Discrete node decision quality evaluation method and system based on trainer

Abstract

The application provides a trainer-based discrete node decision quality evaluation method and system, relates to the technical field of artificial intelligence, and comprises the following steps: acquiring a discrete decision node in a battlefield simulation training environment, performing semantic modeling based on an ontology model to construct a knowledge graph, identifying a node sequence meeting a cause-effect transmission characteristic to form a candidate association path set, and screening an association path meeting a condition through topological structure analysis and semantic consistency verification to obtain a decision node association relationship. The application can accurately identify a battlefield decision cause-effect chain and improve the scientificity and objectivity of decision quality evaluation.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to a method and system for evaluating the decision quality of discrete nodes based on a trainer. Background Technology

[0002] Traditional battlefield simulation training systems primarily evaluate decision quality through expert experience assessments or simple quantitative indicators. This approach often relies on manual analysis and lacks in-depth analysis of causal relationships within the decision-making process. Existing technologies still have significant shortcomings in evaluating the quality of decisions at discrete nodes: current evaluation methods struggle to effectively express and analyze the complex semantic relationships between decision nodes, resulting in an inability to fully grasp the causal transmission mechanisms within the decision-making chain. Entities and behaviors in the decision-making process exhibit multi-dimensional and multi-layered relationships, but existing methods often simplify these relationships, losing crucial semantic information.

[0003] Existing technologies lack the ability to dynamically track decision-making paths and cannot accurately identify and extract decision sequences that satisfy causal transmission characteristics. In complex battlefield environments, the impact of decisions often propagates along nonlinear paths, while existing methods mostly employ static analysis models, making it difficult to capture this dynamic evolution process and thus affecting the accuracy of assessment results.

[0004] Existing decision quality assessment methods lack a mechanism to verify the reliability of decision-related paths. They typically focus only on a single dimension of connection strength or semantic matching, failing to comprehensively consider the topological characteristics and semantic consistency of the paths. This can lead to logical contradictions or inconsistencies with actual battlefield realities in the extracted decision-related relationships, ultimately affecting the effectiveness and credibility of decision quality assessment. Summary of the Invention

[0005] The present invention provides a method and system for evaluating the decision quality of discrete nodes based on a trainer, which can solve the problems in the prior art.

[0006] A first aspect of the present invention provides a method for evaluating the decision quality of discrete nodes based on a trainer, comprising:

[0007] Multiple discrete decision nodes are acquired in a battlefield simulation training environment. Each discrete decision node includes a decision subject identifier, a decision time sequence identifier, and decision status information.

[0008] Based on a pre-defined ontology model, semantic modeling is performed on the discrete decision nodes to obtain a knowledge graph containing entity nodes and relation edges. The entity nodes represent decision subjects, decision objects, and decision behaviors, and the relation edges represent the semantic associations between entity nodes.

[0009] identify a node sequence satisfying a causal transmission characteristic in the knowledge graph based on a graph structure traversal rule and a semantic constraint condition, the causal transmission characteristic representing that decision state information of a preceding node has an influence on decision state information of a subsequent node, to obtain a candidate association path set;

[0010] perform topological structure analysis and semantic consistency verification on each association path in the candidate association path set, the topological structure analysis being used to calculate a connectivity strength between nodes in a path, and the semantic consistency verification being used to verify whether a relationship edge between adjacent nodes in the path conforms to a preset causal logic rule, to filter out an association path that conforms to a connectivity strength threshold and passes the semantic consistency verification, to obtain a decision node association relationship.

[0011] perform semantic modeling on the discrete decision nodes based on a preset ontology model, to obtain a knowledge graph including entity nodes and relationship edges, including:

[0012] analyze decision state information in the discrete decision nodes, to identify decision attributes and decision content therein, map the decision attributes to entity type labels, and map the decision content to attribute values of entity nodes;

[0013] convert the decision nodes carrying the entity type labels and the attribute values into entity nodes according to an entity classification rule defined in the preset ontology model, the entity classification rule defining a hierarchical attribution relationship of a decision subject, a decision object, and a decision behavior in the ontology model;

[0014] analyze semantic dependency relationships between the entity nodes based on a pre-defined relationship mode in the preset ontology model, determine a connection mode between entity nodes according to the semantic dependency relationships, generate a relationship edge representing the connection mode, and the relationship edge carrying a relationship type identifier and relationship strength information;

[0015] organize the entity nodes and the relationship edges into a graph data structure, to obtain the knowledge graph including the entity nodes and the relationship edges.

[0016] identify a node sequence satisfying a causal transmission characteristic in the knowledge graph based on a graph structure traversal rule and a semantic constraint condition, to obtain a candidate association path set, including:

[0017] perform path expansion along relationship edges according to the graph structure traversal rule, record entity nodes and relationship edges passed in the path expansion process, to form a node sequence, and the graph structure traversal rule defining direction constraints and depth constraints of path expansion;

[0018] extracting decision state information of a preceding entity node and decision state information of a subsequent entity node for adjacent entity nodes in the node sequence, and judging whether the decision state information of the preceding entity node has an influence relationship on the decision state information of the subsequent entity node based on the semantic constraint condition;

[0019] performing a causal transmission characteristic verification on the node sequence judged by the semantic constraint condition, the causal transmission characteristic verification being used to detect whether a complete state transmission chain exists in the node sequence, the state transmission chain representing a transmission path in which decision state information from a starting node to a terminal node successively has an influence;

[0020] marking the node sequence verified by the causal transmission characteristic verification as a candidate association path, collecting all the candidate association paths, and obtaining a candidate association path set.

[0021] performing a causal transmission characteristic verification on the node sequence judged by the semantic constraint condition includes:

[0022] performing time sequence sorting on a time sequence coordinate system of the entity nodes in the node sequence after time sequence alignment, obtaining a node chain in time sequence order, extracting decision state information of each entity node in the node chain in time sequence order, and constructing a state information sequence;

[0023] traversing the state information sequence, analyzing a dependency relationship between a state change amount in preceding state information and a state change amount in subsequent state information for adjacent state information pairs, the dependency relationship representing whether the preceding state change amount is a trigger condition for the subsequent state change amount;

[0024] constructing a state transmission matrix according to the dependency relationship, the state transmission matrix recording state transmission relationships between the entity nodes in the node chain in time sequence order;

[0025] detecting, based on the state transmission matrix, whether a continuous state transmission relationship exists from a starting node to a terminal node of the node chain in time sequence order, and if the continuous state transmission relationship exists, determining that the node sequence has a causal transmission characteristic, the continuous state transmission relationship representing that there is no transmission interruption in a state transmission chain.

[0026] performing a topological structure analysis and a semantic consistency verification on each association path in the candidate association path set, screening out association paths that meet a connectivity strength threshold and pass the semantic consistency verification, and obtaining a decision node association relationship including:

[0027] extracting entity nodes and relationship edges in the association path, counting a number of the relationship edges and relationship strength information carried by the relationship edges, and calculating a connectivity strength of the association path based on the number of the relationship edges and the relationship strength information.

[0028] screening the associated paths with the connectivity strength greater than or equal to the connectivity strength threshold value to obtain a path set satisfying the connectivity strength requirement;

[0029] For each associated path in the path set satisfying the connectivity strength requirement, the relationship type identifier carried by the relationship edge between adjacent entity nodes in the associated path is extracted, and whether the relationship type identifier conforms to the semantic constraint of causal reasoning is verified according to a preset causal logic rule.

[0030] The associated path verified by the causal logic rule is marked, and the marked associated path is taken as an associated path conforming to the connectivity strength threshold value and passing the semantic consistency verification.

[0031] The starting entity node and the terminal entity node are extracted from the associated path conforming to the connectivity strength threshold value and passing the semantic consistency verification, and an associated relationship record between the starting entity node and the terminal entity node is established, the associated relationship record containing an associated path identifier and associated strength information, to obtain the decision node associated relationship.

[0032] The second aspect of the embodiment of the application provides a discrete node decision quality evaluation system based on a trainer, which comprises:

[0033] A first unit is configured to acquire a plurality of discrete decision nodes in a battlefield simulation training environment, wherein the discrete decision nodes include decision subject identifiers, decision time sequence identifiers and decision state information.

[0034] A second unit is configured to perform semantic modeling on the discrete decision nodes based on a preset ontology model to obtain a knowledge graph comprising entity nodes and relationship edges, wherein the entity nodes represent decision subjects, decision objects and decision behaviors, and the relationship edges represent semantic associations between the entity nodes.

[0035] A third unit is configured to identify a node sequence satisfying a causal transmission feature in the knowledge graph based on a graph structure traversal rule and a semantic constraint condition, wherein the causal transmission feature represents that the decision state information of a previous node has an impact on the decision state information of a subsequent node, and a candidate associated path set is obtained.

[0036] A fourth unit is configured to perform topological structure analysis and semantic consistency verification on each associated path in the candidate associated path set, wherein the topological structure analysis is used to calculate the connectivity strength between nodes in the path, and the semantic consistency verification is used to verify whether the relationship edges of adjacent nodes in the path conform to a preset causal logic rule, and an associated path conforming to the connectivity strength threshold value and passing the semantic consistency verification is screened out to obtain a decision node associated relationship.

[0037] The third aspect of the embodiment of the application,

[0038] An electronic device is provided, comprising:

[0039] a processor;

[0040] a memory for storing processor-executable instructions;

[0041] wherein the processor is configured to invoke the instructions stored by the memory to perform the method as described above.

[0042] A fourth aspect of the embodiments of the present application,

[0043] A computer-readable storage medium is provided, which stores computer program instructions, and the computer program instructions are executed by a processor to implement the method as described above.

[0044] The present application can effectively capture the complex semantic association between the decision subject, the decision object and the decision behavior by acquiring a plurality of discrete decision nodes in the battlefield simulation training environment and performing semantic modeling to construct a knowledge graph, thereby realizing comprehensive representation of the decision process and improving the accuracy of the decision node quality evaluation.

[0045] The present application identifies node sequences with causal transmission characteristics based on graph structure traversal rules and semantic constraint conditions, reveals the implicit causal relationship between decision nodes, and enables the evaluation process to consider the continuity of the decision, overcoming the limitations of traditional methods that only focus on a single decision point and ignore the overall effect of the decision link.

[0046] The present application filters the associated path by combining topological structure analysis and semantic consistency verification, considering the connectivity strength between nodes and verifying the causal logic rationality of the relationship edge, ensuring the accuracy and interpretability of the decision node association relationship, and providing a reliable data basis for subsequent decision quality evaluation. BRIEF DESCRIPTION OF DRAWINGS

[0047] Figure 1 A flowchart of the method for evaluating the quality of discrete nodes based on a trainer according to an embodiment of the present application. DETAILED DESCRIPTION

[0048] To make the objectives, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative labor fall within the scope of the present application.

[0049] The technical solutions of the present application are described in detail below with specific examples. The following specific examples can be combined with each other, and some examples may not be described in detail for the same or similar concepts or processes.

[0050] Figure 1 The flowchart of the method for evaluating the quality of discrete node decision-making based on a trainer according to an embodiment of the present application is shown in FIG. 1, which comprises the following steps: Figure 1

[0051] Obtaining a plurality of discrete decision-making nodes in a battlefield simulation training environment, wherein the discrete decision-making nodes include decision-making subject identification, decision-making time sequence identification, and decision-making state information.

[0052] Performing semantic modeling on the discrete decision-making nodes based on a preset ontology model to obtain a knowledge graph comprising entity nodes and relationship edges, wherein the entity nodes represent decision-making subjects, decision-making objects, and decision-making behaviors, and the relationship edges represent semantic associations between the entity nodes.

[0053] Identifying node sequences that satisfy a causal transmission feature in the knowledge graph based on graph structure traversal rules and semantic constraint conditions, wherein the causal transmission feature represents that the decision-making state information of a preceding node has an impact on the decision-making state information of a subsequent node, and obtaining a candidate association path set.

[0054] Performing topological structure analysis and semantic consistency verification on each association path in the candidate association path set, wherein the topological structure analysis is used to calculate the connectivity strength between nodes in the path, and the semantic consistency verification is used to verify whether the relationship edges of adjacent nodes in the path conform to a preset causal logic rule, and filtering out association paths that conform to a connectivity strength threshold and pass the semantic consistency verification to obtain decision-making node association relationships.

[0055] In an optional implementation, performing semantic modeling on the discrete decision-making nodes based on a preset ontology model to obtain a knowledge graph comprising entity nodes and relationship edges comprises:

[0056] Analyzing the decision-making state information in the discrete decision-making nodes to identify decision-making attributes and decision-making content therein, mapping the decision-making attributes to entity type labels, and mapping the decision-making content to attribute values of entity nodes.

[0057] Converting the decision-making nodes carrying the entity type labels and attribute values into entity nodes according to entity classification rules defined in the preset ontology model, wherein the entity classification rules define hierarchical attribution relationships of decision-making subjects, decision-making objects, and decision-making behaviors in the ontology model.

[0058] ​Based on a relationship mode predefined in the preset ontology model, semantic dependency relationships between the entity nodes are analyzed, a connection mode between the entity nodes is determined according to the semantic dependency relationships, and a relationship edge representing the connection mode is generated, the relationship edge carrying a relationship type identifier and relationship strength information;

[0059] The entity nodes and the relationship edge are organized into a graph data structure to obtain the knowledge graph containing the entity nodes and the relationship edge.

[0060] For example, a preset ontology model is first loaded, and the ontology model constructs a three-layer classification system for a battlefield simulation training scene. The top layer contains three basic categories of decision subjects, decision objects and decision behaviors. The middle layer is subdivided into categories such as command units, combat units, support units, enemy targets, friendly equipment, battlefield resources, reconnaissance behaviors, strike behaviors, maneuver behaviors and protection behaviors according to military combat characteristics. The bottom layer contains specific instance types such as red command post, blue armored battalion, reconnaissance unmanned aerial vehicle, air defense position, fire strike and tactical retreat. The relationship types include command relationship, support relationship, strike relationship, protection relationship and reconnaissance relationship, each relationship type defines domain and value domain constraints to ensure that only entity pairs that meet the military combat logic can establish a specific relationship.

[0061] The battlefield simulation trainer collects decision node data in real time during the training process, and each decision node is stored in a structured record form. An example of decision node data generated in a certain confrontation training is as follows: the node identifier is "TN20251119001", the decision type is "fire strike", the decision subject is "red artillery battalion", the decision object is "blue front-line position", the decision content is "implement three-wave fire coverage on target coordinates X107Y238", the decision time is "simulation time T+0325", and the decision state is "in execution". The record also contains battlefield situation information such as the amount of ammunition reserve of our side is 65%, the enemy defense strength evaluation value is 0.78, and the expected damage effect is moderate suppression.

[0062] When parsing the decision node, the attribute extraction module performs semantic analysis on the node data. The module identifies decision attributes based on a predefined military domain dictionary and template library. For the above example node, the identified decision attributes include nine dimensions: decision type, decision subject, decision object, decision content, decision time, decision state, ammunition status, enemy situation assessment, and expected effect. The decision type is mapped to the behavior entity type label "strike behavior", the decision subject is mapped to the subject entity type label "combat unit / fire unit", and the decision object is mapped to the object entity type label "enemy target / position facility". The decision content "implement three-wave fire coverage on target coordinates X107Y238" is parsed into three attribute values: attack coordinates, strike intensity, and strike wave. The decision time is converted to an integer value of the time stamp in the simulation time coordinate system, the decision state is mapped to a state code, and battlefield situation parameters such as ammunition status, enemy situation assessment, and expected effect are retained as numerical attributes.

[0063] The entity node conversion process is performed according to the entity classification rules of the ontology model. Three types of entity nodes are generated for the above decision node. The subject entity node has the identification "E20251119001", the type "combat unit / fire unit", the name "Red Artillery Regiment", and the attribute fields including unit organization as a regiment, equipment type as self-propelled artillery, current position as coordinates X095Y245, combat readiness state as good, and ammunition reserve as 65%. The object entity node has the identification "E20251119002", the type "enemy target / position facility", the name "Blue Front Line Position", and the attribute fields including target position as coordinates X107Y238, defense intensity as 0.78, target value assessment as high-value target, and current state as the existence of living forces. The behavior entity node has the identification "E20251119003", the type "strike behavior / fire coverage", the name "fire strike task TN20251119001", and the attribute fields including execution time as simulation time T+0325, strike wave as three times, expected damage as moderate suppression, and execution state as in progress.

[0064] The relationship edge is generated based on the relationship mode predefined by the ontology model. The relationship mode stipulates that the combat unit class entity can strike the behavior class entity by performing the relationship connection, and the behavior class entity can connect the enemy target class entity by the action relationship. It is analyzed that there is an execution relationship between the subject entity "Red Artillery Regiment" and the behavior entity "Fire Strike Task TN20251119001", and the identification of the relationship edge is generated as "R20251119001", the starting point is "E20251119003", the ending point is "E20251119003", the relationship type is "execution", and the relationship strength information is calculated by comprehensively considering the factors such as the combat readiness state, ammunition reserve, command response speed and the like of the decision subject. The calculation process is to map the good combat readiness state to the basic strength 0.9, the ammunition reserve 65% to the adjustment coefficient 0.93, and the fast command response speed to the bonus coefficient 1.05, and the relationship strength 0.8793 is obtained by multiplying the three. The second relationship edge is generated with the identification "R20251119002", the starting point "E20251119003", the ending point "E20251119002", the relationship type "action on", and the relationship strength is calculated by the strike wave, the expected damage, the target defense strength and the like. The strike wave is three times corresponding to the basic strength 0.85, the expected damage is moderate suppression mapping to the effect coefficient 0.7, and the target defense strength 0.78 is taken as the attenuation factor, and the relationship strength 0.4655 is obtained by comprehensive calculation.

[0065] The knowledge graph is constructed by using an attribute graph model as the underlying data structure. Entity nodes are stored in a node table, and each row record includes a node identifier, a node type, a node name, and a key-value pair set of attribute fields. Relationship edges are stored in an edge table, and each row record includes an edge identifier, a starting node identifier, an ending node identifier, a relationship type, a relationship strength, and a key-value pair set of relationship attributes. To support efficient graph traversal, an adjacency index structure is maintained, and the out-edge list and the in-edge list of each node are recorded. The out-edge list of the node "E20251119001" includes the edge "R20251119001", and the in-edge list includes the instruction relationship edge from the superior command node according to the training promotion. A type index is also established to support fast retrieval by node type or relationship type, and a time sequence index is established to support query of decision nodes by simulation time range.

[0066] The complex decision chain in the battlefield simulation training environment involves multi-level and multi-unit cooperation. The decision chain generated by a certain comprehensive confrontation training includes the following links: the red side command center issues reconnaissance instructions, the reconnaissance team executes the reconnaissance task, the reconnaissance team reports enemy situation, the red side command center issues strike instructions, the artillery battalion executes fire strikes, and the armored battalion implements tactical attacks. A knowledge subgraph including 15 entity nodes and 21 relationship edges is constructed for the decision chain. The entity nodes include 4 command unit nodes, 3 combat unit nodes, 2 reconnaissance unit nodes, 3 enemy target nodes and 3 behavior nodes. The relationship edge types cover command relationship, execution relationship, reconnaissance relationship, reporting relationship, strike relationship and support relationship. The relationship strength is dynamically calculated according to multi-dimensional parameters such as the combat readiness state, equipment performance, battlefield situation and command efficiency, and the numerical range is between 0.3 and 0.98, reflecting the reliability and influence difference of different decision links.

[0067] The knowledge graph is updated continuously with the training process, and the newly generated decision nodes are converted into entity nodes in real time and inserted into the graph, and the newly identified associated relationships generate relationship edges and update the adjacency index. The graph consistency is checked regularly to check whether there are abnormal situations such as isolated nodes, hanging edges and type conflicts. For the decision nodes whose state changes, the attribute fields of the corresponding entity nodes are updated, and the strength information of the related relationship edges is recalculated to ensure that the knowledge graph accurately reflects the current battlefield situation and decision state. The size of the graph increases with the training time, and a single confrontation training can generate hundreds to thousands of entity nodes and thousands to tens of thousands of relationship edges, providing a complete knowledge base for subsequent causal path identification and decision quality evaluation.

[0068] The application converts the discrete decision nodes in the battlefield simulation training into structured knowledge representation with semantic association by constructing a knowledge graph based on an ontology model, breaking through the limitation of traditional methods that only rely on time sequence relationship for decision analysis. Through semantic modeling of entity nodes and relationship edges, the complex association between decision subjects, decision objects and decision behaviors can be accurately described, significantly improving the accuracy and completeness of decision relationship mining.

[0069] In an optional implementation, based on the graph structure traversal rule and the semantic constraint condition, a node sequence satisfying the causal transmission characteristic is identified in the knowledge graph to obtain a candidate association path set, including:

[0070] According to the graph structure traversal rule, the path is expanded along the relationship edge, and the entity nodes and relationship edges passed in the path expansion process are recorded to form a node sequence, and the graph structure traversal rule defines the direction constraint and depth constraint of the path expansion;

[0071] For adjacent entity nodes in the node sequence, the decision state information of the preceding entity node and the decision state information of the subsequent entity node are extracted, and it is judged whether the decision state information of the preceding entity node constitutes an influence relationship for the decision state information of the subsequent entity node based on the semantic constraint condition;

[0072] Causal transmission characteristic verification is performed on the node sequence judged by the semantic constraint condition, and the causal transmission characteristic verification is used to detect whether there is a complete state transmission chain in the node sequence, and the state transmission chain represents a transmission path in which the decision state information from the starting node to the terminal node successively produces an influence;

[0073] The node sequence verified by the causal transmission characteristic verification is marked as a candidate association path, all the candidate association paths are collected, and the candidate association path set is obtained.

[0074] The present application converts the discrete decision nodes in the battlefield simulation training into structured knowledge representation with semantic association by constructing the knowledge graph based on the ontology model, and breaks through the limitation of the traditional method of relying only on the time sequence relationship for decision analysis. Through semantic modeling of entity nodes and relationship edges, the complex association between decision subjects, decision objects and decision behaviors can be accurately described, and the accuracy and completeness of decision relationship mining are significantly improved.

[0075] For example, the graph structure traversal rule is designed for battlefield decision causal chain characteristics. The direction constraint specifies that the path expansion is along the direction of the relationship edge, and is only expanded from the starting node to the terminal node, to reflect the transmission direction of the decision influence. For military command relationship, the path is expanded from the superior command node to the subordinate execution node. For the combat behavior chain, the path is expanded from the reconnaissance node to the strike node and then to the effect evaluation node. The depth constraint sets the maximum length of the path to five levels, that is, the node reached by expanding the starting node four times, covering the complete links of typical combat decision chains such as command decision, intelligence acquisition, firepower strike, force maneuver and effect evaluation. The depth setting is determined according to the statistical analysis of the length of the common decision chain in the battlefield simulation training, which not only ensures the integrity of the key causal path, but also avoids the exponential growth of the calculation complexity caused by the too long path.

[0076] The path expansion starts from the selected start node in the knowledge graph. The behavior node "E_CMD_20251119_R001" corresponding to the reconnaissance order issued by the red command post is selected as the starting point. The out-edge list of this node is read, and it is found that there is an execution relationship edge "R_EXEC_001" pointing to the reconnaissance unit node "E_UNIT_RECON_001". The traversal algorithm adds "E_UNIT_RECON_001" to the path, and the current path is "E_CMD_20251119_R001->R_EXEC_001->E_UNIT_RECON_001", and the path depth is one level. Continue to read the out-edge of "E_UNIT_RECON_001", and find that the execution relationship edge "R_EXEC_002" points to the reconnaissance behavior node "E_ACT_RECON_001". The node is added to the path, and the path is updated to "E_CMD_20251119_R001->R_EXEC_001->E_UNIT_RECON_001->R_EXEC_002->E_ACT_RECON_001", and the depth is two levels. The out-edge of the reconnaissance behavior node includes the reconnaissance relationship edge "R_RECON_001" pointing to the enemy target node "E_TARGET_ENEMY_001", and the reporting relationship edge "R_REPORT_001" pointing to the intelligence node "E_INTEL_001". The traversal algorithm follows the breadth-first strategy, and adds two branch paths to the expansion queue respectively.

[0077] For the path branch containing reconnaissance relationship "E_CMD_20251119_R001->...->E_ACT_RECON_001->R_RECON_001->E_TARGET_ENEMY_001", the decision state information of the adjacent node pair is extracted. The state information of the reconnaissance behavior node "E_ACT_RECON_001" includes the reconnaissance execution state as completed, the reconnaissance coverage range as 85% of the target area, and the intelligence acquisition quality as high credibility. The state information of the enemy target node "E_TARGET_ENEMY_001" includes the target exposure state as discovered, the target position accuracy as less than 50 meters of coordinate error, and the target threat level as medium threat. The battlefield causal rules are stored in the semantic constraint condition library, and the rule "if the coverage range of reconnaissance behavior is higher than 70% and the acquisition quality is high credibility, then the exposure state of the target node is affected" is applicable to the current node pair. It is determined that the reconnaissance behavior node has an influence relationship with the target node, the influence type is information acquisition influence, and the influence strength is calculated as 0.91 according to the coverage range and quality parameters.

[0078] For the path branch containing the reporting relationship, the state information of the reconnaissance behavior node and the intelligence node is extracted. The state information of the intelligence node "E INTEL 001" includes the intelligence content as the position of enemy armored forces, the intelligence timeliness as real-time intelligence, and the intelligence integrity as complete. The semantic constraint rule "if the reconnaissance behavior is executed and the acquisition quality is high, then the content integrity and timeliness of the intelligence node are directly affected" matches the current scenario. It is determined that there is an influence relationship, the influence type is information transmission influence, and the influence strength is 0.88. The path continues to expand, and the out-edge of the intelligence node includes the support relationship edge "R SUPPORT 001" pointing to the strike decision node "E CMD 20251119 S001". The state information of the strike decision node includes the decision basis as intelligence E INTEL 001, the decision timeliness as timely response, and the decision accuracy assessment as high probability of success. The constraint rule "if the timeliness of the intelligence node is real-time and the integrity is complete, then the accuracy and timeliness of the decision node are positively affected" is applicable. It is determined that the influence relationship is established, the influence type is decision support influence, and the influence strength is 0.93.

[0079] The path expansion process continues until the depth constraint is reached or there is no expandable out-edge.

[0080] The causal transmission characteristic verification is performed on the node sequence determined by the semantic constraint. The verification process detects whether there is a complete state transmission chain in the path, requiring that there is a confirmed influence relationship between each pair of adjacent nodes in the path, and the influence type meets the causal logic. The verification analysis of the above path is as follows. The reconnaissance instruction node to the reconnaissance detachment node is a command execution relationship, the state of the instruction directly affects the task receiving state of the detachment, and the influence relationship is confirmed. The reconnaissance detachment node to the reconnaissance behavior node is a task execution relationship, the combat readiness state of the detachment affects the execution quality of the behavior, and the influence relationship is confirmed. The reconnaissance behavior node to the intelligence node is an information generation relationship, the execution result of the behavior has a decisive influence on the content and quality of the intelligence, and the influence relationship is confirmed. The intelligence node to the strike decision node is a decision basis relationship, the accuracy and timeliness of the intelligence have a key influence on the correctness of the decision, and the influence relationship is confirmed. The strike decision node to the artillery battalion node is a task allocation relationship, the decision content directly affects the combat task of the unit, and the influence relationship is confirmed. The artillery battalion node to the fire strike node is a behavior execution relationship, the firepower capability of the unit affects the strength of the strike behavior, and the influence relationship is confirmed.

[0081] The state transmission chain integrity check requires that there is no transmission interruption in the path. Transmission interruption is defined as the connection between adjacent nodes although there is a relationship edge, but the state change of the previous node has no actual effect on the subsequent node, or the influence strength is lower than the minimum threshold 0.3. The verification algorithm traverses all adjacent node pairs in the path, extracts the influence strength value of each pair, and checks whether all are greater than or equal to 0.3. The influence strength in the above path is 0.95, 0.87, 0.91, 0.93, 0.89, 0.86 respectively, all meet the threshold requirement, and it is determined that there is no transmission interruption. The verification also checks the logical consistency of the influence type, which requires that the influence type sequence in the path conforms to the operational process specification. The type sequence of reconnaissance influence intelligence generation, intelligence support decision making, and decision driven action execution conforms to the military decision logic, and the logical consistency verification is passed.

[0082] The node sequence verified by the causal transmission characteristic is marked as a candidate associated path. Assign a unique identifier "PATH_CAND_20251119_001" to the path, record the node sequence, relationship edge sequence, strength value of each influence relationship, total length of the path, start and end node identifier, and other metadata contained in the path. The causal transmission score of the path is calculated by multiplying all the influence strength values, the score is 0.95 multiplied by 0.87 multiplied by 0.91 multiplied by 0.93 multiplied by 0.89 multiplied by 0.86, which is equal to 0.5568, representing the overall causal reliability of the path. The marked candidate path is stored in the candidate set.

[0083] The traversal process is performed for multiple starting nodes in the knowledge graph, covering decision nodes of different command levels and operational types. Starting from the defense deployment decision node of the blue command post, a defense decision chain path containing position construction, firepower configuration, early warning reconnaissance, interception preparation, etc. is identified. Starting from the attack decision node of the red armored battalion, an attack decision chain path containing target selection, route planning, fire cover, attack implementation, and battle result consolidation is identified. In the knowledge graph generated by a single confrontation training, dozens to hundreds of node sequences that meet the depth constraint and direction constraint can be identified. After semantic constraint judgment, the sequences that meet the influence relationship requirements are retained, and then filtered through the causal transmission characteristic verification, finally forming a candidate set containing 20 to 80 candidate associated paths. The length of the candidate path is distributed between two and five levels, and the causal transmission score is distributed between 0.3 and 0.9, providing input data for subsequent connectivity strength analysis and semantic consistency verification.

[0084] The application adopts a combination of graph structure traversal rules and semantic constraint conditions to identify a cause-effect transmission path, and overcomes the problem that a traditional correlation analysis method is susceptible to false correlation interference. By verifying the actual influence of a preceding node decision state on a subsequent node, a node sequence with a cause-effect transmission characteristic is screened, and a path adjacent in time but lacking a cause-effect relationship is effectively filtered, so that the identified decision correlation relationship has military logic rationality and interpretability.

[0085] In an optional embodiment, the cause-effect transmission characteristic verification on the node sequence determined by the semantic constraint condition comprises:

[0086] The entity nodes in the node sequence are time-sequentially sorted according to a time-sequential coordinate system after time-sequential alignment, a time-sequentially ordered node chain is obtained, decision state information of each entity node in the time-sequentially ordered node chain is extracted, and a state information sequence is constructed;

[0087] The state information sequence is traversed, a dependency relationship between a state change amount in a preceding state information and a state change amount in a subsequent state information is analyzed for adjacent state information pairs, and the dependency relationship represents whether the preceding state change amount is a trigger condition of the subsequent state change amount;

[0088] A state transmission matrix is constructed according to the dependency relationship, and the state transmission matrix records a state transmission relationship between each entity node in the time-sequentially ordered node chain;

[0089] Whether there is a continuous state transmission relationship from a starting node to a terminal node of the time-sequentially ordered node chain is detected based on the state transmission matrix, if the continuous state transmission relationship exists, it is determined that the node sequence has the cause-effect transmission characteristic, and the continuous state transmission relationship represents that there is no transmission interruption in a state transmission chain.

[0090] For example, a node sequence is subjected to semantic constraint judgment and then enters a time-sequential alignment processing stage. A battlefield simulation trainer adopts a unified simulation time coordinate system, and the time stamps of all decision nodes are recorded based on the coordinate system. A node sequence to be verified includes five entity nodes, which correspond to a red side reconnaissance unmanned aerial vehicle take-off decision, an unmanned aerial vehicle reconnaissance behavior execution, an enemy target discovery event, a red side fire distribution decision, and a cannon fire strike behavior. The original time stamps of the nodes are T+0120, T+0135, T+0142, T+0148, and T+0155, and the time unit is second. The time stamp fields of the nodes are read, and the nodes are arranged in ascending order according to the numerical values, so that a time-sequentially ordered node chain is obtained. After sorting, the node order is a reconnaissance decision node, a reconnaissance execution node, a target discovery node, a fire decision node, and a strike execution node, and the time stamp sequence is 0120, 0135, 0142, 0148, and 0155.

[0091] The state information extraction is performed for each entity node in the node chain. The decision state information of the reconnaissance decision node includes that the decision type is reconnaissance task allocation, the number of available UAVs is 3, the target area range is from 110 degrees east to 110.5 degrees east and from 40 degrees north to 40.5 degrees north, the reconnaissance priority is high priority, and the decision execution state is issued. The state information of the reconnaissance execution node includes that the reconnaissance payload type is an optical pod, the flight height is 5000 meters, the reconnaissance coverage area is 20 square kilometers, the reconnaissance duration is 12 minutes, and the number of targets obtained is 8 suspected targets. The state information of the target discovery node includes that the target type is an armored vehicle, the target position is 110.23 degrees east and 40.15 degrees north, the number of targets is 6, the recognition confidence is 0.85, and the threat level assessment is a medium threat. The state information of the fire decision node includes that the strike target selection is the armored vehicle identified by the target discovery node, the assigned fire unit is the third artillery battalion of the red side, the strike ammunition type is high-explosive shrapnel, the strike intensity is two waves, and the expected damage is a suppressive strike. The state information of the strike execution node includes that the actual number of fired ammunition is 12, the strike wave is two waves, the hit accuracy is a 50-meter circular probable error, the damage effect assessment is target area smoke coverage, and the execution completion state is strike end.

[0092] The state information sequence is constructed as an ordered array structure, and each element corresponds to a state information dictionary of a node. The first element in the sequence is the state of the reconnaissance decision node, and the key-value pairs include fields such as decision type, available resources, target range, priority, and execution state. The second element is the state of the reconnaissance execution node, including fields such as payload type, flight parameters, coverage area, duration, and number of discoveries. The third to fifth elements are the state information of the target discovery, fire decision, and strike execution nodes, respectively. The state information sequence corresponds to the timestamp sequence, supporting access to the state of each node in chronological order.

[0093] The state change amount analysis is realized by comparing the differences in key fields of adjacent state information pairs. The first pair of adjacent nodes is the reconnaissance decision node and the reconnaissance execution node. The decision execution state in the previous node state is issued, and the actual execution parameters such as flight height 5000 meters and coverage area 20 square kilometers appear in the subsequent node state. The state change amount is characterized by the transition of the decision from the issued state to the execution state, the transition of the resources from 3 available UAVs to 1 actually used UAV, and the transition of the task range from the planned coverage range to the actual coverage of 20 square kilometers. The dependency relationship is determined based on a pre-defined trigger rule library. The rule "if the previous node is a task decision and the execution state is issued, and the subsequent node is a corresponding task execution and actual execution parameters appear, then determine that the previous state change triggers the subsequent state change" is applicable to the current node pair. The issued state of the reconnaissance decision is confirmed as the trigger condition for the start of the reconnaissance execution, the dependency relationship is established, and the dependency strength is set to 0.95.

[0094] The second pair of adjacent nodes is the reconnaissance execution node and the target discovery node. The number of targets obtained by reconnaissance in the preceding node state is 8, and the coverage area is 20 square kilometers. In the subsequent node state, the target type is an armored vehicle, the number of targets is 6, and the confidence is 0.85. The state change amount is from a suspected target in the reconnaissance process to a confirmed identified target, and from 8 to 6 high-confidence vehicles. The trigger rule "if the number of targets discovered by the preceding node reconnaissance execution is greater than zero and the coverage range contains the location of the subsequent node target, then the reconnaissance result triggers target confirmation" is applicable. The confirmed reconnaissance execution discovery result is the trigger condition for target confirmation, the dependency relationship is established, and the dependency strength is calculated as 0.82 according to the confidence of 0.85 and the coverage rate.

[0095] The third pair of adjacent nodes is the target discovery node and the fire decision node. In the preceding node state, the target type is an armored vehicle, and the threat level is medium threat. In the subsequent node state, the strike target selection is the armored target, and the fire unit assigned is an artillery battalion. The state change amount is from target information in intelligence to strike object in combat decision, and the threat level affects the selection of strike intensity. The trigger rule "if the threat level of the target in the preceding node reaches medium and above, and the subsequent node is a strike decision for the target, then the target threat state triggers the strike decision" is applicable. The dependency relationship is established, and the dependency strength is 0.9.

[0096] The fourth pair of adjacent nodes is the fire decision node and the strike execution node. In the preceding node state, the strike intensity is two waves, and the ammunition type is high-explosive grenade. In the subsequent node state, 12 rounds are actually fired, and two waves are executed. The state change amount is from decision plan to actual execution, and the planned wave is consistent with the actual wave. The trigger rule "if the strike decision has been made in the preceding node and the fire unit assigned is clear, and the subsequent node is the strike execution of the corresponding unit, then the decision content triggers the execution behavior" is applicable. The dependency relationship is established, and the dependency strength is 0.93.

[0097] The state transfer matrix is constructed as a 5x5 two-dimensional array, with row and column indices corresponding to the 5 nodes in the node chain. The matrix element (i, j) stores the state transfer relationship strength from node i to node j, and when i and j are not adjacent, the element value is 0, and when they are adjacent and there is a dependency relationship, the element value is the dependency strength. The value of the first row and second column element of the matrix is 0.95, indicating the transfer strength from reconnaissance decision to reconnaissance execution. The value of the second row and third column element is 0.82, indicating the transfer strength from reconnaissance execution to target discovery. The value of the third row and fourth column element is 0.9, indicating the transfer strength from target discovery to fire decision. The value of the fourth row and fifth column element is 0.93, indicating the transfer strength from fire decision to strike execution. The remaining element values are 0, indicating that there is no direct transfer relationship.

[0098] The continuous state transfer path detection is based on matrix connectivity analysis. The detection algorithm starts from the starting node, i.e., the first node, and finds the non-zero element in the first row of the matrix. It is found that the value of the (1, 2) element is 0.95, and it is confirmed that there is a transfer to the second node. The second row of the matrix is continuously searched, and it is found that the value of the (2, 3) element is 0.82, and it is confirmed that there is a transfer to the third node. The third and fourth rows are sequentially searched, and it is confirmed that the values of the (3, 4) and (4, 5) elements are 0.9 and 0.93, respectively. A path 1->2->3->4->5 is formed from the first node to the fifth node, and all adjacent node pairs in the path have non-zero transfer strength in the matrix, and it is determined that there is a continuous state transfer relationship.

[0099] The transfer interruption detection verifies whether there is a link with a transfer strength of 0 or lower than the threshold value in the path. The transfer interruption threshold is set to 0.3, and a value lower than the threshold is considered to be too weak to constitute an interruption. The transfer strength sequence 0.95, 0.82, 0.9, 0.93 in the path is checked, and all values are greater than 0.3, and there is no link lower than the threshold. The detection also verifies whether there is a jump transfer, i.e., a situation where there is no transfer relationship between adjacent nodes but there is a transfer relationship between interval nodes. It is found that all non-zero elements are located in the corresponding positions of adjacent nodes in the matrix, and there is no jump transfer. The comprehensive determination is that the state transfer chain of the node sequence is complete and continuous, and there is no transfer interruption.

[0100] The final determination of the causal transfer characteristics of the node sequence is based on the existence of the continuous transfer relationship. The detection result confirms that there is a complete state transfer path from the starting node to the terminal node, and each state change in the path triggers the next state, and the transfer strength meets the threshold requirement, which logically constitutes a complete causal chain. It is marked that the node sequence has causal transfer characteristics, and a causal score is assigned to it. The causal score calculation uses the geometric mean of the transfer strength, and the fourth root of the four transfer strength values 0.95, 0.82, 0.9, and 0.93 is calculated to obtain a causal score of 0.897, representing the overall causal reliability level of the sequence.

[0101] Another to be verified node sequence contains red electronic interference decision, interference device power-on, blue communication interruption, blue command delay, blue defense response delay 5 nodes. After timing sorting, the state information is extracted, the first pair of inter-node dependence is that the interference decision triggers the device power-on, and the transmission strength is 0.92. The second pair of nodes is that the interference device power causes the communication interruption, and the transmission strength is 0.78. The third pair of nodes is that the communication interruption causes the command delay, and the transmission strength is 0.85. The fourth pair of nodes is originally the command delay causes the defense response delay, but analysis finds that the blue side starts a standby command channel during the command delay, the defense response delay is mainly caused by other factors, the triggering effect of the previous state change on the subsequent state is not obvious, and the transmission strength is only 0.25. The value is lower than the interruption threshold 0.3, it is judged that there is transmission interruption between the third and fourth nodes. The value of the (4, 5) unit in the state transmission matrix is 0.25, the detection algorithm finds that the transmission strength is insufficient at this position, and it is judged that the state transmission chain is not complete. The node sequence is judged as not having the causal transmission characteristic, and is not included in the candidate association path set.

[0102] The causal transmission characteristic verification provides a key screening mechanism for decision quality evaluation. Only the node sequence with a complete causal transmission chain can accurately reflect the transmission process of the decision influence, and support the scientific evaluation of the decision quality. The verification process identifies the real causal relationship, filters the pseudo-association and weak association, and improves the accuracy and interpretability of the subsequent decision association relationship extraction through steps such as timing alignment, state extraction, dependence analysis, matrix construction and connection detection.

[0103] The application realizes the double quality control of the candidate association path by analyzing the topological structure to calculate the path connection strength and verifying the causal logic rationality by combining semantic consistency verification. The connection strength quantifies the reliability of the decision influence transmission, and the semantic verification ensures that the path conforms to the battlefield operation law, and the synergistic effect of the two ensures that the finally extracted decision node association relationship has statistical significance and conforms to military professional knowledge, thereby providing high-quality data support for decision quality evaluation.

[0104] In an optional implementation, topological structure analysis and semantic consistency verification are performed on each association path in the candidate association path set, association paths that meet the connection strength threshold and pass the semantic consistency verification are screened out, and the decision node association relationship includes:

[0105] Extracting entity nodes and relationship edges in the association path, counting the number of relationship edges and the relationship strength information carried by the relationship edges, and calculating the connection strength of the association path based on the number of relationship edges and the relationship strength information;

[0106] The association paths with the connection strength greater than or equal to the connection strength threshold are screened out, and a path set meeting the connection strength requirement is obtained.

[0107] For each associated path in the set of paths satisfying the connectivity strength requirement, the relationship type identifier carried by the relationship edge between adjacent entity nodes in the associated path is extracted, and whether the relationship type identifier conforms to the semantic constraints of causal reasoning is verified according to a preset causal logic rule;

[0108] The associated path that passes the verification of the causal logic rule is marked, and the marked associated path is taken as an associated path that conforms to the connectivity strength threshold and passes the semantic consistency verification;

[0109] The starting entity node and the terminating entity node are extracted from the associated path that conforms to the connectivity strength threshold and passes the semantic consistency verification, and an associated relationship record between the starting entity node and the terminating entity node is established, the associated relationship record containing an associated path identifier and associated strength information, to obtain the decision node associated relationship.

[0110] For example, the set of candidate associated paths enters the topological structure analysis stage. A certain candidate path is identified as "PATH_CAND_20251119_015", which corresponds to the fire coordination decision chain of the red command post. The path metadata is read, and the entity node sequence and the relationship edge sequence contained by the path are extracted. The node sequence includes the decision node of the red command post, the combat unit node of the artillery battalion, the fire strike behavior node of the artillery, the combat unit node of the armored battalion, and the armored assault behavior node, a total of 5 entity nodes. The relationship edge sequence includes the command relationship edge from the command post to the artillery battalion, the execution relationship edge from the artillery battalion to the fire strike, the coordination relationship edge from the fire strike to the armored battalion, and the execution relationship edge from the armored battalion to the armored assault, a total of 4 relationship edges.

[0111] The number of relationship edges is 4. The relationship strength information is read from the attribute field of each relationship edge. The relationship strength carried by the command relationship edge is 0.92, reflecting the command authority and response timeliness of the command post to the artillery battalion. The strength of the execution relationship edge from the artillery battalion to the fire strike is 0.88, reflecting the ability and actual strike effect of the artillery battalion to execute the fire task. The strength of the coordination relationship edge from the fire strike to the armored battalion is 0.75, reflecting the support degree of the fire cover to the armored assault. The strength of the execution relationship edge from the armored battalion to the armored assault is 0.9, reflecting the execution efficiency of the armored battalion assault action.

[0112] Connectivity strength is calculated by multiplying the geometric mean of relation strengths by a path length attenuation factor. The geometric mean is the fourth root of four relation strengths: 0.92, 0.88, 0.75, and 0.9, yielding 0.86. The path length attenuation factor is determined by the number of edges; for four edges, the attenuation factor is 1 minus the number of edges multiplied by 0.05, i.e., 1 minus 0.2 equals 0.8. The connectivity strength is the geometric mean multiplied by the attenuation factor, i.e., 0.86 multiplied by 0.8 equals 0.688. This value characterizes the overall reliability of the path's connectivity; a higher value indicates a tighter connection between links in the path and more stable causal transmission.

[0113] The connectivity strength threshold is set to 0.5, which is determined based on the reliability requirements of the decision chain in battlefield simulation training. Paths below this threshold are considered to have insufficient connectivity and cannot be used as reliable evidence of causal association. The connectivity strength of the path "PATH_CAND_20251119_015" is 0.688, which is greater than the threshold of 0.5. Therefore, it is included in the set of paths that meet the connectivity strength requirements through connectivity strength screening.

[0114] Another candidate path, "PATH_CAND_20251119_023", contains 6 nodes and 5 relation edges. The nodes are: Blue Team reconnaissance node, intelligence transmission node, command and decision-making node, defense unit node, position construction behavior node, and early warning radar deployment node. The relation edge strengths are 0.65, 0.7, 0.58, 0.72, and 0.63, respectively. The geometric mean, calculated as the fifth root of the five strength values, is 0.653. The path length attenuation factor is 1 - 5 × 0.05 = 0.75. The connectivity strength is 0.653 × 0.75 ≈ 0.4898. This value is less than the threshold of 0.5, and therefore fails the connectivity strength screening, and is excluded from the candidate set.

[0115] The set of paths that meet the connectivity strength requirements enters the semantic consistency verification phase. For the path "PATH_CAND_20251119_015", the relationship type identifiers of the edges between adjacent nodes are extracted. The first edge is a command relationship, with the type identifier "CMD_COMMAND". The second edge is an execution relationship, with the type identifier "ACT_EXECUTE". The third edge is a cooperation relationship, with the type identifier "COOP_SUPPORT". The fourth edge is an execution relationship, with the type identifier "ACT_EXECUTE".

[0116] The causal logic rule base defines semantic constraint rules for battlefield decision chains. The rule base contains valid patterns for relation type sequences, such as rules like "command relationship can be followed by execution relationship or coordination relationship," "execution relationship can be followed by effect relationship or coordination relationship," and "coordination relationship can be followed by execution relationship." The verification algorithm checks whether the relation type sequences in the path conform to these patterns.

[0117] The type sequence of the path "PATH_CAND_20251119_015" is "CMD_COMMAND -> ACT_EXECUTE -> COOP_SUPPORT -> ACT_EXECUTE". The first pair of types is a command relationship to an execution relationship, which matches the rule "command relationship followed by an execution relationship", and passes the verification. The second pair of types is an execution relationship to a cooperation relationship, which matches the rule "execution relationship followed by a cooperation relationship", and passes the verification. The third pair of types is a cooperation relationship to an execution relationship, which matches the rule "cooperation relationship followed by an execution relationship", and passes the verification. The sequence of relationship types in the entire path logically forms a reasonable operational cooperation process: the commander issues an order, the artillery executes a fire strike, the fire covers the armored unit, and the armored unit executes a breakthrough, which conforms to the tactical logic of fire and breakthrough cooperation. The semantic consistency verification passes.

[0118] Another path "PATH_CAND_20251119_018" that satisfies the connectivity strength contains the relationship type sequence "CMD_COMMAND -> INFO_REPORT -> ACT_EXECUTE". The first pair of types is a command relationship to an intelligence reporting relationship, and the rule library is checked to find that "a command relationship is followed by an execution relationship or a cooperation relationship", which does not contain an intelligence reporting relationship. This type sequence does not conform to the cause-effect logic, because after the commander issues a command, an execution behavior should be triggered rather than an intelligence report. The semantic consistency verification fails, and this path is excluded.

[0119] The cause-effect logic rules also verify the rationality of the relationship direction. The rule "the starting point of a strike relationship should be a combat unit or a strike behavior, and the endpoint should be an enemy target" is used to constrain the use of strike relationships. If a strike relationship edge appears in a path, the verification algorithm checks whether the starting node type of the edge is a combat unit or a strike behavior category, and whether the endpoint node type is an enemy target category. A certain candidate path contains a strike relationship edge, the starting point of which is a red artillery fire strike behavior node, and the endpoint of which is a blue armored target node, which conforms to the rule constraint and passes the verification. If the starting point of a strike relationship edge in a path is an intelligence node and the endpoint is a friendly unit node, it violates the rule and fails the verification.

[0120] The paths that pass the semantic consistency verification are marked. The marking process adds a "semantic_validated" field in the path metadata, sets the value to true, and records the timestamp of the verification passing and the verification rule version. The path "PATH_CAND_20251119_015" is marked as a relevant path that meets the connectivity strength threshold and passes the semantic consistency verification, and the status is updated to "VALIDATED".

[0121] The decision node association relationship extraction is performed on the verified path. The start node and end node identifiers of the path "PATH_CAND_20251119_015" are read. The start node is the red command post decision node "E_CMD_HQ_20251119_F001", and the end node is the armored assault action node "E_ACT_ASSAULT_001". An association record is created, and the record structure includes fields such as association identifier, start node identifier, end node identifier, association path identifier, association strength, association type, and establishment time. The association identifier is generated as "REL_20251119_F001_TO_ASSAULT_001", the start node identifier is "E_CMD_HQ_20251119_F001", the end node identifier is "E_ACT_ASSAULT_001", the association path identifier is "PATH_CAND_20251119_015", the association strength is the path connectivity strength 0.688, the association type is marked as "CAUSAL_CHAIN", and the establishment time is the current timestamp.

[0122] The association record is stored in the association table of the decision knowledge base. The table adopts a relational database structure and supports queries by start node, end node, and association strength. After the record is inserted, the relevant indexes are updated, including the start node index, the end node index, and the strength range index. The query module can quickly retrieve all subsequent action nodes triggered by the decision of the command post decision node, or it can query the source of the decision that leads to the action according to the assault action node.

[0123] Multiple paths have the same start and end nodes but pass through different intermediate nodes. The path "PATH_CAND_20251119_015" and the path "PATH_CAND_20251119_027" both start from the red command post decision node and finally reach the armored assault action node, but the former passes through artillery fire support, and the latter passes through air strike support. Association records are established for the two paths, and the association identifiers are "REL_20251119_F001_TO_ASSAULT_001_VIA_ARTY" and "REL_20251119_F001_TO_ASSAULT_001_VIA_AIR", respectively. The association strengths of the two records are 0.688 and 0.72, respectively, reflecting the reliability differences of different coordination methods. The decision analysis module can compare the strengths of different paths to identify the optimal coordination scheme.

[0124] The association relationship record supports multi-dimensional analysis of decision quality evaluation. The evaluation module counts the number of association relationships of each decision node as a starting node, reflecting the influence range of the decision. The red command post establishes 15 association relationships in a comprehensive decision, involving firepower attack, troop movement, intelligence reconnaissance, logistics support and other fields, indicating that the decision has a wide range of influence. The evaluation module calculates the average strength of the association relationship, reflecting the reliability of the decision influence. The average strength of the association relationship of a certain decision node is 0.78, higher than the global average of 0.65, indicating that the execution effect of the decision is stable and reliable. The evaluation module also analyzes the average length of the association path, and a shorter path indicates that the transmission chain from decision to effect is compact and the response is rapid.

[0125] After a single confrontation training is completed, all decision node association relationships are summarized to generate a decision causal network diagram. The nodes in the network diagram represent decision entities, the edges represent association relationships, the thickness of the edges reflects the association strength, and the color of the edges identifies the association type. During training review, commanders can visually view decision links, analyze the influence range and transmission path of key decision points, identify decision error links, and optimize subsequent training schemes. The decision node association relationship provides an objective and quantitative basis for training quality evaluation, supporting data-based training improvement and tactical optimization.

[0126] The decision node association relationship record established by the present application covers multi-dimensional information such as association path and association strength, supports comprehensive quality evaluation and traceability analysis of training decisions, and commanders can intuitively identify decision error links, analyze decision influence range and transmission path, and quantitatively evaluate the execution effect of different decision schemes, providing a scientific basis for training review and summary and tactical scheme optimization, effectively improving the quality and efficiency of battlefield simulation training.

[0127] In a second aspect of the embodiment of the present application, a structural schematic diagram of a discrete node decision quality evaluation system based on a training device is provided, and the system comprises:

[0128] A first unit is configured to acquire a plurality of discrete decision nodes in a battlefield simulation training environment, wherein the discrete decision nodes include decision subject identification, decision time sequence identification and decision state information.

[0129] A second unit is configured to perform semantic modeling on the discrete decision nodes based on a preset ontology model to obtain a knowledge graph comprising entity nodes and relationship edges, wherein the entity nodes represent decision subjects, decision objects and decision behaviors, and the relationship edges represent semantic associations between the entity nodes.

[0130] A third unit is configured to identify a node sequence meeting a causal transmission feature in the knowledge graph based on a graph structure traversal rule and a semantic constraint condition, wherein the causal transmission feature represents that the decision state information of a previous node has an influence on the decision state information of a subsequent node, and a candidate association path set is obtained.

[0131] A fourth unit is configured to perform topology analysis and semantic consistency verification on each of the candidate association paths in the candidate association path set, the topology analysis is configured to calculate the connectivity strength between nodes in the path, and the semantic consistency verification is configured to verify whether the relationship edges of adjacent nodes in the path conform to preset causal logic rules, and the association path that meets the connectivity strength threshold and passes the semantic consistency verification is screened out to obtain the decision node association relationship.

[0132] A third aspect of the embodiments of the present application,

[0133] An electronic device is provided, comprising:

[0134] A processor;

[0135] A memory for storing processor-executable instructions;

[0136] The processor is configured to invoke the instructions stored in the memory to execute the method described above.

[0137] A fourth aspect of the embodiments of the present application,

[0138] A computer-readable storage medium is provided, which stores computer program instructions, and the computer program instructions are executed by a processor to implement the method described above.

[0139] The present application can be a method, device, system and / or computer program product. The computer program product can include a computer readable storage medium having computer readable program instructions stored therein, which are used to perform various aspects of the present application.

[0140] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application, and not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that: it can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement to part or all of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for evaluating the quality of decisions of discrete nodes based on a trainer, characterized in that, include: Multiple discrete decision nodes are acquired in a battlefield simulation training environment. Each discrete decision node includes a decision subject identifier, a decision time sequence identifier, and decision status information. Based on a pre-defined ontology model, semantic modeling is performed on the discrete decision nodes to obtain a knowledge graph containing entity nodes and relation edges. The entity nodes represent decision subjects, decision objects, and decision behaviors, and the relation edges represent the semantic associations between entity nodes. Based on graph structure traversal rules and semantic constraints, node sequences satisfying causal transitivity are identified in the knowledge graph. Causal transitivity represents the influence of the decision state information of preceding nodes on the decision state information of subsequent nodes, resulting in a candidate association path set, including: The path is extended along the relation edges according to the graph structure traversal rules. During the path extension process, the entity nodes and relation edges traversed are recorded to form a node sequence. The graph structure traversal rules define the direction constraints and depth constraints of the path extension. For adjacent entity nodes in the node sequence, the decision state information of the preceding entity node and the decision state information of the subsequent entity node are extracted. Based on the semantic constraints, it is determined whether the decision state information of the preceding entity node has an influence relationship on the decision state information of the subsequent entity node. The node sequence determined by the semantic constraints is subjected to causal transitivity verification. The causal transitivity verification is used to detect whether there is a complete state transit chain in the node sequence. The state transit chain represents the transit path in which the decision state information from the starting node to the ending node sequentially generates influence. The node sequences verified by the causal transitivity property are marked as candidate association paths, and all the candidate association paths are collected to obtain the candidate association path set; For each associated path in the candidate associated path set, topological structure analysis and semantic consistency verification are performed. The topological structure analysis is used to calculate the connectivity strength between nodes in the path, and the semantic consistency verification is used to verify whether the relationship edges between adjacent nodes in the path conform to the preset causal logic rules. Associated paths that meet the connectivity strength threshold and pass the semantic consistency verification are selected to obtain the association relationship of the decision nodes.

2. The method of claim 1, wherein, Based on a pre-defined ontology model, semantic modeling is performed on the discrete decision nodes to obtain a knowledge graph containing entity nodes and relation edges, including: The decision state information in the discrete decision nodes is analyzed to identify the decision attributes and decision content. The decision attributes are mapped to entity type labels, and the decision content is mapped to the attribute values ​​of the entity nodes. According to the entity classification rules defined in the preset ontology model, decision nodes carrying entity type labels and attribute values ​​are converted into entity nodes. The entity classification rules define the hierarchical relationship between decision subjects, decision objects, and decision behaviors in the ontology model. Based on the predefined relation patterns in the preset ontology model, the semantic dependencies between the entity nodes are analyzed, the connection methods between the entity nodes are determined according to the semantic dependencies, and relation edges representing the connection methods are generated. The relation edges carry relation type identifiers and relation strength information. The entity nodes and relation edges are organized into a graph data structure to obtain the knowledge graph containing entity nodes and relation edges.

3. The method according to claim 1, characterized in that, Verifying the causal transitivity of the node sequence determined by the semantic constraints includes: The entity nodes in the node sequence are sorted temporally according to the temporally aligned temporal coordinate system to obtain a temporally ordered node chain. The decision state information of each entity node in the temporally ordered node chain is extracted to construct a state information sequence. Traverse the state information sequence, and for adjacent state information pairs, analyze the dependency relationship between the state change in the preceding state information and the state change in the subsequent state information. The dependency relationship characterizes whether the state change in the preceding state is a triggering condition for the state change in the subsequent state. A state transit matrix is ​​constructed based on the dependencies, and the state transit matrix records the state transit relationships between entity nodes in the temporally ordered node chain; Based on the state transit matrix, it is detected whether there is a continuous state transit relationship from the starting node to the ending node of the temporally ordered node chain. If the continuous state transit relationship exists, it is determined that the node sequence has causal transit characteristics. The continuous state transit relationship indicates that there is no transit interruption in the state transit chain.

4. The method according to claim 1, characterized in that, For each associated path in the candidate associated path set, topological structure analysis and semantic consistency verification are performed. Associated paths that meet the connectivity strength threshold and pass the semantic consistency verification are selected, resulting in the following decision node association relationships: Extract the entity nodes and relationship edges in the associated path, count the number of relationship edges and the relationship strength information carried by the relationship edges, and calculate the connectivity strength of the associated path based on the number of relationship edges and the relationship strength information; Filter out associated paths with connectivity strength greater than or equal to the connectivity strength threshold to obtain a set of paths that meet the connectivity strength requirement; For each associated path in the set of paths that meet the connectivity strength requirements, the relation type identifier carried by the relation edge between adjacent entity nodes in the associated path is extracted, and the relation type identifier is verified to meet the semantic constraints of causal reasoning according to the preset causal logic rules. The associated paths verified by the causal logic rules are marked, and the marked associated paths are regarded as associated paths that meet the connectivity strength threshold and pass the semantic consistency check; Extract the starting entity node and the ending entity node from the association path that meets the connectivity strength threshold and passes the semantic consistency check, establish the association relationship record between the starting entity node and the ending entity node, the association relationship record includes the association path identifier and association strength information, and obtain the association relationship of the decision node.

5. A discrete node decision quality evaluation system based on a trainer, used to implement the method as described in any one of claims 1-4, characterized in that, include: The first unit is used to acquire multiple discrete decision nodes in a battlefield simulation training environment. The discrete decision nodes include decision subject identifiers, decision time sequence identifiers, and decision status information. The second unit is used to perform semantic modeling on the discrete decision nodes based on a preset ontology model, thereby obtaining a knowledge graph containing entity nodes and relation edges. The entity nodes represent decision subjects, decision objects, and decision behaviors, and the relation edges represent the semantic associations between entity nodes. The third unit is used to identify node sequences that satisfy causal transitivity in the knowledge graph based on graph structure traversal rules and semantic constraints. Causal transitivity represents the influence of the decision state information of the preceding node on the decision state information of the subsequent node, thereby obtaining a set of candidate association paths. The fourth unit is used to perform topological structure analysis and semantic consistency verification on each associated path in the candidate associated path set. The topological structure analysis is used to calculate the connectivity strength between nodes in the path, and the semantic consistency verification is used to verify whether the relationship edges of adjacent nodes in the path conform to the preset causal logic rules. The associated paths that meet the connectivity strength threshold and pass the semantic consistency verification are selected to obtain the association relationship of the decision nodes.

6. An electronic device, characterized in that, include: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the method according to any one of claims 1 to 4.

7. A computer-readable storage medium having computer program instructions stored thereon, characterized in that, When the computer program instructions are executed by the processor, they implement the method described in any one of claims 1 to 4.

Images