$, the PConvKB model represents entities and relations using k-dimensional vectors, thus each triple can be represented as a matrix:

[0130] ;

[0131] Subsequently, each triple is sequentially input into the convolutional layer and the fully connected layer for processing, resulting in a different scalar score for each triple, which is the score function of PConvKB.

[0132] For each triplet, the scoring function of PConvKB for:

[0133] ;

[0134] in, The sigmoid function represents a non-linear activation function. This indicates the average pooling operation. and Let $\mathbf$ and $\mathbf$ represent the global and local importance weights of the $i$-th path, respectively. This represents the path vector of the i-th entity. and This represents the network parameters of the model. These represent the k-dimensional embedding vectors of the head entity, relation, and tail entity, respectively.

[0135] The scoring function described above quantifies the probability that a candidate triple in the knowledge graph is "true". Figure 11 This is a schematic diagram of the knowledge graph reasoning path and confidence propagation provided in this specific embodiment, such as... Figure 11 As shown, the signal is first observed through a narrow-width continuous wave frequency jump, at which point the confidence level is 1.0.

[0136] Subsequently, a knowledge graph is used for assisted reasoning, specifically through triplet matching to achieve reasoning from signal feature nodes to behavior nodes to intent nodes. Feature extraction is performed on the observed signal, where the modulation type is FM and the bandwidth is 2MHz, with a confidence level of 0.95. Based on the signal features, the observed signal is classified as interference behavior, with a confidence level of 0.91. Furthermore, a knowledge graph mapping is performed on the observed signal, and intent inference is performed, resulting in the observed signal being classified as suppression interference / spoofing interference, with a confidence level of 0.87.

[0137] Finally, based on this intention, a threat assessment was performed on the observed signal, and the threat level of the observed signal was determined to be high, with a confidence level of 0.85.

[0138] For example, the PConvKB model can infer implicit triples (Sig_001, possibly interfering, AN / APG-77 radar) with high confidence based on a large number of known path patterns such as "a certain reconnaissance signal usually interferes with a certain type of radar". Adding such high-scoring new triples to the knowledge graph significantly expands the breadth and depth of knowledge.

[0139] Step S360: Perform graph query and association rule mining on the knowledge graph.

[0140] Specifically, the Neo4j graph database is used to store attribute layers, efficiently managing entities, attributes, and binary relationships, and enabling efficient neighborhood queries and path traversal. Neo4j utilizes Cypher, the native declarative query language of graph databases, to mine frequently occurring association rules. For example, a query reveals that when a reconnaissance intent signal appears and operates in the X band, there is an 80% probability that an interference intent signal will subsequently appear.

[0141] Step S370: Obtain new knowledge triples, update and improve the knowledge graph.

[0142] Specifically, the newly identified association rules and the new classification results generated after continuously inputting signals to the intent classification neural network model trained in step S330 are used as new knowledge triples to feed back and update the knowledge graph, enabling it to reflect the evolution of the battlefield situation in real time.

[0143] Step S380: Use Neo4j to store the knowledge graph in the form of a graph database.

[0144] Specifically, using the graph database Neo4j, the entities (nodes), relationships (edges), and their attributes in the updated and improved knowledge graph are rendered according to the analysis topic. This facilitates subsequent real-time graph queries, incremental updates, and graph analysis, and provides powerful knowledge service capabilities for upper-layer applications.

[0145] The knowledge graph construction method described above can help improve the adaptability of knowledge graphs to dynamically evolving electromagnetic signals, and further improve the accuracy and efficiency of signal intent analysis of electromagnetic signals.

[0146] It should be noted that the steps shown in the above process or in the flowcharts in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions.

[0147] This embodiment also provides a knowledge graph construction device based on electromagnetic situation, which is used to implement the above embodiments and preferred embodiments, and will not be repeated as described above. The terms "module," "unit," "subunit," etc., used below can refer to combinations of software and / or hardware that achieve a predetermined function. Although the device described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0148] Figure 12 This is a structural block diagram of the knowledge graph construction device based on electromagnetic situation provided in the embodiments of this application, such as... Figure 12 As shown, the device includes: a construction module 10, an implicit relationship acquisition module 20, and an extension module 30.

[0149] Module 10 is used to construct the initial triples of the initial knowledge graph based on electromagnetic signal instances; the electromagnetic signal instances include the signal intent classification of the electromagnetic signals.

[0150] The implicit relation acquisition module 20 is used to input the initial triples into a preset knowledge graph embedding model; in the knowledge graph embedding model, the implicit triples with entity relations in the initial triples are determined according to the preset implicit relation; the preset knowledge graph embedding model has a path attention mechanism, which is used to map the graph structure composed of entities and relations in the initial knowledge graph to a continuous low-dimensional vector space; the preset implicit relation includes indirect relations derived from knowledge in the electromagnetic domain.

[0151] Extension module 30 is used to extend the initial knowledge graph based on implicit triples.

[0152] It should be noted that the above modules can be functional modules or program modules, and can be implemented through software or hardware. For modules implemented through hardware, the above modules can reside in the same processor; or the above modules can be located in different processors in any combination.

[0153] This embodiment also provides an electronic device including a memory and a processor, the memory storing a computer program and the processor being configured to run the computer program to perform the steps in any of the above method embodiments.

[0154] Optionally, the electronic device may further include a transmission device and an input / output device, wherein the transmission device is connected to the processor and the input / output device is connected to the processor.

[0155] Optionally, in this embodiment, the processor can be configured to perform the following steps via a computer program:

[0156] S1. Based on electromagnetic signal instances, construct initial triples for the initial knowledge graph; electromagnetic signal instances include the signal intent classification of electromagnetic signals.

[0157] S2, input the initial triples into the preset knowledge graph embedding model; in the knowledge graph embedding model, determine the implicit triples with entity relations in the initial triples according to the preset implicit relationships; the preset knowledge graph embedding model has a path attention mechanism, which is used to map the graph structure composed of entities and relations in the initial knowledge graph to a continuous low-dimensional vector space; the preset implicit relationships include indirect relationships derived from knowledge in the electromagnetic domain.

[0158] S3 expands the initial knowledge graph based on implicit triples.

[0159] It should be noted that the specific examples in this embodiment can refer to the examples described in the above embodiments and optional implementations, and will not be repeated in this embodiment.

[0160] Furthermore, in conjunction with the knowledge graph construction method based on electromagnetic situation provided in the above embodiments, this embodiment can also provide a storage medium for implementation. This storage medium stores a computer program; when executed by a processor, the computer program implements any of the knowledge graph construction methods based on electromagnetic situation described in the above embodiments.

[0161] It should be understood that the specific embodiments described herein are merely illustrative of the application and not intended to limit it. All other embodiments derived by those skilled in the art based on the embodiments provided in this application without inventive effort are within the scope of protection of this application.

[0162] Obviously, the accompanying drawings are merely some examples or embodiments of this application. Those skilled in the art can apply this application to other similar situations based on these drawings without any creative effort. Furthermore, it is understood that although the work done in this development process may be complex and lengthy, for those skilled in the art, certain design, manufacturing, or production modifications made based on the technical content disclosed in this application are merely conventional technical means and should not be considered as insufficient disclosure of this application.

[0163] The term "embodiment" in this application refers to a specific feature, structure, or characteristic described in connection with an embodiment that may be included in at least one embodiment of this application. The appearance of this phrase in various places in the specification does not necessarily imply the same embodiment, nor does it imply that it is mutually exclusive with or independent of other embodiments. It will be clearly or implicitly understood by those skilled in the art that the embodiments described in this application may be combined with other embodiments without conflict.

[0164] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of patent protection. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the appended claims.

Claims

1. A method for constructing a knowledge graph based on electromagnetic situation, characterized in that, The method includes: Based on electromagnetic signal instances, initial triples are constructed for an initial knowledge graph; the electromagnetic signal instances include the signal intent classification of the electromagnetic signals. The initial triples are input into a preset knowledge graph embedding model; in the knowledge graph embedding model, implicit triples with entity relationships are determined from the initial triples according to preset implicit relationships; the preset knowledge graph embedding model has a path attention mechanism to map the graph structure composed of entities and relationships in the initial knowledge graph to a continuous low-dimensional vector space; the preset implicit relationships include indirect relationships derived from knowledge in the electromagnetic domain; The initial knowledge graph is expanded based on the hidden triples.

2. The knowledge graph construction method based on electromagnetic situation as described in claim 1, characterized in that, The construction of initial triples for the initial knowledge graph based on electromagnetic signal instances includes: Based on a preset ontology editing tool, an ontology model of an initial knowledge graph is constructed according to knowledge in the electromagnetic domain; the ontology model of the initial knowledge graph contains the entity and attribute content of the electromagnetic signal; Based on the electromagnetic signals containing signal intent classification in the electromagnetic signal examples, a preset sequential deep model is trained to obtain an intent classification neural network. The electromagnetic signal instance is input into the intent classification neural network to obtain the signal intent classification of the electromagnetic signal instance; Based on the signal intent classification of the electromagnetic signal instance, and the entity and attribute content of the electromagnetic signal, the initial triples of the initial knowledge graph are constructed.

3. The knowledge graph construction method based on electromagnetic situation according to claim 2, characterized in that, The initial triplet is input into a preset knowledge graph embedding model; In the knowledge graph embedding model, based on preset implicit relationships, the implicit triples with entity relationships in the initial triples are determined, including: The initial triples are converted into entity embedding vectors and relation embedding vectors, and then input into the preset knowledge graph embedding model. Using the preset knowledge graph embedding model and based on the path attention mechanism, the path scores of multiple entity paths between the entity embedding vectors corresponding to the initial triples are calculated; the entity paths contain the preset implicit associations. The hidden triples are determined based on the path scores.

4. The knowledge graph construction method based on electromagnetic situation according to claim 3, characterized in that, The step involves calculating path scores for multiple entity paths between entity embedding vectors corresponding to the initial triples using the preset knowledge graph embedding model and based on a path attention mechanism. This includes: The paths of the head and tail entities of the entity embedding vector corresponding to the initial triple are marked to obtain the marked entity paths; Based on a preset nonlinear activation function, the labeled entity paths, as well as the global importance weights and local importance weights of the entity paths, are processed to obtain the path scores of multiple entity paths between the entity embedding vectors.

5. The knowledge graph construction method based on electromagnetic situation according to any one of claims 2 to 4, characterized in that, The expansion of the initial knowledge graph based on the hidden triples includes: The entities, attributes, and signal intents corresponding to the implicit triples are classified and added to the initial knowledge graph to obtain an extended knowledge graph.

6. The knowledge graph construction method based on electromagnetic situation according to claim 5, characterized in that, The method further includes: The extended knowledge graph is stored in a preset graph database, and the association information rules in the extended knowledge graph are mined through the graph database; the association information rules are determined by association information in the extended knowledge graph that appears more frequently than a preset frequency. Based on the association information rules and the signal intent classification of the electromagnetic signal instances, a new knowledge triplet is established; The extended knowledge graph is updated based on the new knowledge triples.

7. The knowledge graph construction method based on electromagnetic situation as described in claim 6, characterized in that, The step of mining the association information rules in the extended knowledge graph through the graph database includes: By using the graph database, we can mine the association rules with dynamic temporal relationships in the extended knowledge graph.

8. A knowledge graph construction device based on electromagnetic situation, characterized in that, The device includes: a construction module, a hidden relationship acquisition module, and an extension module; The construction module is used to construct initial triples of the initial knowledge graph based on electromagnetic signal instances; the electromagnetic signal instances include the signal intent classification of the electromagnetic signals. The implicit relation acquisition module is used to input the initial triples into a preset knowledge graph embedding model; in the knowledge graph embedding model, according to preset implicit relationships, the module determines the implicit triples with entity relationships in the initial triples; the preset knowledge graph embedding model has a path attention mechanism, used to map the graph structure composed of entities and relationships in the initial knowledge graph to a continuous low-dimensional vector space; the preset implicit relationships include indirect relationships derived based on electromagnetic domain knowledge. The extension module is used to extend the initial knowledge graph based on the hidden triples.

9. An electronic device comprising a memory and a processor, characterized in that, The memory stores a computer program, and the processor is configured to run the computer program to execute the knowledge graph construction method based on electromagnetic situation as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the knowledge graph construction method based on electromagnetic situation as described in any one of claims 1 to 7.

Images