Multi-source reasoning-based confrontation environment target behavior analysis method and system

Abstract

The invention relates to the technical field of intelligent decision making, in particular to a multi-source reasoning-based confrontation environment target behavior analysis method and system. Comprising the following steps: detecting a blue target under a simulation system, obtaining target position and state information, and generating a knowledge graph; dynamically extracting an entity sub-graph associated with the blue target through a space-time constraint graph sampler; constructing a dual-channel decision generator to respectively output an action probability vector driven by an expert rule and an action probability vector derived by a Bayesian network; designing a high-dimensional decision space projection module to map the two types of probability vectors into n-dimensional space coordinate points, and performing probability trajectory fusion and error correction through an improved Kalman filter; and a game utility corrector is introduced to optimize the final action decision based on the Nash equilibrium principle. The problem that in the prior art, a static knowledge graph is difficult to support real-time fusion of multi-source decision information in a dynamic confrontation environment is solved, and the timeliness and accuracy of complex situation behavior reasoning are remarkably improved.

Description

Technical Field

[0002] The present invention relates to the technical field of intelligent decision-making, and particularly relates to a method and system for analyzing the behavior of an adversarial environment target based on multi-source reasoning. Background Art

[0003] Currently, the methods for analyzing the behavior of an adversarial environment target mainly rely on traditional rules and models for decision-making analysis. These methods usually perform certain analyses under static information conditions based on the reasoning of expert systems or prior knowledge. However, when faced with complex and dynamic adversarial environments, these methods are insufficient to cope with their uncertainty and variability. Especially in simulation systems, traditional methods for analyzing target behavior are difficult to process a large amount of sensor data in real time and make a quick response. The rigidity of the rule model results in a lack of sufficient adaptability in the face of dynamic environments and the diversity of target behaviors, thus being unable to effectively predict and make decisions, affecting the real-time response ability and accuracy of the system.

[0004] Advanced intelligent decision-making technologies adopt methods for analyzing target behavior based on multi-source reasoning. Such methods are expected to more accurately and quickly simulate the dynamic behavior of targets by combining multiple data sources and reasoning mechanisms. However, existing multi-source reasoning technologies still face many challenges in the integration with real-time target states and environmental data. Especially when it comes to how to cope with complex and changing adversarial environments and achieve fast and accurate decision-making analysis, existing technologies often suffer from insufficient robustness and efficiency. For a target behavior analysis system with extremely high real-time requirements, existing technologies are difficult to provide accurate, timely, and effective decision-making support, thus being difficult to meet the requirements of target behavior prediction and dynamic decision-making in complex environments in practical applications, restricting the overall effectiveness of the system and the efficient implementation of decisions. Summary of the Invention

[0005] In order to solve the problem of enabling intelligent decision-making quickly and accurately in a target behavior analysis system with extremely high real-time requirements, the present invention provides a method and system for analyzing the behavior of an adversarial environment target based on multi-source reasoning.

[0006] In a first aspect, a method for analyzing the behavior of an adversarial environment target based on multi-source reasoning provided by the present invention adopts the following technical solution: A method for analyzing the behavior of an adversarial environment target based on multi-source reasoning includes: Detect the blue-side target under the simulation system, obtain the target position and status information, and generate a knowledge graph; Dynamically extract the entity sub-graph associated with the blue-side target through a spatio-temporal constraint graph sampler, where the sub-graph contains entities and relationships related to the target status; Construct a dual-channel decision generator. Based on information such as the state and relationships of the blue-side entities, generate a first action probability vector by combining expert system rules respectively, and generate a second action probability vector by combining Bayesian network derivation. Map the two types of probability vectors to an n-dimensional decision space to form coordinate points, and perform probability trajectory fusion and error correction through an improved Kalman filter. Introduce a game utility modifier to optimize the filtering result based on the Nash equilibrium principle and output the final action decision.

[0007] Furthermore, the detection of the blue-side target in the simulation system, obtaining the target position and status information and generating a knowledge graph includes: Use multiple sensors to conduct all-round detection of the blue-side target, and collect information such as the target's position coordinates, movement speed, and equipment status. Preprocess the collected data, including noise filtering, data standardization, and outlier removal, to generate a spatio-temporal state data matrix in a unified format. According to domain knowledge and ontology models, determine the entity types and relationship types in the knowledge graph, and construct an initial knowledge graph to provide basic data for subsequent analysis.

[0008] Furthermore, the dynamic extraction of the entity subgraph associated with the blue-side target by the spatio-temporal constraint graph sampler includes: Partition the initial target knowledge graph according to spatio-temporal constraints, and extract the entities and their relationships that are closely related to the current target state. Adopt a preset weight scoring mechanism to evaluate each entity and relationship in the subgraph, and screen out the subgraph data with strong representativeness. Output the blue-side target entity subgraph containing key state information, behavior clues, and environmental interaction relationships for subsequent decision-making analysis.

[0009] Furthermore, the construction of the dual-channel decision generator includes: One channel, based on expert system rules, performs rule matching on the entity state and relationship information of the blue-side target to generate a first action probability vector. Another channel, based on Bayesian network reasoning, performs probability calculations on the target's historical behavior, environmental factors, and uncertainties to generate a second action probability vector. Normalize the probability vectors generated by the two channels respectively to ensure their comparability in the same decision space.

[0010] Furthermore, the one channel based on expert system rules includes: Input the state and relationship information in the blue-side entity subgraph into the expert system. The expert system analyzes and reasons the input information according to the pre-set rule base. For each possible action, the expert system determines its likelihood of occurrence according to the rules and gives the corresponding probability value.

[0011] Furthermore, the two-channel inference based on the Bayesian network includes: Taking the observed data in the blue-side entity subgraph as the input of the Bayesian network; For each action, the Bayesian network calculates the probability of the action occurring under the current observed data; Organize these probability values into a second action probability vector, which also contains n dimensions, and each dimension corresponds to the probability of an action.

[0012] Furthermore, the mapping of the two types of probability vectors to the n-dimensional decision space to form coordinate points includes: Taking the first action probability vector and the second action probability vector as the coordinates of two points in the n-dimensional decision space respectively, and using an improved Kalman filter to fuse these two points according to the dynamic model of the system and the current observed data; During the fusion process, by calculating the prediction error and the measurement error, the probability trajectory is corrected to reduce the error and make the fusion result closer to the real situation; After multiple iterative calculations, a fused probability vector is obtained as the basis for the preliminary action decision.

[0013] Furthermore, the introduction of a game utility modifier to optimize the filtering result includes: Construct an adversarial game model based on the Nash equilibrium principle, and use the fused probability vector as the initial action decision; Calculate the utility values of each decision plan in the game model to evaluate the advantages and disadvantages and mutual influences between different decisions; Adjust the filtering result according to the utility value, optimize and output the final action decision.

[0014] In a second aspect, an adversarial environment target behavior analysis system based on multi-source inference includes: A target detection module for detecting the blue-side targets in the simulation system to obtain target position, movement speed, and equipment status information; A knowledge graph generation module connected to the target detection module for constructing an initial knowledge graph including entity types and relationship types; A spatio-temporal constraint graph sampling module connected to the knowledge graph generation module for dynamically extracting entity subgraphs related to the target state and screening representative data through a weight scoring mechanism; A two-channel decision generation module connected to the spatio-temporal constraint graph sampling module, including an expert system channel and a Bayesian network channel, which respectively generate a first action probability vector and a second action probability vector; A decision space mapping module, connected to the dual-channel decision generation module, for mapping a probability vector to an n-dimensional decision space to form a coordinate point; A Kalman fusion module, connected to the decision space mapping module, using an improved Kalman filter for probability trajectory fusion and error correction; A game optimization module, connected to the Kalman fusion module, constructing an adversarial game model based on the Nash equilibrium principle to optimize decision-making utility; A decision output module, connected to the game optimization module, for outputting a final action decision.

[0015] Thirdly, the present invention provides a computer-readable storage medium, in which multiple instructions are stored, and the instructions are adapted to be loaded and executed by a processor of a terminal device for the method for constructing a knowledge graph related to the situation and action of marine equipment.

[0016] Fourthly, the present invention provides a terminal device, including a processor and a computer-readable storage medium. The processor is used to implement each instruction; the computer-readable storage medium is used to store multiple instructions, and the instructions are adapted to be loaded and executed by the processor for the method for constructing a knowledge graph related to the situation and action of marine equipment.

[0017] In summary, the present invention has the following beneficial technical effects: 1. The method and system for analyzing target behavior in an adversarial environment based on multi-source reasoning proposed by the present invention deeply integrates multiple technologies such as knowledge graphs, expert systems, Bayesian networks, Kalman filters, and game theory, achieving efficient fusion of multi-source information. With the multi-source reasoning architecture, the unique advantages of each technology are fully explored and utilized, enabling comprehensive and accurate analysis of target behavior from multiple dimensions and levels, significantly improving the credibility and accuracy of the analysis results. Compared with traditional analysis methods using a single technology, the present invention shows significant superiority in accuracy, comprehensiveness, and reliability.

[0018] 2. The present invention uses a spatio-temporal constraint graph sampler to dynamically extract entity subgraphs and screens representative data samples with a preset weight scoring mechanism. This technical means can accurately focus on key information closely related to target behavior analysis, effectively excluding interference from irrelevant information. It greatly improves the efficiency and real-time performance of the analysis process, enabling the system to quickly perceive and respond to the dynamic changes in the adversarial environment, providing timely, accurate, and powerful support for the real-time decision-making process.

[0019] 3. The present invention constructs a dual-channel decision generator, which conducts independent decision-making analysis relying on expert system rules and Bayesian networks respectively. The expert system rule channel fully absorbs the rich experience and professional knowledge of domain experts to judge target behaviors based on prior knowledge; the Bayesian network channel conducts probability reasoning based on data-driven, mining potential rules and trends from massive data. Subsequently, the analysis results of the two channels are deeply fused through normalization processing and the Kalman filtering algorithm. This innovative decision-making generation method can not only make full use of the wisdom of experts but also combine the objective laws of data, effectively coping with and processing the uncertainties and conflicting information commonly existing in adversarial environments, and greatly improving the scientificity and rationality of decision-making.

[0020] 4. The present invention innovatively introduces a game utility modifier to optimize and adjust the results after Kalman filtering based on the Nash equilibrium principle. This modifier fully considers the strategic interaction relationship between the two adversarial parties during the decision-making process, calculates the utility values of each decision-making plan in the game environment accurately, and screens out the optimal decision-making plan from numerous alternative plans. This optimization mechanism can ensure the maximization of benefits and effects in complex and changeable adversarial environments, significantly improving the efficiency of combat command and the overall combat capabilities of combat units. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 is the flowchart of the method in Embodiment 1 of the present invention; Figure 2 is the schematic structural diagram of the system in Embodiment 2 of the present invention; DETAILED DESCRIPTION OF THE INVENTION

[0022] The present invention will be further described in detail below with reference to the accompanying drawings.

[0023] Embodiment 1 Refer to Figure 1 , a method for constructing a knowledge graph related to the situation and actions of maritime equipment in this embodiment includes: S1. Detect the blue-side targets under the simulation system, and obtain the target position and status information and generate a knowledge graph, including: S11. Use a variety of sensors to conduct all-round detection of the blue-side targets, and collect information such as the position coordinates, movement speed, and equipment status of the targets; In the simulation system, the comprehensive use of a variety of sensors is of great significance. Different types of sensors have unique detection advantages. Through all-round detection, detailed information of the blue-side targets can be collected from multiple angles to ensure the integrity and accuracy of the information.

[0024] S12. Preprocess the collected data, including noise filtering, data standardization, and outlier removal, to generate a spatio-temporal state data matrix in a unified format; The collected data often has problems such as noise, inconsistent data scales, and outliers, which will seriously affect the subsequent analysis results. Noise filtering can remove the interference signals in the data and improve the data quality. Data standardization is to unify different types of data to the same scale for easy comparison and analysis. Outlier removal is to identify and remove the values that significantly deviate from the normal range in the data to avoid these outliers misleading the analysis results. The finally generated spatio-temporal state data matrix in a unified format provides a standardized and ordered data basis for the subsequent construction of the knowledge graph.

[0025] S13. Determine the entity types and relationship types in the knowledge graph according to the domain knowledge and ontology model, and construct an initial knowledge graph to provide basic data for subsequent analysis; Domain knowledge and ontology model are important bases for constructing the knowledge graph. Domain knowledge contains the professional knowledge and experience in the adversarial environment and stipulates the entity types and relationship types. By combining domain knowledge and ontology model, the entity types in the knowledge graph and the relationship types between them can be accurately determined. Constructing an initial knowledge graph presents the collected and processed data in a graphical way, intuitively showing the associations between various entities and providing data support for subsequent intelligent decision-making.

[0026] S2. Dynamically extract the entity subgraph associated with the blue side's target through the spatio-temporal constraint graph sampler. The subgraph contains entities and relationships related to the target state, including: S21. Divide the initial target knowledge graph according to spatio-temporal constraints, and extract the entities closely related to the current target state and the relationships between them; Spatio-temporal constraints refer to extracting information within a specific time range and spatial area. In the adversarial environment, the behavior and state of the target change with time and space. By setting spatio-temporal constraint conditions, the information irrelevant to the current target state in the initial knowledge graph can be excluded, and only the entities and relationships related to the current time and space range are extracted. This can reduce unnecessary information interference and make the analysis more focused.

[0027] S22. Use a preset weight scoring mechanism to evaluate each entity and relationship in the subgraph, and screen out the subgraph data with stronger representativeness; The preset weight scoring mechanism can assign corresponding weights to entities and relationships according to their importance. The evaluation of importance can consider multiple factors. By scoring each entity and relationship, the entities and relationships with higher scores can be screened out to form subgraph data with stronger representativeness. This can further optimize the quality of the subgraph and ensure that the information contained in the subgraph is the most valuable information for analyzing the target behavior.

[0028] S23. Output the sub - graph of the blue - side target entity containing key status information, behavior clues, and environmental interaction relationships for subsequent decision - making analysis; After spatio - temporal constraint segmentation and weight - scoring screening, the output entity sub - graph contains key information closely related to the current state of the blue - side target. This information includes the target's key status information, such as current location, movement speed, equipment status, etc.; behavior clues, such as the target's historical movement trajectory, recent action patterns, etc.; and environmental interaction relationships, such as the positional relationships, cooperation, or confrontation relationships with other surrounding entities. This information provides a direct and effective basis for subsequent decision - making analysis, enabling analysts to more accurately predict the target's behavior and formulate corresponding countermeasures.

[0029] S3. Construct a two - channel decision generator. Based on information such as the state and relationships of the blue - side entities, generate the first action probability vector by combining expert - system rules respectively, and generate the second action probability vector by combining Bayesian network derivation, including: S31. One channel, based on expert - system rules, performs rule matching on the entity state and relationship information of the blue - side target to generate the first action probability vector; Expert - system rules are a series of rules summarized based on the knowledge and experience of domain experts. These rules describe the actions that the target may take and their occurrence probabilities under different entity states and relationships. Input the state and relationship information in the blue - side entity sub - graph into the expert system, and the expert system will perform rule matching according to the pre - set rule base. For each possible action, the expert system will judge its possibility according to the rules and give the corresponding probability value. These probability values form the first action probability vector. The advantage of the expert system is that it can utilize the rich experience and professional knowledge of domain experts to make judgments on target behavior based on prior knowledge.

[0030] S32. The other channel, based on Bayesian network reasoning, performs probability calculations on the target's historical behavior, environmental factors, and uncertainties to generate the second action probability vector; A Bayesian network is a model based on probabilistic reasoning that can handle uncertainties and complex causal relationships. Input the observed data in the blue - side entity sub - graph as the input of the Bayesian network, and the Bayesian network will perform reasoning calculations according to the pre - learned probability relationships (conditional probability tables) between nodes. For each action, the Bayesian network will calculate the probability of the action occurring under the current observed data. These probability values are organized into the second action probability vector. The advantage of the Bayesian network is that it can mine potential laws and information from data, taking into account the target's historical behavior, environmental factors, and various uncertainties.

[0031] S33. Normalize the probability vectors generated by the two channels respectively to ensure their comparability within the same decision - making space; Since the first action probability vector and the second action probability vector are generated by different methods, their value ranges and scales may be different. Normalization is to adjust all elements in the probability vector so that their sum is 1. Through normalization, the two probability vectors can be unified into the same decision space, facilitating comparison and fusion. This can ensure that in the subsequent decision-making process, the probability information generated by the two channels can participate in decision-making fairly and effectively.

[0032] S4. Mapping the two types of probability vectors to the n-dimensional decision space to form coordinate points, and the probability trajectory fusion and error correction through the improved Kalman filter include: S41. Taking the first action probability vector and the second action probability vector as the coordinates of two points in the n-dimensional decision space respectively, and using the improved Kalman filter to fuse these two points according to the dynamic model of the system and the current observation data; The n-dimensional decision space is an abstract mathematical space, where each dimension corresponds to the probability of an action. Taking the first action probability vector and the second action probability vector as the coordinates of two points in the space can visually represent the probability information generated by the two channels. The improved Kalman filter is a filtering algorithm based on the optimal estimation theory, which can fuse these two points according to the dynamic model of the system and the current observation data. The dynamic model of the system describes the variation law of the target behavior, and the observation data is the current target state information obtained. By fusing these two points, the information of the two channels can be comprehensively considered to obtain a more accurate probability estimate.

[0033] S42. During the fusion process, by calculating the prediction error and the measurement error, the probability trajectory is corrected to reduce the error and make the fusion result closer to the real situation; During the fusion process, the prediction error refers to the difference between the probability value predicted according to the system dynamic model and the actually observed probability value, and the measurement error refers to the error existing in the observation data itself. By calculating these two errors, the improved Kalman filter can correct the probability trajectory. If the prediction error is large, it indicates that there may be a deviation in the system dynamic model and the model needs to be adjusted; if the measurement error is large, it indicates that the accuracy of the observation data is low and the data needs to be further processed. By continuously correcting the probability trajectory, the error can be reduced and the fusion result can be made closer to the real situation.

[0034] S43. After multiple iterative calculations, the fused probability vector is obtained as the basis for the preliminary action decision; Multiple iterative calculations can enable the improved Kalman filter to continuously optimize the fusion result. In each iteration, the filter corrects the probability trajectory based on new observation data and error information, gradually improving the accuracy of the fusion result. After multiple iterations, the fused probability vector obtained can be used as the basis for preliminary action decisions. This probability vector synthesizes the information from two channels and has undergone error correction, being able to more accurately reflect the likelihood of the target taking various actions.

[0035] S5. Introduce a game utility corrector to optimize the filtering result based on the Nash equilibrium principle, and output the final action decision, including: S51. Construct an adversarial game model based on the Nash equilibrium principle, and use the fused probability vector as the initial action decision; The Nash equilibrium principle describes an equilibrium state in an adversarial environment where both sides adopt optimal strategies. Construct an adversarial game model based on the Nash equilibrium principle, and use the fused probability vector as the initial action decision. In this model, the possible actions of the enemy and the strategic interaction between the two sides are considered. By analyzing the gains and losses under different strategy combinations, the optimal decision-making scheme is sought.

[0036] S52. Calculate the utility values of each decision-making scheme in the game model, and evaluate the advantages and disadvantages of different decisions and their mutual influences; In the adversarial game model, each decision-making scheme corresponds to a utility value. The utility value comprehensively considers various factors and is used to evaluate the advantages and disadvantages of the decision-making scheme. By calculating the utility values of each decision-making scheme, the advantages and disadvantages of different decisions can be compared, and their mutual influences can be analyzed. By comparing the utility values, the most favorable decision-making scheme in the current situation can be selected.

[0037] S53. Adjust the filtering result according to the utility value, optimize and output the final action decision; Adjust the filtering result according to the calculated utility value. If the utility value of a certain decision-making scheme is high, increase the weight of this scheme in the final decision; if the utility value of a certain decision-making scheme is low, reduce its weight. In this way, optimize the filtering result and select the action with the maximum utility value as the final action decision. The finally output action decision is the optimal decision obtained after considering the likelihood of the target's behavior, the strategic interaction of the enemy, and the comprehensive influence of various factors, and can achieve the maximum benefit and effect in the adversarial environment.

[0038] The method for analyzing the target behavior in an adversarial environment based on multi-source reasoning proposed in this embodiment aims to address the deficiencies of existing technologies in integrating multi-source information, handling uncertainties, and adapting to complex adversarial environments. The limitations of existing technologies make it difficult to achieve accurate and real-time decision-making, restricting their application in adversarial scenarios. To accurately and real-time analyze the behavior of the blue-side targets and make decisions in the simulation system, this method operates according to the following process. First, comprehensively detect the blue-side targets, collect information such as position, speed, and equipment status using various sensors, and generate a spatio-temporal state data matrix through preprocessing. Then, construct an initial knowledge graph based on domain knowledge and ontology models. Next, through a spatio-temporal constraint graph sampler, segment the knowledge graph according to spatio-temporal constraints, extract relevant entities and relationships, and use a weight scoring mechanism to screen out representative sub-graph data, outputting an entity sub-graph containing key information. Then, construct a dual-channel decision generator. One channel is based on expert system rules, inputs target entity status and relationship information for rule matching, and generates a first action probability vector; the other channel is based on Bayesian network reasoning, calculates probabilities with observed data as input, and generates a second action probability vector, and normalizes the two to ensure comparability. After that, map the two types of probability vectors to the n-dimensional decision space to form coordinate points, fuse them using an improved Kalman filter, calculate the error-corrected probability trajectory, and obtain the fused probability vector as the preliminary decision-making basis through multiple iterations. Finally, introduce a game utility corrector, construct an adversarial game model based on the Nash equilibrium principle, use the fused probability vector as the initial decision, calculate the utility value to evaluate the quality of the decision and make adjustments, and output the final action decision. This method improves the accuracy of target behavior analysis and the scientific nature of decision-making, provides strong support for command and control systems in adversarial environments, etc., and enhances the effectiveness and real-time nature of adversarial decision-making.

[0039] Embodiment 2 Referring to Figure 2 , this embodiment provides a knowledge graph construction system for the association between the situation and actions of maritime equipment, including: A target detection module, which is used to detect the blue-side targets in the simulation system and obtain target position, movement speed, and equipment status information; A knowledge graph generation module, which is connected to the target detection module and is used to construct an initial knowledge graph including entity types and relationship types; A spatio-temporal constraint graph sampling module, which is connected to the knowledge graph generation module and is used to dynamically extract entity sub-graphs related to the target state and screen representative data through a weight scoring mechanism; A dual-channel decision generation module, which is connected to the spatio-temporal constraint graph sampling module and includes an expert system channel and a Bayesian network channel, respectively generating a first action probability vector and a second action probability vector; A decision space mapping module, which is connected to the dual-channel decision generation module and is used to map the probability vectors to the n-dimensional decision space to form coordinate points; The Kalman fusion module, connected to the decision space mapping module, uses an improved Kalman filter for probabilistic trajectory fusion and error correction; The game optimization module, connected to the Kalman fusion module, constructs an adversarial game model based on the Nash equilibrium principle to optimize the decision-making utility; The decision output module, connected to the game optimization module, is used to output the final action decision.

[0040] A computer-readable storage medium stores multiple instructions, and the instructions are adapted to be loaded and executed by a processor of a terminal device for the method for constructing a knowledge graph related to the situation and action of a maritime equipment.

[0041] A terminal device includes a processor and a computer-readable storage medium. The processor is used to implement each instruction; the computer-readable storage medium is used to store multiple instructions, and the instructions are adapted to be loaded and executed by the processor for the method for constructing a knowledge graph related to the situation and action of a maritime equipment.

[0042] The above are all preferred embodiments of the present invention. The protection scope of the present invention is not limited thereby. Therefore, all equivalent changes made according to the structure, shape, and principle of the present invention shall be covered by the protection scope of the present invention.

Claims

1. A method for analyzing target behavior in an adversarial environment based on multi-source reasoning, characterized in that: include: Detect the blue team target in the simulation system, obtain the target position and status information and generate a knowledge graph; Dynamically extract the entity subgraph associated with the blue team's target through the spatiotemporal constraint graph sampler, where the subgraph contains entities and relationships related to the target state; Construct a dual-channel decision generator, based on the blue party's entity status and relationship information, combine the expert system rules to generate the first action probability vector, and combine the Bayesian network to derive the second action probability vector; The two types of probability vectors are mapped to the n-dimensional decision space to form coordinate points, and the probability trajectory fusion and error correction are performed through the improved Kalman filter; A game utility corrector is introduced to optimize the filtering results based on the Nash equilibrium principle and output the final action decision.

2. According to the method of analyzing target behavior in a confrontation environment based on multi-source reasoning in claim 1, it is characterized in that: Detecting the blue target under the simulation system, obtaining the target position and state information and generating a knowledge graph includes: Use a variety of sensors to conduct all-round detection of blue targets and collect information such as the target's location coordinates, movement speed, and equipment status; Preprocess the collected data, including noise filtering, data standardization, and outlier removal, to generate a spatiotemporal state data matrix in a unified format; Based on domain knowledge and ontology models, the entity types and relationship types in the knowledge graph are determined, and an initial knowledge graph is constructed to provide basic data for subsequent analysis.

3. According to the method of analyzing target behavior in a confrontation environment based on multi-source reasoning in claim 1, it is characterized in that: The method of dynamically extracting the entity subgraph associated with the blue team target through the spatiotemporal constraint graph sampler includes: The initial target knowledge graph is segmented according to the time and space constraints to extract entities closely related to the current target state and their relationships; Use the preset weight scoring mechanism to evaluate each entity and relationship in the subgraph, and select subgraph data with strong representativeness; Output the blue target entity subgraph containing key status information, behavioral clues and environmental interaction relationships for subsequent decision analysis.

4. The method for analyzing target behavior in a confrontation environment based on multi-source reasoning according to claim 1, characterized in that: The construction of the dual-channel decision generator includes: One channel matches the entity state and relationship information of the blue team's target based on the expert system rules to generate the first action probability vector; The other channel uses Bayesian network reasoning to perform probability calculations on the target’s historical behavior, environmental factors, and uncertainties to generate a second action probability vector; The probability vectors generated by the two channels are normalized separately to ensure their comparability in the same decision space.

5. The method for analyzing target behavior in a confrontation environment based on multi-source reasoning according to claim 4 is characterized in that: The one channel based on expert system rules includes: Input the state and relationship information of the blue square entity subgraph into the expert system; Expert systems analyze and reason about input information based on a pre-set rule base; For each possible action, the expert system determines the possibility of its occurrence based on the rules and gives a corresponding probability value.

6. The method for analyzing target behavior in a confrontation environment based on multi-source reasoning according to claim 4, characterized in that: The two-channel Bayesian network reasoning includes: The observation data in the blue entity subgraph is used as the input of the Bayesian network; For each action, the Bayesian network calculates the probability of that action occurring given the current observed data; These probability values ​​are organized into a second action probability vector, which also contains n dimensions, and each dimension corresponds to the probability of an action.

7. The method for analyzing target behavior in a confrontation environment based on multi-source reasoning according to claim 1, characterized in that: Mapping the two types of probability vectors to the n-dimensional decision space to form coordinate points includes: The first action probability vector and the second action probability vector are respectively used as the coordinates of two points in the n-dimensional decision space, and the improved Kalman filter is used to fuse the two points according to the dynamic model of the system and the current observation data; In the fusion process, the prediction error and measurement error are calculated to correct the probability trajectory, reduce the error, and make the fusion result closer to the actual situation; After multiple iterative calculations, the fused probability vector is obtained as the basis for preliminary action decisions.

8. The method for analyzing target behavior in a confrontation environment based on multi-source reasoning according to claim 1, characterized in that: The introducing of the game utility modifier to optimize the filtering result includes: Construct an adversarial game model based on the Nash equilibrium principle, and use the fused probability vector as the initial action decision; Calculate the utility value of each decision-making plan in the game model and evaluate the pros and cons and mutual influence between different decisions; The filtering results are adjusted according to the utility value to optimize and output the final action decision.

9. A system for analyzing target behavior in an adversarial environment based on multi-source reasoning, comprising: The target detection module is used to detect the blue team's targets in the simulation system and obtain the target position, movement speed and equipment status information; The knowledge graph generation module is connected to the target detection module and is used to construct an initial knowledge graph containing entity types and relationship types; The spatiotemporal constraint graph sampling module is connected to the knowledge graph generation module to dynamically extract entity subgraphs related to the target state and filter representative data through a weighted scoring mechanism; A dual-channel decision generation module, connected to the spatiotemporal constraint graph sampling module, includes an expert system channel and a Bayesian network channel, and generates a first action probability vector and a second action probability vector respectively; A decision space mapping module, connected to the dual-channel decision generation module, is used to map the probability vector to the n-dimensional decision space to form a coordinate point; The Kalman fusion module is connected to the decision space mapping module and uses an improved Kalman filter to perform probability trajectory fusion and error correction; The game optimization module is connected with the Kalman fusion module to build an adversarial game model based on the Nash equilibrium principle to optimize the decision-making utility; The decision output module is connected to the game optimization module and is used to output the final action decision.

Images