An air combat situation analysis method and system

Abstract

The application discloses an air combat situation analysis method and system. First, the target air combat area situation data is acquired and normalized semantic coding is performed to generate an air combat situation semantic description. Subsequently, the RAG technology is used to accurately retrieve prior knowledge from a preset knowledge base, and the prior knowledge and real-time semantic description are fused to construct a situation semantic context, and a tactical intent hypothesis is generated by relying on a preset large language model to fit the actual combat scene. The key link is to use the causal knowledge graph constructed based on air combat rules and historical data to verify the hypothesis in both forward and reverse directions to ensure that the stored hypothesis conforms to the actual causal law. Finally, the preset large language model generates a situation analysis report for the verified intent hypothesis, effectively avoiding the disadvantages of traditional methods relying on experience or deterministic rules, and achieving double improvement of situation analysis rigor and reliability in a complex and dynamic air combat environment.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of artificial intelligence, in particular to an air combat situation analysis method and system. BACKGROUND

[0002] At present, the air combat situation analysis scheme mainly relies on artificial analysis of expert knowledge or a system based on specific analysis rules. However, with the development trend of modern air combat environment towards complexity and dynamicity, whether based on expert knowledge or through specific analysis rules, both of them rely on deterministic analysis logic and rules. When new threats or new tactical strategies appear in the air combat environment, the above two situation analysis methods cannot quickly adapt to the changes of the air combat environment and provide accurate air combat situation analysis data, so the current air combat situation analysis has the problem of poor situation analysis reliability. SUMMARY

[0003] Based on the above problems, in order to improve the reliability of air combat situation analysis, the embodiments of the present application provide an air combat situation analysis method and system.

[0004] The embodiments of the present application disclose the following technical solutions:

[0005] In a first aspect, the embodiments of the present application provide an air combat situation analysis method, comprising:

[0006] Obtaining situation representation data of a target air combat area, and performing semantic coding on the situation representation data to generate an air combat situation semantic description; the target air combat area includes at least one hostile target;

[0007] Taking the air combat situation semantic description as a retrieval input of a preset situation analysis knowledge base, determining a first target priori knowledge from the preset situation analysis knowledge base through RAG technology; the RAG technology is a retrieval enhancement generation technology;

[0008] Based on the first target priori knowledge and the air combat situation semantic description, generating a situation semantic context, and performing air combat situation analysis according to the situation semantic context and a preset large language model to generate at least one tactical intention hypothesis for the hostile target in the target air combat area;

[0009] According to a preset tactical causal knowledge graph, performing forward causal traversal and reverse causal exclusion on the tactical intention hypothesis to determine an intention verification result for the tactical intention hypothesis; the intention verification result is used to represent whether the tactical intention hypothesis conforms to causal rationality; the preset tactical causal knowledge graph is determined based on an air combat tactical rule base and historical air combat confrontation data;

[0010] In a case where the intention verification result is verified, a situation analysis report for the target air combat area is generated according to the tactical intention hypothesis and the preset large language model.

[0011] In a possible implementation, the forward causal traversal and the reverse causal exclusion are performed on the tactical intention hypothesis according to the preset tactical causal knowledge graph to determine an intention verification result for the tactical intention hypothesis, including:

[0012] Based on the preset tactical causal knowledge graph, a first target causal path set with the tactical intention hypothesis as a terminal node is extracted, wherein the first target causal path is composed of a series connection relationship among a tactical element node, a tactical action node and a tactical intention node;

[0013] For each first target causal path in the first target causal path set, the tactical element node in each path is extracted to generate a necessary causal condition set composed of a plurality of tactical element nodes;

[0014] The forward causal traversal is performed according to the situation representation data and the necessary causal condition set to determine a forward causal traversal result;

[0015] In a case where the forward causal traversal result is not verified, the forward causal traversal result is determined as the intention verification result;

[0016] In a case where the forward causal traversal result is verified, the first target causal path set is subjected to reverse causal exclusion based on the tactical action node in each first target causal path and the situation representation data to determine a second target causal path set;

[0017] The intention verification result is determined according to the situation representation data and the second target causal path set.

[0018] In a possible implementation, the forward causal traversal is performed according to the situation representation data and the necessary causal condition set to determine a forward causal traversal result, including:

[0019] The situation representation data is subjected to matching verification with all the tactical element nodes in the necessary causal condition set to determine whether the situation representation data completely covers all the tactical element nodes;

[0020] In a case where any tactical element node is determined not to be completely covered by the situation representation data, the forward causal traversal result is determined as not verified;

[0021] In a case where all the tactical element nodes in the necessary causal condition set are completely covered by the situation characterization data, it is determined that the forward causal traversal result is verified.

[0022] In a possible implementation, the matching verification of the situation characterization data and all the tactical element nodes in the necessary causal condition set comprises:

[0023] According to the element type, data format, and verification rule of each tactical element node in the necessary causal condition set, the necessary causal condition set is information-structured to generate a verification rule set;

[0024] Based on the verification rule set, data alignment processing is performed on the situation characterization data to generate a feature vector set and situation data;

[0025] Based on the element type of each tactical element node in the verification rule set, matching verification is performed on each tactical element node and the feature vector set or the situation data set.

[0026] In a possible implementation, the element type of the tactical element node comprises a numerical type node, a Boolean type node, and an enumeration type node.

[0027] The matching verification of each tactical element node and the feature vector set or the situation data set based on the element type of each tactical element node in the verification rule set comprises:

[0028] In a case where the element type is the numerical type node, a numerical value interval corresponding to the tactical element node in the feature vector set is determined, and it is determined whether the numerical value represented by the tactical element node falls within the numerical interval;

[0029] In a case where the element type is the Boolean type node or the enumeration type node, it is determined whether there is a signal feature matching the tactical element node in the situation data set.

[0030] In a possible implementation, the reverse causal exclusion of the first target causal path set based on the tactical action node and the situation characterization data in each first target causal path comprises:

[0031] For the tactical action node in each first target causal path, it is determined whether the execution order of the tactical action node conforms to the tactical action execution order specified in the air combat rule library;

[0032] The execution order of the tactical action node is pruned from a first target causal path that does not conform to the air combat rule base to determine the second target causal path set.

[0033] In a possible implementation, the method further includes:

[0034] In response to a situation analysis request, determining, according to query content in the situation analysis request and the situation characterization data, a second target priori knowledge from the preset situation analysis knowledge base by the RAG technology;

[0035] Embedding the second target priori knowledge and the query content into a preset prompt template to generate a model optimization prompt; the model optimization prompt is used to optimize an output result of the preset large language model.

[0036] In a possible implementation, the situation analysis report includes at least one of a situation threat level assessment, a tactical intention inference, a threat ranking and target identification, a counter-tactical suggestion, and a natural language explanation of an inference process; and the preset situation analysis knowledge base includes at least one of a weapon system library, a tactical rule base, an air combat database, a threat assessment and target database, and a sensor and electronic equipment library.

[0037] In a possible implementation, after the situation analysis report for the target air combat area is generated, the method further includes:

[0038] Based on the situation analysis report, visual information processing is performed to generate visual situation analysis data, and a plurality of feedback information options are output through a visual interactive interface;

[0039] According to the feedback information option determined through the visual interactive interface, model optimization processing is performed on the preset large language model.

[0040] In a second aspect, an embodiment of the present application provides an air combat situation analysis system, including:

[0041] An acquisition module is configured to acquire situation characterization data of a target air combat area, and perform semantic coding on the situation characterization data to generate a conforming air combat situation semantic description; the target air combat area includes at least one hostile target.

[0042] A first determination module is configured to take the air combat situation semantic description as a retrieval input of a preset situation analysis knowledge base, and determine a first target priori knowledge from the preset situation analysis knowledge base by a RAG technology; the RAG technology is a retrieval enhancement generation technology.

[0043] The first generation module is configured to generate a situation semantic context based on the first target prior knowledge and the air combat situation semantic description, perform air combat situation analysis according to the situation semantic context and a preset large language model, and generate at least one tactical intention hypothesis for the hostile target in the target air combat area.

[0044] The second determination module is configured to perform forward causal traversal and reverse causal exclusion on the tactical intention hypothesis according to a preset tactical causal knowledge graph to determine an intention verification result for the tactical intention hypothesis; the intention verification result is used to represent whether the tactical intention hypothesis is in accordance with causal rationality; and the preset tactical causal knowledge graph is determined based on an air combat tactical rule library and historical air combat confrontation data.

[0045] The second generation module is configured to generate a situation analysis report for the target air combat area according to the tactical intention hypothesis and the preset large language model when the intention verification result is verified.

[0046] Compared with the prior art, the present application has the following beneficial effects:

[0047] The embodiment of the application provides a kind of air combat situation analysis method and system, in its method, the situation characterization data of target air combat area is acquired and is carried out semantic coding, generates air combat situation semantic description, ensures the normativity and semantic accuracy of original data, and lays reliable data foundation for subsequent analysis. Then, the first target priori knowledge is retrieved from the preset situation analysis knowledge base by RAG technology, which combines retrieval enhancement generation technology, can rely on the data in the preset situation analysis knowledge base, avoid the limitation brought by single dependence on large language model, make the analysis process have solid knowledge support. Subsequently, situation semantic context is generated based on priori knowledge and semantic description, and tactical intent hypothesis is generated by means of preset large language model, which makes the generation of intent hypothesis more in line with actual air combat scene by fusing priori knowledge and real-time situation data. After that, the preset tactical causal knowledge graph is used to perform forward causal traversal and reverse causal exclusion on the tactical intent hypothesis, which is constructed based on air combat tactical rule base and historical confrontation data, and can verify the rationality of intent hypothesis from the causal logic level. The dual verification mechanism of forward causal traversal and reverse causal exclusion can exclude unreasonable tactical intent hypothesis from the logic level, to ensure that the remaining hypothesis conforms to the causal logic and actual combat rules in air combat field. Finally, only the hypothesis that passes the verification is used to generate situation analysis report, so that the report content is based on reliable hypothesis verified by multiple layers, avoids the limitation of traditional method relying on deterministic rules or expert experience, can improve the rigor and reliability of situation analysis in complex and dynamic air combat environment through data semantic coding, knowledge retrieval enhancement, causal logic verification and other multi-link cooperation, provides more reliable basis for air combat situation analysis, and then achieves the effect of improving the reliability of air combat situation analysis. BRIEF DESCRIPTION OF DRAWINGS

[0048] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed to be used in the embodiments or prior art description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and those skilled in the art can obtain other drawings according to these drawings without creating any inventive labor.

[0049] Figure 1 A flowchart of an air combat situation analysis method is provided for the embodiments of the present application.

[0050] Figure 2 A flowchart of a tactical intent hypothesis verification is provided for the embodiments of the present application.

[0051] Figure 3 A flowchart of a matching verification is provided for the embodiments of the present application.

[0052] Figure 4A structural schematic diagram of an air combat situation analysis system provided by an embodiment of the present application. DETAILED DESCRIPTION

[0053] As described above, the current air combat situation analysis scheme mainly relies on manual analysis based on expert knowledge or a system based on specific analysis rules. However, with the development trend of modern air combat environment tending to be complex and dynamic, whether the air combat situation analysis is based on expert knowledge or through specific analysis rules, both of them rely on deterministic analysis logic and rules. When new threats or new tactical strategies appear in the air combat environment, the above two situation analysis methods cannot quickly adapt to the changes in the air combat environment and provide accurate air combat situation analysis data, so the current air combat situation analysis has the problem of poor situation analysis reliability.

[0054] To solve the above problem, the embodiment of the present application provides an air combat situation analysis method and system. In the method, the situation representation data of the target air combat area is obtained and semantically encoded, an air combat situation semantic description is generated, ensuring the standardization and semantic accuracy of the original data, and laying a reliable data foundation for subsequent analysis. Then, the first target prior knowledge is retrieved from the preset situation analysis knowledge base through the RAG technology. This technology combines retrieval enhancement generation technology and can rely on the data in the preset situation analysis knowledge base to avoid the limitations brought by single dependence on large language models, so that the analysis process has solid knowledge support. Subsequently, the situation semantic context is generated based on the prior knowledge and the semantic description, and the tactical intention hypothesis is generated by means of the preset large language model. This way of fusing prior knowledge and real-time situation data makes the generation of intention hypothesis more in line with the actual air combat scene. After that, the tactical intention hypothesis is subjected to forward causal traversal and reverse causal exclusion by using the preset tactical causal knowledge graph. This knowledge graph is constructed based on the air combat tactical rule base and historical confrontation data, and can verify the rationality of the intention hypothesis from the causal logic level. The dual verification mechanism of forward causal traversal and reverse causal exclusion can exclude unreasonable tactical intention hypothesis from the logic level, ensuring that the remaining hypothesis conforms to the causal logic and actual combat rules in the air combat field. Finally, only the verified hypothesis is used to generate a situation analysis report, so that the report content is based on reliable hypotheses verified by multiple layers, avoiding the limitations of traditional methods relying on deterministic rules or expert experience. In the complex and dynamic air combat environment, through the cooperation of data semantic encoding, knowledge retrieval enhancement, causal logic verification and other multiple links, the rigor and credibility of situation analysis are improved, a more reliable basis for air combat situation analysis is provided, and the effect of improving the reliability of air combat situation analysis is achieved.

[0055] In the following, the technical solutions in the embodiments of the present application will be described clearly and completely in conjunction with the drawings in the embodiments of the present application, so that those skilled in the art can better understand the present application. Obviously, the described embodiments are only some of the embodiments of the present application, but not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work fall within the scope of the present application.

[0056] Referring to Figure 1 The figure provides a flowchart of an air combat situation analysis method according to an embodiment of the present application, which specifically includes the following steps:

[0057] S101: Obtain situation representation data of a target air combat area, and perform semantic coding on the situation representation data to generate an air combat situation semantic description; the target air combat area includes at least one hostile target.

[0058] The situation representation data is the core information basis supporting the entire air combat situation analysis. The essence of the situation representation data is the structured information formed by multi-source heterogeneous sensors after collection and preprocessing. These data are not simply piled up from a single source, but cover observation results of multiple platforms such as air, ground, and satellite. Specifically, the situation representation data includes at least one of the following: motion parameters such as position, speed, heading, and height of the enemy target, target attribute information such as radar reflection characteristics and infrared radiation signals, and environmental elements such as weather conditions, electromagnetic environment, and airspace rules. For example, target track data from airborne radar, enemy radar signal characteristics captured by electronic support measurement systems, visual image information obtained by optical / electronic sensors, and friendly position, early warning aircraft intelligence, and the like transmitted in data link communication all belong to the category of situation representation data. These data have significant heterogeneity, which is reflected in the diversity of data formats (such as binary detection signals, text instructions, image pixel matrix), time stamp, and spatial coordinate system. Therefore, a unified structured representation needs to be formed through preprocessing to provide effective input for subsequent semantic coding.

[0059] In the process of acquiring situation characterization data, multiple types of front-end equipment such as radar detection, electronic support measurement, and photoelectric / infrared sensors are integrated, and real-time access to collaborative information from early warning aircraft, ground radar stations, and satellites is achieved through high-speed bus or wireless link to ensure multi-dimensional battlefield situation awareness. For example, when our aircraft enters the task area, the airborne sensor captures the raw echo signal of the enemy target in real time, and at the same time, through the data link, the early warning aircraft obtains the collaborative tracking data of the target, forming a multi-source cross-verification raw data set. Secondly, in the data preprocessing link of the raw data set, abnormal values (such as sensor noise and interference caused false alarms) need to be removed through data cleaning, and the out-of-sync sensor data is unified to the same time reference through a time and space alignment algorithm, the target position in different coordinate systems is converted to a unified global battlefield coordinate system through a coordinate conversion algorithm, and the target correlation and trajectory fusion technology is used to eliminate repeated tracking, forming a unique and continuous target track, so as to convert the scattered sensor data into an ordered target motion track and attribute feature set.

[0060] Finally, in the process of semantic encoding of situation characterization data, specific tactical domain knowledge in the air combat scene needs to be injected, and simple numerical values, state parameters, and air combat rules, weapon performance, threat level, etc. are associated to establish a correlation, giving the situation characterization data a causal logic and threat meaning in the air combat scene, so that physical quantities such as speed and heading are upgraded to semantic descriptions with decision-making value such as attack preparation situation and evasion maneuver tendency. Finally, the generated semantic description not only retains the original data accuracy, but also builds a context environment containing tactical intent direction, providing direct support for generating tactical hypotheses, verifying causal reasonableness, and driving large language model reasoning in air combat situation analysis, thereby shortening the cognitive and logical distance between data acquisition and tactical decision-making to ensure that the acquired situation characterization data can effectively serve the air combat situation analysis of the target air combat area.

[0061] S102: input the air combat situation semantic description as a preset situation analysis knowledge base, and determine the first target prior knowledge from the preset situation analysis knowledge base through RAG technology; the RAG technology is a retrieval enhancement generation technology.

[0062] The preset situation analysis knowledge base is a complex collection of multi-dimensional air combat field knowledge, covering weapon system library (such as missile range, radar detection range), tactical rule library (such as different aircraft cooperative tactics, attack and defense action execution logic), air combat database (such as tactical evolution process in typical confrontation scenarios), threat assessment and target database (such as target threat level determination rules), and sensor and electronic equipment library, etc. Heterogeneous knowledge forms, these knowledge are stored in structured, semi-structured, unstructured and other forms, as basic data to support situation analysis.

[0063] When the air combat situation semantic description carrying real-time battlefield information is used as the retrieval input of the preset situation analysis knowledge base, the semantic description is deeply analyzed by the RAG (Retrieval-Augmented Generation) technology, the input text is converted into a high-dimensional vector that can be used for similarity calculation with the knowledge base by using the vector embedding technology in natural language processing, and at the same time, the shallow retrieval technologies such as keyword matching and entity recognition are combined to accurately locate the knowledge subset in the knowledge base that is highly related to the current situation. For example, when the air combat situation semantic description contains the text "the enemy target releases a chaff jamming and performs a large-angle dive", the retrieval process will capture the tactical use of "chaff jamming" and the corresponding tactical rule base of "large-angle dive" to determine the tactical rule corresponding to the tactical action, as well as the associated knowledge such as the typical countermeasures in historical cases. In the retrieval process, the "retrieval-generation" interactive mechanism unique to the RAG technology plays a key role. Its retrieval process not only returns directly matched knowledge items, but also generates an extended knowledge network based on the associated relationships in the knowledge base (such as associating a single tactical action with weapon performance limitations, battlefield environment constraints, and enemy aircraft characteristics), forming a knowledge set containing direct factual knowledge and indirect reasoning logic. Finally, through a confidence evaluation algorithm, the candidate knowledge is screened to determine the first target prior knowledge that best fits the current air combat situation semantic description. These knowledge includes not only direct reference factual data (such as the minimum launch height of a certain type of missile), but also tactical rules that need to be further reasoned with real-time data (such as the condition triggering rule "when the enemy target enters a certain area, be alert to the cluster attack"). This process deeply couples real-time situation data with domain knowledge, ensuring that the subsequent generated tactical intent hypothesis is based on reliable knowledge that has been verified, injecting domain expertise and logical rigor into air combat situation analysis, and effectively improving the accuracy and reliability of air combat situation analysis in complex battlefield environments.

[0064] S103: Based on the first target prior knowledge and the air combat situation semantic description, a situation semantic context is generated, and air combat situation analysis is performed according to the situation semantic context and a preset large language model, to generate at least one tactical intent hypothesis for the enemy target in the target air combat area.

[0065] The essence of this step is to deeply integrate the professional knowledge in air combat field with real-time battlefield situation, and then optimize the analysis logic of the preset large language model for air combat situation, so as to improve the reliability of air combat situation analysis. First, the construction of situation semantic context needs to integrate the first target prior knowledge obtained through RAG technology, such as weapon system performance parameters, standard tactical process, typical tactical patterns summarized in historical confrontation, and the previously generated air combat situation semantic description, to form a comprehensive information body combining background knowledge and real-time data. It needs to be noted that the process of integrating prior knowledge and semantic description is to match the tactical rules and causal logic in prior knowledge with specific parameters in real-time situation through semantic association algorithm, for example, to compare the current height, speed and heading data of the enemy aircraft in the semantic description with the required condition parameters of "mid-range missile attack preparation" in the prior knowledge, and to clearly mark the possible triggering tactical scene boundary in the formed situation semantic context. The situation semantic context formed in this way not only contains factual description of the battlefield (such as "the enemy F-16 aircraft is located 50 kilometers away from me at 3 o'clock direction, with a speed of 1.8 Mach, and the radar signal characteristics match AN / APG-66 fire control radar"), but also integrates the potential meaning given by the domain knowledge (such as "this model has the ability to launch AIM-120 missiles at the current speed, and in historical similar scenarios, 60% probability will enter the attack route").

[0066] After receiving the situation semantic context, the preset large language model uses its deep neural network structure to perform multi-layer semantic analysis and logical deduction on the information. The model first identifies key elements in the context such as target type, maneuver characteristics, equipment status, environmental constraints, etc., and then infers the potential relationship between these elements and the possible tactical purpose based on the air combat field logical relationships learned during the training process. For example, when observing that the enemy target is releasing a chaff jammer while maneuvering towards the direction of the air defense array, the model will combine rules such as "chaff jammer is often used in conjunction with penetration tactics" and "air defense array direction maneuver may indicate suppressive attack" in prior knowledge to generate the tactical intent hypothesis "enemy may implement air defense suppression tactics". Due to the complexity of air combat scenarios, the model will consider multiple possible tactical logic paths and generate multiple different tactical intent hypotheses, covering attack, reconnaissance, evasion, coordination and even innovative inference of new tactics. These hypotheses are not simply probability outputs, but logical deduction results based on context information, each hypothesis contains preliminary judgments on target action motivation, tactical cooperation possibility, and potential threat level. Through this process, fragmented battlefield data and domain knowledge are transformed into intent inferences with tactical guidance significance, providing multi-dimensional analysis perspectives for air combat decision-making.

[0067] S104: According to the preset tactical causal knowledge graph, forward causal traversal and reverse causal exclusion are performed on the tactical intention hypothesis to determine an intention verification result for the tactical intention hypothesis; the intention verification result is used to represent whether the tactical intention hypothesis conforms to causal rationality; and the preset tactical causal knowledge graph is determined based on an air combat tactical rule library and historical air combat confrontation data.

[0068] The process of performing forward causal traversal and reverse causal exclusion on the tactical intention hypothesis according to the preset tactical causal knowledge graph to determine the intention verification result is essentially a process of performing causal logic verification on the intention verification result by means of a causal logic network in the air combat field, to ensure that the tactical intention hypothesis conforms to objective laws and historical experience in actual combat. The preset tactical causal knowledge graph is constructed based on an air combat tactical rule library and historical confrontation data, and includes multiple nodes and edges formed by connections between the nodes. The nodes cover tactical elements, tactical actions, and tactical intentions. Figure Three The core entities include: the tactical element node records the physical state of the target, such as observable data such as speed, height, equipment parameters, and environmental conditions; the tactical action node represents the dynamic behavior of the target, such as identifiable actions such as maneuvering actions, weapon operations, and electronic countermeasure measures; the tactical intention node corresponds to an abstract tactical purpose, such as intention categories such as attack, reconnaissance, evasion, and coordination; and the edge represents a causal relationship between elements that has been verified in actual combat, such as “radar turned on” pointing to “target search”, “entering a missile non-escape zone” necessarily associated with “attack initiation condition mature”, to form a causal logic network covering the whole process of air combat.

[0069] Next, the process of verifying the tactical intention hypothesis by the preset tactical causal knowledge graph in step S104 to determine the intention verification result will be introduced in combination with specific flow embodiments and drawings.

[0070] Referring to Figure 2 The figure is a flowchart of a tactical intention hypothesis verification provided by an embodiment of the present application, specifically including the following steps:

[0071] S1041: Based on the preset tactical causal knowledge graph, a first target causal path set with the tactical intention hypothesis as a terminal node is extracted; wherein the first target causal path is composed of a series connection relationship of a tactical element node, a tactical action node, and a tactical intention node.

[0072] As known from the foregoing, the nodes in the preset tactical causal knowledge graph cover tactical elements, tactical actions, and tactical intentions. Figure ThreeThe core entity, accordingly, the first target causal path is also a directed chain composed of the three types of nodes. Among them, the starting point of the path is the observable tactical element node, such as "enemy aircraft height 10000 meters, speed 2.0 Mach, located in the radar detection boundary of our side", the middle is the transition through the tactical action node, such as "continuous dive and open fire control radar", and finally converges to the tactical intent assumption as the terminal node, such as "high-speed penetration attack". This path composition design has a double meaning. From a technical point of view, it disassembles the abstract tactical intent into a verifiable "data-action-intent" logical chain, so that each intent assumption can be corresponded to specific battlefield observation data and intermediate action characteristics, avoiding the ambiguity of relying solely on probability inference. For example, the reconnaissance intent must be associated with the observable actions such as route detour, low signal interception processing, and element nodes such as non-attack heading and intermittent sensor activation, forming a clear causal evidence chain.

[0073] Extracting the causal path set with the tactical intent assumption as the terminal is essentially constructing an "intent backtracking evidence network" in the knowledge graph. Specifically, taking the assumed tactical intent as the anchor point, all causal paths that may lead to the intent are retrieved in reverse. For example, "attack intent" may correspond to "radar lock + missile preheating" and "high-speed approach + weapon bay opening", each of which contains a complete causal chain from the initial observation data to the final intent. The significance of these paths lies in providing multi-dimensional evidence support for intent verification. On the one hand, the tactical element nodes and action nodes in each path constitute the necessary conditions for the intent to be established. On the other hand, the cross-validation of multiple paths can enhance the confidence of intent judgment. In this way, the tactical intent assumption is transformed into a causal chain that can be visualized and traced back, solving the problem of explainability of intelligent analysis results and providing a reasoning path based on factual evidence for air combat decision-making, ensuring that each step of the tactical intent assumption can find clear causal basis in the knowledge graph, effectively improving the reliability of situation analysis in complex confrontation environments.

[0074] S1042: For each of the first target causal path set, extract the tactical element node in each path to generate a necessary causal condition set composed of multiple tactical element nodes.

[0075] When traversing each causal path, the tactical element nodes representing objective facts in the path are screened out one by one. For example, in the causal path of "high-altitude and high-speed approach - fire control radar start - anti-ship missile attack preparation", the corresponding height, speed, heading parameters of high-altitude and high-speed approach, and the signal characteristic parameters of fire control radar start are extracted as tactical element nodes. These nodes are the necessary conditions for the calibration of the knowledge graph causal relationship. That is, the absence of any node may cause the causal chain to break, thereby making the tactical intent assumption lose factual support. Through the extraction and aggregation of all target element nodes of the causal path, the final necessary causal condition set is actually a verification index library containing multi-dimensional factual data, and each index corresponds to a key factual basis supporting the establishment of a specific tactical intent.

[0076] The purpose of this step is to materialize the verification standard of the tactical intent into observable and comparable factual conditions. On the one hand, each tactical element node in the set points to specific battlefield observation data (such as "target speed exceeds 1.5 Mach" and "radar signal matches the characteristics of a certain type of fire control radar"), enabling the system to directly verify whether the conditions are met through real-time data collection, avoiding the ambiguity of relying solely on experience. On the other hand, by aggregating the element nodes of multiple causal paths, the necessary causal condition set can cover multiple possible fact combinations supporting the same tactical intent, preserving the flexibility of causal reasoning while ensuring that each intent assumption has a clear factual verification dimension.

[0077] S1043: According to the situation characterization data and the necessary causal condition set, perform forward causal traversal to determine a forward causal traversal result, and determine whether the result passes.

[0078] According to the process of performing forward causal traversal based on the situation characterization data and the necessary causal condition set to determine whether the result passes, the essence is to compare and verify the core factual basis supporting the tactical intent assumption with real-time battlefield data one by one, and provide direct evidence support for the reliability of the intent assumption through condition and data matching analysis. Specifically, this process is achieved through the following three steps:

[0079] Step one, match the situation characterization data with all tactical element nodes in the necessary causal condition set to determine whether the situation characterization data completely covers all tactical element nodes.

[0080] The matching verification between the situation characterization data and the tactical element nodes in the necessary causal condition set is essentially to ensure that each factual basis supporting the tactical intent assumption can find an exact landing point in the real-time battlefield observation through the accurate comparison between the data and the conditions. The situation characterization data, as a digital mirror image of the real-time state of the battlefield, covers multi-dimensional observable information such as the position parameters, motion characteristics, equipment state and environmental parameters of the target, while the tactical element nodes in the necessary causal condition set are the core factual indexes supporting the establishment of a specific tactical intent, which are extracted from the causal path. The execution logic of the matching verification is to convert the tactical element nodes in each causal path into specific data retrieval instructions. The system checks whether there is an observation value completely corresponding to each element node in the situation characterization data, for example, for the element node of “radar signal characteristics match a certain type of fire control radar”, the radar signal data in the current electromagnetic environment is extracted, and whether the frequency, pulse width, modulation mode and other parameters are completely consistent with the characteristics defined by the node. The matching verification is an accurate matching based on the node definition in the knowledge graph, that is, to verify that each element node contains specific attributes such as data type, threshold range and feature mode. Only when the situation data completely meets the node attribute requirements in the corresponding dimension, can it be considered to “cover” the node.

[0081] Specifically, the specific process of matching verification between the situation characterization data and the necessary causal condition set can be seen from Figure 3 The figure is a process schematic diagram of matching verification provided by an embodiment of the present application, which specifically includes the following steps:

[0082] S301: According to the element type, data format and verification rule of each tactical element node in the necessary causal set, the necessary causal condition set is information structured to generate a verification rule set.

[0083] The core of this step is to transform the tactical element nodes in the necessary causal set into a standardized rule framework that can be recognized and executed by a computer. Each tactical element node contains three key attributes: element type, data format, and verification rule. The element type is used to define the battlefield data field to which the node belongs, such as motion characteristics, equipment status, environmental parameters, etc. The data format is used to specify the presentation standard of the node data, such as the floating-point type of Mach number for speed, and the string of specific encoding for radar signals. The verification rule is used to specify the judgment logic for the node to be established, such as threshold comparison and pattern matching. In the structured processing, first, classify the nodes by element type, and aggregate nodes in the same field (such as speed and heading into the motion characteristics category), to facilitate subsequent rapid positioning of data sources. Then, calibrate the data format, and convert heterogeneous data from different sensors into a unified form required by the rules (such as dividing the speed value in kilometers per hour by 1225 to convert it to Mach number, and decoding binary radar signals into strings), to eliminate format ambiguity. Finally, convert the verification rule into an executable logical expression (such as "speed > 1.5" corresponds to numerical comparison logic, and "radar signal matches a certain type of feature" corresponds to regular expression matching logic). After processing, each node is transformed from a natural language description of the condition to a structured rule unit containing field classification, standard format, and execution logic. All units are organized according to the ontology architecture of the knowledge graph, forming a verification rule set. This step provides a clear rule basis for subsequent automated verification, ensuring that the system can interpret tactical conditions according to a unified standard.

[0084] S302: Perform data alignment processing on the situation characterization data based on the verification rule set to generate a feature vector set and situation data.

[0085] This step is the process of matching and adapting real-time situation data with the verification rule set. Situation characterization data is multi-source real-time battlefield information collected by sensors (such as target position, speed, radar signals, etc.). However, these data often have problems such as scattered dimensions, heterogeneous formats, and inconsistent time sequences. The purpose of data alignment processing is to solve these problems. First, according to the element type in the verification rule set, extract the corresponding data fields from the situation characterization data. Then, perform format conversion to transform the extracted data into the standard format required by the rules. Then, process the time consistency according to the time-related requirements in the rules (such as "radar illumination data in the last 30 seconds"), extract the situation data within the corresponding time window, and ensure that the timeliness of the data matches the requirements of the rules. After alignment, the situation data corresponding to each tactical element node is transformed into a feature vector set, and structured situation data is generated. The role of this step is to transform the original scattered situation data into standardized data that can be directly used by the verification rule set, ensuring that the rules can accurately call the corresponding data for logical judgment in the subsequent verification process, avoiding verification errors caused by data mismatch, and laying a data foundation for reliable verification of tactical intent.

[0086] S303: Match each of the tactical element nodes with the feature vector set or the situation data set based on the element type of each of the tactical element nodes in the set of verification rules.

[0087] The core of the steps is to match each node with the corresponding real-time data set according to the type of the tactical element node, and use the adaptive logic to determine whether the node meets the requirements of the tactical rules. First, each tactical element node has a clear type - numerical, Boolean or enumeration, and the type directly determines whether the feature vector set or the situation data set is used for verification.

[0088] Among them, the numerical type node focuses on battlefield elements with continuous numerical characteristics, such as target flight speed, distance from our side, radar signal strength, and other indicators that can be quantified with specific numerical values. The core of such data is the numerical value interval, and the key to verification is to determine whether the real-time data falls within the interval set by the rules. In matching verification, since the feature vector set is a structured numerical dimension after alignment processing, the numerical value interval corresponding to the numerical type node (such as the rule "high-speed maneuver" requiring speed greater than 1.5 Mach, and "close-range early warning" requiring distance less than 40 kilometers) needs to be extracted from the feature vector set first, and then the real-time numerical value represented by the tactical element node (such as the target current speed 1.7 Mach, distance 35 kilometers) is compared with this interval. If the real-time numerical value falls within the interval, it means that the node is covered by the situation data; otherwise, it is not covered. The core of this logic is that the "quantity" of numerical data directly reflects the degree difference of battlefield state - speed exceeding the threshold means that the intention of maneuvering is upgraded, and distance decreasing means that the threat is approaching, and interval judgment can accurately capture this degree of change, ensuring that the verification of numerical elements meets the quantitative requirements of tactical rules.

[0089] Boolean nodes are binary judgments, such as weapon bay door opening, electronic jamming starting, and the result is only yes or no; enumeration nodes are a set of fixed options, such as radar operating mode (search, tracking, fire control), target flight attitude (flat flight, climbing, diving), and the result is limited to pre-defined options. The verification core of these two types of nodes is not "numerical range", but "feature matching". They reflect the "quality" difference of battlefield state, such as the opening of the bay door is the state switching from preparation to standby, and the radar mode is the functional change from reconnaissance to attack. Therefore, the situation data set (i.e. the real-time battlefield signal data after alignment) is used for verification. In the specific verification process, the system will retrieve from the situation data set whether there is a signal feature matching the node, such as the Boolean type "weapon bay door opening", whether the sensor returns the trigger signal of the bay door opening; the enumeration type "radar in fire control mode", whether the coding, pulse width and other characteristics of the radar signal meet the pre-defined template of the fire control mode; the enumeration type "target in climbing attitude", whether the elevation data returned by the attitude sensor meets the characteristics of climbing (such as the elevation angle is greater than 15 degrees and lasts for more than 5 seconds). If there is a completely matching signal feature in the situation data set, it means that the node is covered; otherwise, it is not covered. The logic of this way is that the recognition of discrete state or mode depends on specific signal features, fire control radar has unique signal coding, bay door opening has specific sensor trigger logic, only these features appear, can confirm the existence of the corresponding state or mode, ensure the accuracy and pertinence of verification.

[0090] Step two, in the case where any of the said tactical element nodes is not completely covered by the said situation representation data, determining that the positive causal traversal result is verification failure;

[0091] Step three, in the case where all the said tactical element nodes are completely covered by the said situation representation data, determining that the positive causal traversal result is verification success.

[0092] In this way, if any one of the tactical element nodes is not completely covered by the situation data (i.e. the value corresponding to the node does not fall within the preset interval, or the required signal feature of the Boolean type or enumeration type node does not appear in the situation data), it is directly determined that the positive causal traversal result is not verified; on the contrary, only when all the tactical element nodes are completely covered by the situation data (the real-time value of each numerical type node conforms to the corresponding interval, and the signal feature of all Boolean type and enumeration type nodes is accurately matched in the situation data), it is determined that the positive causal traversal result is verified. This process takes "full node coverage" as the core determination standard. As long as there is an unsatisfied key element node, the effectiveness of the overall tactical intent is denied. It is ensured that only when all the preset tactical conditions are satisfied by the real-time situation data, the tactical intent verification is confirmed to be passed, providing a rigorous logical basis for subsequent decision-making.

[0093] S1044: determining the positive causal traversal result as the intent verification result;

[0094] S1045: in the case where the positive causal traversal result is verified, performing reverse causal exclusion on the first target causal path set based on the tactical action nodes in each of the first target causal paths and the situation representation data, to determine a second target causal path set;

[0095] S1046: determining the intent verification result according to the situation representation data and the second target causal path set.

[0096] If the forward traversal shows that all tactical elements are satisfied by the real-time situation, a preliminary basis for supporting the intent is obtained; if there are uncovered nodes, it is directly concluded that the tactical intent assumption is not established, and the intent assumption result is determined to be verified. Further, under the premise that the forward traversal is verified, the reverse causal exclusion mechanism is started. At this time, although the key elements of the tactical intent are preliminarily confirmed to match the situation, it is still necessary to exclude those paths that do not conform to the actual causal logic. Specifically, the system will analyze whether these paths have logical contradictions or conflicts with actual data based on the tactical action nodes in each first target causal path (i.e. specific action steps constituting the tactical intent, such as force mobilization, weapon operation, etc.) and the real-time situation representation data. For example, a causal path assumes that a specific sensor signal will be accompanied when the enemy implements a certain tactical action, but the signal does not appear in the situation data, or the time sequence of the action does not match the data record. Such a path will be excluded, and finally a second target causal path set that is more consistent with the actual situation is formed.

[0097] Specifically, the process of performing reverse causal exclusion in step S1045 is implemented through the following two steps:

[0098] Step one, for each of the first target causal path, determine whether the execution order of the tactical action node conforms to the execution order of the tactical action specified in the air combat rule base;

[0099] Step two, the execution order of the tactical action node is not in line with the first target causal path of the air combat rule base is eliminated, to determine the second target causal path set.

[0100] In the process of reverse causal exclusion, the goal of step one is to check the execution order of the tactical action. For each of the first target causal path, the execution order of the tactical action node (i.e. the specific action link that constitutes the tactical intent, such as target detection, maneuvering, weapon launching, etc.) needs to be confirmed one by one. The air combat rule base condenses the basic logic and practical experience of air combat tactics, such as "first reconnaissance and then attack" and "first establish effective communication link and then develop coordinated action". These rules essentially define the causal chain of reasonable tactical action, that is, any action sequence that violates the conventional tactical logic, such as launching missiles before completing target locking or engaging in close combat before electronic jamming takes effect, may make the causal path lose practical significance and become a hypothetical scenario that does not conform to the actual combat scene. Therefore, this step is equivalent to installing a "logic filter" for each causal path, which eliminates those paths that violate the basic tactical logic based on the sequence specifications in the air combat rule base.

[0101] Step two is to eliminate the path based on the sequence check. Once it is found that the execution order of the tactical action node in a certain first target causal path conflicts with the air combat rule base (for example, the reconnaissance action occurs after the attack action, or the time sequence of key coordinated action is reversed), this path will be directly eliminated from the set, and the remaining paths collectively constitute the second target causal path set. The significance of this screening mechanism is that even if the forward causal traversal verifies the completeness of the tactical elements, it is also necessary to ensure that the causal path supporting the intent conforms to the tactical action execution sequence in the actual combat logic. The establishment of a tactical intent hypothesis not only needs to ensure that each tactical element "exists", but also needs to ensure that these elements are connected in the correct tactical rhythm and sequence. For example, the "implementing a surprise attack" intent requires that the "hidden enemy" action occur before the "fire attack" action. If the order of these two actions in a certain causal path is reversed, even if other elements are met, this path will be excluded because it violates the basic tactical principles. Through these two links, the causal path is purified from the perspective of "action sequence compliance", ensuring that the path entering the subsequent verification stage is not only complete in terms of tactical elements, but also conforms to the air combat logic, providing a more solid support for the reliability of the final intent verification result.

[0102] Finally, after reverse causal exclusion, the core task of S1046 is to combine the situation representation data with the filtered second target causal path set to finally confirm the intent verification result. The logic of this step is that forward traversal solves the problem of whether the tactical elements "exist", and reverse exclusion solves the problem of whether the causal path is "reasonable". Even if all tactical elements are covered, if the causal path supporting the intent has logical breaks or data contradictions in the actual situation (such as the time sequence of actions and results being reversed, the signal of key actions being missing, etc.), the authenticity of the intent still needs to be reexamined. Therefore, the system will deeply match the second target causal path set with real-time situation data to check whether each link of each reserved path can find a corresponding evidence chain in the data, ensuring that the establishment of the tactical intent not only meets the element coverage condition, but also meets the causal logic in the actual battlefield. Through this series of steps, the intent verification result not only covers the integrity verification of tactical elements, but also includes the rationality screening of causal paths, thereby forming a more accurate and reliable conclusion to provide solid logical support for subsequent tactical decision-making.

[0103] After reverse causal exclusion, the goal of S1046 is to combine the situation representation data with the filtered second target causal path set to finally confirm the intent verification result. The logic of this step is that forward traversal solves the problem of whether the tactical elements "exist", and reverse exclusion solves the problem of whether the causal path is "reasonable". Through this series of steps, the intent verification result not only covers the integrity verification of tactical elements, but also includes the rationality screening of causal paths, thereby forming a more accurate and reliable conclusion to provide solid logical support for subsequent tactical decision-making.

[0104] The above is the introduction of step S104 in the embodiments of the present application. Next, the combination of Figure 1 The last step S105 in the embodiments of the present application will be introduced:

[0105] S105: In the case where the intent verification result is verified, a situation analysis report for the target air combat area is generated according to the tactical intent hypothesis and the preset large language model.

[0106] On the basis of determining that the tactical intention assumption passes, the tactical intention assumption is combined with the preset large language model, and deep information integration and analysis are carried out for the target air combat area. Specifically, the tactical intention assumption provides a core framework for analysis, which clearly indicates which key tactical elements (such as enemy force deployment, weapon use mode, and maneuver trajectory) need to be focused on, and the preset large language model is responsible for converting massive situation representation data, verified causal path information, and professional knowledge in the tactical rule library into natural language text that is easy for humans to understand, and finally forming a complete structure, clear logic situation analysis report.

[0107] Specifically, the situation analysis report includes at least one of situation threat level assessment, tactical intention inference, threat sorting and target identification, countermeasures suggestion, and natural language explanation of reasoning process. Among them, for the situation threat level assessment, the preset large language model will quantify the battlefield threat level according to the real-time data of enemy force deployment, equipment status, etc., combined with air combat rules, to measure the situation threat level of the target air combat area in the form of "high risk", "medium risk", "low risk" and the like. Secondly, it is the tactical intention inference, based on the previously verified causal path, to clearly indicate the specific performance of the current enemy or our tactical intention (such as "the enemy is implementing a split and surround tactic"), and combined with historical cases to supplement the possible development direction of the intention, to provide a logical basis for predicting the situation.

[0108] And the threat sorting and target identification module focuses on the priority division of battlefield targets, and the preset large language model will dynamically sort the targets according to their threat level, tactical value and other factors (such as listing the enemy early warning aircraft as the "primary threat target"), helping the combat unit to quickly lock the core attack or monitoring object; The countermeasures suggestion module directly interfaces with the actual combat demand, extracts the tactical scheme matching the current situation from the rule library, and generates specific action guidance combined with weather, weapon performance and other parameters (such as "suggesting to suppress the enemy radar first, and then launching a formation attack"), which has both pertinence and operability; Finally, the natural language explanation module of the reasoning process restores the logical chain of data verification and rule matching through simple language, so that decision makers can understand the basis behind the conclusion and enhance the credibility of the report. These contents together constitute a complete logical chain from threat assessment to action suggestion, making the report a bridge connecting battlefield data and tactical decision-making, providing clear and practical information support for efficient command.

[0109] In one possible implementation, the embodiments of the present application can also provide more personalized air combat situation analysis according to the situation analysis request input by the user from the interactive terminal, and the method includes the following two steps:

[0110] Step one, in response to the situation analysis request, according to the query content in the situation analysis request and the situation characterization data, the second target prior knowledge is determined from the preset situation analysis knowledge base by the RAG technology.

[0111] When receiving the situation analysis request (i.e. the specific analysis demand put forward by the user or the command system for the target air combat area), the query content in the request is first extracted, and the real-time updated situation characterization data is called synchronously. On this basis, with the help of RAG technology, deep retrieval is carried out from the preset situation analysis knowledge base to filter out the second target prior knowledge most matched with the current analysis demand according to the dual dimensions of query content and situation data (for example, if the query content focuses on "low-altitude attack response", the tactical disposal scheme of the same kind of scene in history, the best interception height range of related weapons, the influence data of weather conditions on interception effect, etc. are retrieved), which will serve as the core supporting materials for subsequent model optimization.

[0112] Step two, the second target prior knowledge and the query content are embedded into the preset prompt word template to generate model optimization prompt words; the model optimization prompt words are used to optimize the output results of the preset large language model.

[0113] This step focuses on converting prior knowledge and query content into model recognizable optimization prompt words: the second target prior knowledge obtained in the first step is organically integrated with the original query content, and information embedding is carried out according to the preset prompt word template. For example, if the template structure is "based on [situation data], combined with [historical cases] and [tactical rules], analyze [query problem] and give [output form] conclusion", the real-time battlefield wind speed, enemy formation height and other situation data will be filled into the "situation data" field, the successful interception scheme of historical low-altitude attack will be filled into the "historical cases" field, and the weapon use restriction in the air combat rule library about close combat will be filled into the "tactical rules" field, and finally the complete model optimization prompt word is generated.

[0114] The core role of this prompt word is to provide more accurate input guidance for the preset large language model. On the one hand, by supplementing the prior knowledge in the professional field, it ensures that the output result conforms to the professional logic of air combat practice. On the other hand, through the explicit format guidance, the answer of the large language model is more in line with the actual decision-making demand. Through these two links, the upgrade from "extensive problem response" to "precise analysis based on professional knowledge enhancement" is realized, which not only solves the fusion problem of real-time data and historical knowledge by using RAG technology, but also converts complex professional information into input structure easy for the large language model to process through the prompt word template, finally improving the accuracy, practicality and decision support value of the situation analysis conclusion.

[0115] In another possible implementation, after generating the situation analysis report, the situation analysis report can also be visualized through a terminal that interacts with external users to present refined situation analysis to the users, while receiving the users' evaluation of this situation analysis, and then optimizing the preset large language model. Specifically, this process is realized through the following two steps:

[0116] Step one, visual information processing based on the situation analysis report to generate visual situation analysis data, and output multiple feedback information options through a visual interactive interface;

[0117] Step two, according to the feedback information options determined through the visual interactive interface, model optimization processing is performed on the preset large language model.

[0118] In the process of visualizing the situation analysis report, the complex information contained in the situation analysis report, such as threat level, tactical intent, target ranking, and response suggestion, needs to be visualized and converted into graphical content, such as dynamic battlefield map, threat heat map (to distinguish the danger level of the area by color depth), data chart, etc., to convert the battlefield situation described in words into intuitive and easy-to-understand graphical content, forming visual situation analysis data. These visual data will be presented through a specially designed interactive interface, which provides multiple feedback information options. These options are closely related to the report content, which may include confirmation of the analysis conclusion (such as "approve the current threat ranking"), further inquiry, and correction suggestion (such as "think that the threat level of a certain target is too high, suggest re-evaluating in combination with weather data"), etc., aiming to enable decision-makers or operators to communicate information with the system through intuitive interaction.

[0119] When the user selects a feedback information option through the visual interactive interface, this selection is used as an important optimization signal to adjust the preset large language model. For example, if the user points out that "the response tactical suggestion does not consider the range limitation of our missiles, resulting in some unfeasible schemes" and selects the corresponding correction feedback, the system will extract the key information (such as the constraint condition "range limitation") contained in this feedback, combine it with the original situation representation data and knowledge base content, form new training samples or adjust the input parameters of the model, so that the large language model can automatically include this type of actual constraint condition emphasized by the user when generating the analysis report in the future, avoiding similar problems from occurring again. In this way, the subsequent generated situation analysis report can better meet the needs of real combat scenarios, realizing the deep collaboration between technical tools and user decisions, and thus ensuring the reliability of air combat situation analysis.

[0120] The embodiment of the application provides a kind of air combat situation analysis method, in its method, the situation characterization data of target air combat area is acquired and is carried out semantic coding, generates air combat situation semantic description, ensure the normativity and semantic accuracy of original data, lay reliable data foundation for subsequent analysis. Then, the first target prior knowledge is retrieved from the preset situation analysis knowledge base by RAG technology, which combines retrieval enhancement generation technology, can rely on the data in the preset situation analysis knowledge base, avoid the limitation brought by single dependence large language model, make the analysis process have solid knowledge support. Subsequently, situation semantic context is generated based on prior knowledge and semantic description, and tactical intent hypothesis is generated by using preset large language model, this kind of fusion prior knowledge and real-time situation data makes the generation of intent hypothesis more in line with actual air combat scene, and after that, the preset tactical causal knowledge graph is used to perform forward causal traversal and reverse causal exclusion on tactical intent hypothesis, which is constructed based on air combat tactical rule base and historical confrontation data, can verify the rationality of intent hypothesis from the level of causal logic, the double verification mechanism of forward causal traversal and reverse causal exclusion can exclude unreasonable tactical intent hypothesis from the logic level, ensure that the remaining hypothesis conforms to the causal logic and actual combat rules in air combat field. Finally, only the assumption that passes the verification generates situation analysis report, so that the report content is based on reliable hypothesis verified by multiple layers, avoids the limitation of traditional method relying on deterministic rules or expert experience, can improve the rigor and reliability of situation analysis in complex dynamic air combat environment through data semantic coding, knowledge retrieval enhancement, causal logic verification and other multi-link cooperation, provide more reliable basis for air combat situation analysis, and then achieve the effect of improving the reliability of air combat situation analysis.

[0121] The embodiment of the application provides an air combat situation analysis system, which can be used for implementing the air combat situation analysis method described above.

[0122] Referring to Figure 4 The figure is a structural schematic diagram of an air combat situation analysis system provided by the embodiment of the application, which specifically includes the following modules:

[0123] The acquisition module 401 is used for acquiring the situation characterization data of the target air combat area, and carrying out semantic coding on the situation characterization data to generate a conforming air combat situation semantic description; the target air combat area includes at least one enemy target;

[0124] The first determination module 402 is used for taking the air combat situation semantic description as the retrieval input of the preset situation analysis knowledge base, and determining the first target prior knowledge from the preset situation analysis knowledge base by RAG technology; the RAG technology is a retrieval enhancement generation technology;

[0125] The first generation module 403 is configured to generate a situation semantic context based on the first target prior knowledge and the air combat situation semantic description, perform air combat situation analysis according to the situation semantic context and a preset large language model, and generate at least one tactical intention hypothesis for the hostile target in the target air combat area.

[0126] The second determination module 404 is configured to perform forward causal traversal and reverse causal exclusion on the tactical intention hypothesis according to a preset tactical causal knowledge graph to determine an intention verification result for the tactical intention hypothesis; the intention verification result is used to represent whether the tactical intention hypothesis is in accordance with causal rationality; and the preset tactical causal knowledge graph is determined based on an air combat tactical rule library and historical air combat confrontation data.

[0127] The second generation module 405 is configured to generate a situation analysis report for the target air combat area according to the tactical intention hypothesis and the preset large language model when the intention verification result is verified to be passed.

[0128] It should be noted that each of the embodiments in the specification is described in a progressive manner, and the same and similar parts between the embodiments can be referred to each other, and each embodiment focuses on the difference from other embodiments. In particular, for the method and system embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and the relevant parts can be referred to the part of the method embodiment. The above-described method and system embodiments are only illustrative, wherein the units described as separate components can be or can not be physically separated, and the components indicated as units can be or can not be physical units, that is, they can be located in one place or distributed on multiple network units. According to actual needs, part or all of the modules can be selected to achieve the purpose of the embodiment. Those skilled in the art can understand and implement it without creative labor.

[0129] The above description is only a specific embodiment of the present application, but the protection scope of the present application is not limited to this. Any skilled person in the art can easily think of changes or replacements within the technical range disclosed in the present application, which should be covered in the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. An air combat situation analysis method, characterized by, The method comprises the following steps: acquiring situation representation data of a target air combat area, and performing semantic coding on the situation representation data to generate an air combat situation semantic description; the target air combat area comprises at least one hostile target; taking the air combat situation semantic description as a retrieval input of a preset situation analysis knowledge base, and determining first target prior knowledge from the preset situation analysis knowledge base through an RAG technology; the RAG technology is a retrieval enhancement generation technology; generating a situation semantic context based on the first target prior knowledge and the air combat situation semantic description, and performing air combat situation analysis according to the situation semantic context and a preset large language model to generate at least one tactical intention hypothesis for the hostile target in the target air combat area; performing forward causal traversal and reverse causal exclusion on the tactical intention hypothesis according to a preset tactical causal knowledge graph to determine an intention verification result for the tactical intention hypothesis; the intention verification result is used to represent whether the tactical intention hypothesis conforms to causal rationality; the preset tactical causal knowledge graph is determined based on an air combat tactical rule base and historical air combat confrontation data; in a case where the intention verification result is verified to be passed, generating a situation analysis report for the target air combat area according to the tactical intention hypothesis and the preset large language model; the method of performing forward causal traversal and reverse causal exclusion on the tactical intention hypothesis according to a preset tactical causal knowledge graph to determine an intention verification result for the tactical intention hypothesis comprises the following steps: extracting a first target causal path set with the tactical intention hypothesis as a terminal node based on the preset tactical causal knowledge graph; wherein a first target causal path is composed of a series connection relationship among a tactical element node, a tactical action node and a tactical intention node; for each first target causal path in the first target causal path set, extracting the tactical element node in each path to generate a necessary causal condition set composed of a plurality of tactical element nodes; performing forward causal traversal according to the situation representation data and the necessary causal condition set to determine a forward causal traversal result; in a case where the forward causal traversal result is verified to be failed, determining the forward causal traversal result as the intention verification result; in a case where the forward causal traversal result is verified to be passed, performing reverse causal exclusion on the first target causal path set based on the tactical action node in each first target causal path and the situation representation data to determine a second target causal path set; determining the intention verification result according to the situation representation data and the second target causal path set.

2. The method of claim 1, wherein, the method of performing forward causal traversal according to the situation representation data and the necessary causal condition set to determine a forward causal traversal result comprises the following steps: performing matching verification on the situation representation data and all tactical element nodes in the necessary causal condition set to determine whether the situation representation data completely covers all tactical element nodes; In a case where it is determined that any of the tactical element nodes is not completely covered by the situation characterization data, it is determined that the forward causal traversal result is verification failure; In a case where it is determined that all of the tactical element nodes are completely covered by the situation characterization data, it is determined that the forward causal traversal result is verification success.

3. The method of claim 2, wherein, The matching verification of the situation characterization data and all of the tactical element nodes in the necessary causal condition set comprises: According to the element type, data format and verification rule of each of the tactical element nodes in the necessary causal set, information structure processing is performed on the necessary causal condition set to generate a verification rule set; Based on the verification rule set, data alignment processing is performed on the situation characterization data to generate a feature vector set and situation data; Based on the element type of each of the tactical element nodes in the verification rule set, matching verification is performed on each of the tactical element nodes and the feature vector set or the situation data.

4. The method of claim 3, wherein, The element type of the tactical element node comprises a numerical type node, a Boolean type node and an enumeration type node; Based on the element type of each of the tactical element nodes in the verification rule set, matching verification is performed on each of the tactical element nodes and the feature vector set or the situation data, comprising: In a case where the element type is the numerical type node, a numerical value interval corresponding to the tactical element node in the feature vector set is determined, and it is determined whether the numerical value represented by the tactical element node falls within the numerical value interval; In a case where the element type is the Boolean type node or the enumeration type node, it is determined whether there is a signal feature matching the tactical element node in the situation data.

5. The method of claim 1, wherein, Based on the tactical action node in each of the first target causal path and the situation characterization data, reverse causal exclusion is performed on the first target causal path set to determine a second target causal path set, comprising: For the tactical action node in each of the first target causal path, it is determined whether the execution order of the tactical action node conforms to the tactical action execution order specified in the air combat tactical rule library; The first target causal path in which the execution order of the tactical action node does not conform to the air combat tactical rule library is excluded to determine the second target causal path set.

6. The method of claim 1, wherein, The method further comprises: In response to a situation analysis request, according to the query content in the situation analysis request and the situation characterization data, a second target priori knowledge is determined from the preset situation analysis knowledge base through the RAG technology; The second target priori knowledge and the query content are embedded into a preset prompt word template to generate a model optimization prompt word; the model optimization prompt word is used to optimize the output result of the preset large language model.

7. The method of claim 1, wherein, The situation analysis report includes at least one of a threat level assessment, a tactical intention inference, a threat ranking and target identification, a counter-tactical suggestion, and a natural language explanation of the reasoning process; and the preset situation analysis knowledge base includes at least one of a weapon system library, a tactical rule library, an air combat database, a threat assessment and target database, and a sensor and electronic device library.

8. The method of claim 1, wherein, After the situation analysis report for the target air combat area is generated, the method further includes: based on the situation analysis report, visual information processing is performed to generate visual situation analysis data, and multiple feedback information options are output through a visual interactive interface; According to the feedback information option determined through the visual interactive interface, the preset large language model is optimized.

9. An air combat situation analysis system, characterized by It includes: The acquisition module is used for acquiring the situation representation data of the target air combat area, and performing semantic coding on the situation representation data to generate a consistent air combat situation semantic description; the target air combat area includes at least one hostile target; The first determination module is used for taking the air combat situation semantic description as a retrieval input of a preset situation analysis knowledge base, and determining first target prior knowledge from the preset situation analysis knowledge base through RAG technology; the RAG technology is a retrieval enhancement generation technology; The first generation module is used for generating a situation semantic context based on the first target prior knowledge and the air combat situation semantic description, and performing air combat situation analysis according to the situation semantic context and a preset large language model to generate at least one tactical intention hypothesis for the hostile target in the target air combat area; The second determination module is used for performing forward causal traversal and reverse causal exclusion on the tactical intention hypothesis according to a preset tactical causal knowledge graph to determine an intention verification result for the tactical intention hypothesis; the intention verification result is used to represent whether the tactical intention hypothesis conforms to causal rationality; The preset tactical causal knowledge graph is determined based on an air combat tactical rule library and historical air combat confrontation data; The second generation module is used for generating a situation analysis report for the target air combat area according to the tactical intention hypothesis and the preset large language model when the intention verification result is verified to be passed; The second determination module is specifically used for: Based on the preset tactical causal knowledge graph, a first target causal path set with the tactical intention hypothesis as a terminal node is extracted; wherein a first target causal path is composed of a series relationship of a tactical element node, a tactical action node, and a tactical intention node; For each first target causal path in the first target causal path set, the tactical element node in each path is extracted to generate a necessary causal condition set composed of multiple tactical element nodes; According to the situation representation data and the necessary causal condition set, forward causal traversal is performed to determine a forward causal traversal result; In the case that the forward causal traversal result is verified to be failed, the forward causal traversal result is determined as the intention verification result; In a case where the forward causality traversal result is verified, based on the tactical action node and the situation representation data in each first target causality path, reverse causality exclusion is performed on the first target causality path set to determine a second target causality path set; According to the situation representation data and the second target causality path set, an intent verification result is determined.

Images