A multi-modal cross-domain war game deduction decision data construction method and device based on trajectory logic migration

Abstract

The application provides a multi-modal cross-domain war game deduction decision data construction method and device based on trajectory logic migration, which defines a decision trajectory structured data model containing a time stamp, a command seat, an operation type, a multi-modal information source and a semantic context, uniformly analyzes text instructions, voice communication records and graphic annotation data in a distributed war game deduction system into standardized decision trajectory segments, performs space-time alignment and logic fusion on trajectory segments of different deduction sessions and seats to form panoramic decision process records, and then performs mode mining and semantic annotation on the basis of a military rule base to generate labeled decision trajectories, searches for candidate historical trajectory segments according to new scenario constraint conditions, and completes synthesis through trajectory logic migration and three consistency checks, so as to guarantee the tactical authenticity and logical self-consistency of the synthesized decision trajectory data.

Description

Technical Field

[0001] This invention relates to the field of military simulation, and in particular to a method and apparatus for constructing multimodal cross-domain wargaming decision data based on trajectory logic transfer. Background Technology

[0002] Distributed wargaming systems are widely used in joint training, generating multi-source heterogeneous decision-making data for each exercise, including command and control instructions, voice communication records, and electronic sand table annotations. This data is scattered across independent systems in different exercise sessions and command positions, exhibiting varying recording formats, inconsistent time bases, and inconsistent semantic expressions. When these historical exercise data are needed to provide decision-making references for new training scenarios, existing methods typically involve directly extracting and splicing historical decision fragments. However, due to significant differences in operational pace, troop composition, and battlefield environment between different exercises, simple splicing disrupts the inherent spatiotemporal logic, causal chains, and resource continuity of the original decision sequence. This leads to logical contradictions in the synthesized results, such as troops being assigned to another battlefield before completing their maneuvers, or the decision sequence ignoring losses and ammunition consumption already incurred in previous stages—all contrary to common military sense.

[0003] In view of the above, this application is hereby submitted. Summary of the Invention

[0004] This invention discloses a method and apparatus for constructing multimodal cross-domain wargaming decision data based on trajectory logic transfer. It aims to solve the technical problem that when synthesizing new training scenarios using historical wargaming data, the lack of structured and unified representation and cross-event logical alignment of multi-source heterogeneous decision data leads to the destruction of the spatiotemporal logical relationships, causal dependency chains and resource state continuity in the original decision sequence by directly splicing historical decision fragments, thereby causing the synthesized data to lose tactical authenticity and logical consistency.

[0005] This invention provides a method for constructing multimodal cross-domain wargaming decision data based on trajectory logic transfer, including:

[0006] Define a structured data model for decision trajectories that includes timestamps, command positions, operation types, multimodal information sources, and semantic context. Based on the data model, perform structured parsing and semantic alignment on text instructions, voice communication records, and graphic annotation data in a distributed wargaming system to generate standardized decision trajectory fragments.

[0007] The standardized decision trajectory fragments from different simulation sessions and command positions are spatiotemporally aligned and logically fused to form a panoramic record of the decision-making process;

[0008] Based on the military rule base, pattern mining and semantic annotation are performed on the panoramic decision-making process records to generate tagged decision trajectories containing tactical intention labels, decision effectiveness levels, and trigger condition features.

[0009] The system receives the constraints of the newly proposed scenario, retrieves candidate historical trajectory fragments from the labeled decision trajectory, and synthesizes decision trajectory data through trajectory logic transfer and triple consistency verification.

[0010] Based on the commander's historical decision-making style profile and real-time situation simulation, personalized and exploratory decision-making knowledge is pushed to the synthesized decision-making trajectory data.

[0011] Based on the multimodal cross-domain wargaming decision data construction method and apparatus provided by this invention, a structured data model of decision trajectory including timestamps, command positions, operation types, multimodal information sources, and semantic context is defined. This model unifies the parsing of text instructions, voice communication records, and graphic annotation data in the distributed wargaming system into standardized decision trajectory fragments. Then, the trajectory fragments of different wargaming sessions and positions are spatiotemporally aligned and logically fused to form a panoramic decision process record. Subsequently, pattern mining and semantic annotation are performed on these fragments based on a military rule base to generate labeled decision trajectories. On this basis, candidate historical trajectory fragments are retrieved according to new scenario constraints and synthesized through trajectory logic transfer and triple consistency verification, thereby ensuring the tactical authenticity and logical consistency of the synthesized decision trajectory data. Attached Figure Description

[0012] Figure 1 This is a flowchart illustrating a method for constructing multimodal cross-domain wargaming decision data based on trajectory logic transfer, provided in the first embodiment of the present invention.

[0013] Figure 2 This is a schematic diagram of a module for constructing multimodal cross-domain wargaming decision data based on trajectory logic migration, provided in the second embodiment of the present invention. Detailed Implementation

[0014] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0015] To better understand the technical solution of the present invention, the embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0016] Please see Figure 1This invention provides a method for constructing multimodal cross-domain wargame decision data based on trajectory logic transfer. This method can be executed by a construction device or system, specifically by one or more processors within the construction device, to at least implement the following steps:

[0017] S101, define a structured data model of decision trajectory including timestamp, command position, operation type, multimodal information source and semantic context, and perform structured parsing and semantic alignment on text instructions, voice communication records and graphic annotation data in the distributed wargaming system based on the data model to generate standardized decision trajectory fragments.

[0018] In this embodiment, a structured data model for the decision trajectory is first defined. This model includes five core fields: timestamp, command position, operation type, multimodal information source, and semantic context. The timestamp records the precise moment the decision occurred using the ISO 8601 standard format, such as "2023-11-25T14:30:22.000Z". The command position records the standard name of the command organization that issued the decision, such as "Command Post of the 1st Mobile Brigade of Party A". The operation type identifies the action category of the decision, such as "support request", "force mobilization", or "information acquisition deployment". The multimodal information source includes three types of data carriers: text instructions, voice clips, and graphic annotations. The semantic context records background information such as the task stage, current task, and associated units at the time the decision occurred, for example, {"task stage":"action preparation","current task":"opening a path","associated units":["support unit","information acquisition unit"]}.

[0019] Based on the defined data model, multi-source heterogeneous data from different subsystems in the distributed simulation system are identified and extracted. In a typical distributed simulation, decision data is distributed and stored in multiple independent systems: historical action instruction records exported from the client's command system include time, seat, and instruction content, for example, {"Time":"2023-11-25 22:30:22","Seat":"Client Mobile Brigade","Instruction Content":"1st Platoon moves to Area A3"}; voice communication archives obtained from the communication record system include call time, speaker, and audio file, for example, {"Call Time":"2023-11-25T14:30:22","Speaker":"Client Mobile Brigade Command Post","Audio File":"Support Request Recording.wav"}; and annotation data exported from the joint simulation electronic sand table includes annotation time, annotation user, and layer file, for example, {"Annotation Time":"14:30:22","Annotation User":"Brigade Commander","Layer File":"Main Direction Annotation.xml"}.

[0020] Because the data from the aforementioned systems differ in time format, position names, and other aspects, format standardization is necessary. Regarding time information, the varying time representations across systems will be converted to the ISO 8601 standard format. For example, "2023-11-25 22:30:22" and "14:30:22" will both be converted to "2023-11-25T14:30:22.000Z". Regarding position names, different representations of the same command position across different systems will be mapped to a standardized position name. For example, "Party A Mobile Brigade" and "Party A Mobile Brigade Command Post" will be mapped to "Party A 1st Mobile Brigade Command Post". After format standardization, the unified timestamps will be used to associate the instruction data, voice data, and annotation data extracted from different systems at the same decision-making time, forming a unified data view. For example, at the timestamp "2023-11-25T14:30:22.000Z", the mobilization instructions from the command system, the voice support request from the communication system, and the main direction markings from the electronic sand table are associated as a multi-source data set at the same decision moment.

[0021] After forming a unified data view, each type of data within it undergoes structured parsing. Voice communication records are processed to convert audio files ("Support Request Recording.wav") into structured text data, such as {"Text Content": "Requesting support forces to cover area B2", "Confidence Level": 0.95}. Graphical annotation data is parsed semantically, identifying and converting graphic elements in the layer file "Main Direction Annotation.xml" into semantic descriptions, such as parsing arrows and circles into {"Annotation Type": "Main Direction", "Direction": "Northeast", "Target Area": ​​"C5 Area"}. Text commands are analyzed structurally, extracting key elements from the command content "Squad 1 moves to area A3" using natural language processing, forming {"Action": "Mobilization", "Executing Unit": "Squad 1", "Target": "Area A3"}.

[0022] Finally, the parsing results of each modality are precisely correlated with the corresponding standardized timestamps, command positions, and semantic information inferred from the task context to generate standardized decision trajectory fragments that conform to the structured data model of decision trajectories. Taking the above decision moment as an example, the generated standardized decision trajectory fragment is as follows: the timestamp of this fragment is "2023-11-25T14:30:22.000Z", the command position is "Command Post of the 1st Mobile Brigade of Party A", the operation type is "force mobilization", the text instruction in the multimodal information source includes the original content "1st detachment mobilizes to area A3" and its structured parsing results, the voice fragment includes an audio file reference and the transcribed text "Request support forces to cover area B2" and a confidence level of 0.95, the graphic annotation includes a layer file reference and the semantic description of "main direction, northeast, area C5", and the semantic context records that the current task stage is "action preparation", the current task is "opening a passage", and the associated units are the support detachment and the information acquisition detachment. This standardized decision trajectory fragment serves as the input data for the spatiotemporal alignment and logical fusion in step two.

[0023] S102, perform spatiotemporal alignment and logical fusion of the standardized decision trajectory segments from different simulation sessions and command positions to form a panoramic record of the decision-making process;

[0024] First, the standardized decision trajectory fragments generated above are preprocessed for normalization. The system receives standardized decision trajectory fragments from multiple simulations, sorts and groups them according to their timestamps, command positions, and event sequence numbers, and establishes a unified metadata index for different simulation sessions. For example, the system received trajectory segments from three different simulations: a segment from the 2023 "Exercise-01" simulation, with a timestamp of "2023-11-25T14:30:22.000Z", a command position of "Party A's 1st Mobile Brigade Command Post", and an operation type of "force mobilization"; a segment from the 2023 "Joint-07" simulation, with a timestamp of "2023-08-10T09:15:00.000Z", a command position of "Party B's Mechanized Infantry Brigade Command Post", and an operation type of "support request"; and a segment from the 2024 "Action-03" simulation, with a timestamp of "2024-03-18T16:45:30.000Z", a command position of "Party A's 1st Mobile Brigade Command Post", and an operation type of "information acquisition and deployment". The metadata index, built according to the simulation session, seats, and time range, records the time range, participating seats, and total number of segments for each simulation session. For example, the time range of "Exercise-01" is from "2023-11-25T08:00:00" to "2023-11-25T18:00:00", the participating seats include "Command Post of the 1st Mobile Brigade of Party A" and "Command Post of the Support Unit of Party A", and the total number of segments is 156.

[0025] After establishing the metadata index, an intelligent alignment method integrating dynamic time warping algorithm and rule constraints is adopted to perform cross-schedule spatiotemporal alignment of trajectory sequences from different simulation sessions. Since the action rhythms of different simulation sessions vary significantly, simply aligning by absolute time or stage labels cannot reflect the true correspondence of task progress. Therefore, a multi-dimensional feature sequence including operation density, state change rate, event interval distribution, and resource consumption rate is first constructed for the trajectory data of each simulation session. Operation density is represented by the number of instructions per unit time through a sliding window, indicating decision intensity; the state change rate is reflected by the dynamic nature of the situation through differential calculation of the update frequency of the position and state of each action unit; the event interval distribution reveals the action rhythm pattern through sequence statistical analysis of the time intervals between key decision nodes; and the resource consumption rate measures the sustainability of action intensity by fitting a material consumption curve using a cumulative function. For example, regarding the trajectory of the 1st Mobile Brigade of Party A in the two simulations "Exercise-01" and "Action-03", the constructed feature sequence shows that "Exercise-01" is a high-intensity, fast-paced mode. In the first 30 minutes after the start of the simulation, the command density is 12 times per minute, the state change rate is 8 times per minute, the average event interval is 2.5 minutes, and the first contact time is 28 minutes after the start. "Action-03" is a steady advancement mode. In the first 60 minutes after the start of the simulation, the command density is 3 times per minute, the state change rate is 2 times per minute, the average event interval is 20 minutes, and the first contact time is 105 minutes after the start.

[0026] After constructing the multidimensional feature sequence, the eigenvector distance between any two time points in the two simulation trajectories is calculated to construct a cost matrix. This eigenvector encompasses the operation type, action unit state, and the aforementioned rhythm quantification index. The minimum cumulative cost path from the start point to the end point is searched within the cost matrix as the optimal matching relationship. This path allows for non-linear "stretching" or "compression" of the sequence on the time axis, thus adapting to rhythm differences between different simulations. The local slope of the path reflects the rhythm ratio between the two simulations in the corresponding time period. For example, the dynamic time warping algorithm calculates that the first 30 minutes of "Exercise-01" and the first 90 minutes of "Action-03" have an optimal matching relationship, with a rhythm ratio of approximately 1:3. The specific mapping relationships are as follows: the information gathering force forward deployment action at the 10th minute after the start of "Exercise-01" matches the same action at the 35th minute after the start of "Operation-03", with a rhythm mapping ratio of 1:3.5; the first contact at the 28th minute after the start of "Exercise-01" matches the first contact at the 105th minute after the start of "Operation-03", with a rhythm mapping ratio of 1:3.8; the support request at the 30th to 35th minute after the start of "Exercise-01" matches the support coordination at the 110th to 130th minute after the start of "Operation-03", with a rhythm mapping ratio of 1:4 based on the similarity of operational density.

[0027] The matching results generated by dynamic time warping need to undergo a rationality check based on rule constraints to ensure that the alignment results are logically valid within the task. The check includes three constraints: First, a stage constraint check, requiring that the matching points maintain the evolutionary order of the task stages, and prohibiting cross-stage misalignment where the matching point moves from the preparation stage to the execution stage; second, a causal relationship constraint check, requiring that events with logical dependencies maintain the correct temporal order after alignment, for example, the information acquisition action must precede the coverage action, and this order cannot be reversed after alignment; third, a rhythm boundary constraint check, requiring that the rhythm compression ratio of different action types be within a reasonable range, for example, the rhythm compression ratio of the contact interaction stage should not exceed 1:5. Taking the aforementioned matching results as an example, the phase sequence check shows that both simulations follow the phase sequence of "information acquisition, contact, and support coordination," and the sequence remains unchanged after alignment, thus it is considered successful. The causal relationship check shows that the information acquisition force is deployed before the first contact, and the sequence is maintained after alignment, thus it is considered successful. The rhythm boundary check shows that the information acquisition time before the first contact in "Drill-01" is 10 minutes, while the corresponding segment in "Action-03" is 70 minutes, with a rhythm ratio of 1:7, which exceeds the 1:5 boundary specified by the rule base, thus triggering an adjustment.

[0028] When the verification fails, the system performs iterative optimization based on constraint relaxation. The optimization mechanism includes three strategies: adding constraint relaxation terms to the cost matrix and re-searching for suboptimal paths that satisfy the boundary conditions; dividing the entire deduction process into task stages, with each stage performing dynamic time warping independently while maintaining temporal constraints between stages; and calling special processing rules from the expert rule base for critical nodes that repeatedly fail the verification, such as "the tolerance upper limit for the first contact time difference between high-intensity rapid action and robust advancement is 1:5". After iterative optimization, the final alignment results for "Exercise-01" and "Action-03" met the requirements of rationality: the information acquisition and deployment phase corresponds to 0 to 15 minutes after the start of "Exercise-01" and 0 to 60 minutes after the start of "Action-03", with a pace ratio of 1:4.0; the initial contact phase corresponds to 15 to 30 minutes after the start of "Exercise-01" and 60 to 120 minutes after the start of "Action-03", with a pace ratio of 1:4.0; the support and coordination phase corresponds to 30 to 45 minutes after the start of "Exercise-01" and 120 to 165 minutes after the start of "Action-03", with a pace ratio of 1:3.0; and the adjustment phase corresponds to 45 to 60 minutes after the start of "Exercise-01" and 165 to 210 minutes after the start of "Action-03", with a pace ratio of 1:3.0.

[0029] The alignment results that pass the verification are converted into a unified logical timeline with rhythm labels. This timeline uses one simulation as a baseline and records the rhythm mapping function and specific alignment point information of the other simulation at each stage. For example, using "Exercise-01" as the baseline simulation, the unified logical timeline records that the rhythm mapping function coefficient for the information acquisition and contact phases is 4.0, and the rhythm mapping function coefficient for the support and coordination phase is 3.0. It also records typical alignment points, such as the information acquisition force advance event at the 10th minute after the start of "Exercise-01" corresponding to the 40th minute after the start of "Action-03" and labeled as "fast rhythm 1:4"; the first contact event at the 28th minute after the start of "Exercise-01" corresponding to the 112th minute after the start of "Action-03" and also labeled as "fast rhythm 1:4"; and the support request event at the 32nd minute after the start of "Exercise-01" corresponding to the 130th minute after the start of "Action-03" and labeled as "medium rhythm 1:3".

[0030] After completing the spatiotemporal alignment across different scenarios, the consistency of state changes and command logic of multiple positions' trajectories within the same time period is verified using a rule base. For detected conflicts, a weighted arbitration mechanism based on command hierarchy, data completeness, and time accuracy is used for logical fusion. For example, near the timestamp "2023-11-25T14:30:22.000Z" in the "Exercise-01" simulation, a trajectory fragment from the 1st Mobile Brigade Command Post of the client side shows a "force mobilization" operation, i.e., detachment 1 moving to area A3; while a trajectory fragment from the support detachment Command Post of the client side shows a "support request" operation, i.e., covering area A3, with the context including intelligence that "unmanned information acquisition equipment reports an enemy cluster in area A3." According to the rule base verification, the rule stipulates that "friendly forces should not be mobilized to areas where friendly forces are currently covering." The two trajectories overlap spatiotemporally, and their intentions contradict each other. The system then initiated weighted arbitration: In terms of command hierarchy, the brigade command post is higher than the platoon command post, and brigade-level instructions have higher weight; in terms of data completeness, brigade-level instructions contain explicit mobilization orders, while platoon requests contain specific intelligence evidence; in terms of time precision, both timestamps have the same precision. Based on the weighted calculations and referencing the rule in the rule base that "when coverage operations conflict with force mobilization, priority is given to ensuring coverage security," the system generated a fused decision record: at this timestamp, the primary event is "coverage operations take priority," based on the unmanned information acquisition equipment confirming a threat from a cluster of adversaries in area A3. The conflict resolution is to cancel the platoon's mobilization order to area A3 and instead reroute it to area A2. The current mission phase is "operation preparation."

[0031] After the above-mentioned normalization preprocessing, cross-segment spatiotemporal alignment and multi-source trajectory logical fusion processing, a panoramic decision-making process record is formed that eliminates logical conflicts and integrates multi-perspective information. This record fully presents the decision-making actions of each command position and their interrelationships on a unified logical timeline, which can be used for pattern mining and semantic annotation in step three.

[0032] S103, Based on the military rule base, perform pattern mining and semantic annotation on the panoramic decision-making process record to generate a labeled decision trajectory containing tactical intention labels, decision effectiveness levels and trigger condition features;

[0033] First, key decision-making node sequences and their corresponding semantic contexts are extracted from the panoramic decision-making process records. A decision-state-temporal correlation analysis matrix is ​​then established based on the extracted decision-making node sequences and semantic contexts. Taking a decision-making process of the 1st Mobile Brigade of the client in the "Exercise-01" simulation as an example, the extracted key decision-making node sequences include: Node T1 issues an order for information acquisition forces to advance at "2023-11-25T14:20:00"; Node T2 receives a report from unmanned information acquisition equipment confirming the presence of an enemy cluster in area A3 at "2023-11-25T14:25:00"; Node T3 decides to cancel the original mobilization plan and prioritize coverage operations at "2023-11-25T14:30:22"; and Node T4 issues an order for the 1st detachment to detour to area A2 at "2023-11-25T14:35:00". The correlation analysis matrix established on this basis records the mapping relationship between the decision content, situational changes, and intelligence basis corresponding to each time node. For example, the decision content of node T1 is to advance for information acquisition, the corresponding situational change is the deployment of information acquisition forces, and there is no intelligence basis; the decision content of node T2 is to receive intelligence, the corresponding situational change is the discovery of the enemy's situation in area A3, and the intelligence basis is the report from unmanned information acquisition equipment; the decision content of node T3 is to cancel the deployment and change to a coverage operation, the corresponding situational change is that the coverage forces enter a state of readiness, and the intelligence basis is the confirmation of the enemy's cluster threat; the decision content of node T4 is to make a detour and reposition, the corresponding situational change is that 1st squad changes its route, and the intelligence basis is to avoid one's own coverage area.

[0034] After establishing the correlation analysis matrix, pattern analysis is performed on the decision node sequence based on the rule base to identify typical behavioral patterns and label each decision node with an intent tag and action style classification. For the decision of node T3, "cancel the original mobilization plan and prioritize coverage action", the system matches the "emergency response when encountering sudden threats" pattern in the rule base, labeling its intent as "mobilization adjustment under coverage cover" and classifying its action style as "combination of coverage and mobilization". For the entire decision sequence T1 to T4, the system identifies the overall pattern as "information acquisition, intelligence acquisition, coverage decision, and mobilization adjustment", matches the typical pattern "coverage-mobilization coordination based on real-time intelligence" in the rule base, labels the overall intent as "integrated information acquisition-strike-mobilization", and classifies its action style as "joint strike and mobilization".

[0035] Subsequently, based on feedback data from the simulation and adjudication system, the decision effectiveness level of each decision node was labeled according to the achievement rate of operational objectives, loss exchange ratio, and situational control indicators. Taking node T3 as an example, the simulation and adjudication feedback showed that after the implementation of this decision, the enemy cluster in area A3 was successfully suppressed without any losses to our side, and the flanking maneuver in area A2 was successfully completed. Based on this, the effectiveness indicators were calculated as follows: the achievement rate of objectives was 100%, the loss exchange ratio was zero losses to our side, and the situational control assessment was improved and the initiative was gained. The overall decision effectiveness level of this node was labeled as "highly efficient decision". At the same time, the enemy and friendly force comparison, environmental conditions, and intelligence completeness corresponding to the decision were extracted as conditional features. Among them, the force comparison was "enemy cluster against our combined arms unit plus support unit", the environmental conditions were "hilly area, clear weather, good visibility", the intelligence completeness was "real-time high-definition acquisition by unmanned information acquisition equipment, high timeliness, and credibility of 0.95", and the time pressure was "3-minute decision window".

[0036] Finally, the information from all the above dimensions is integrated to form a labeled decision trajectory. Each decision node includes an intent label, action style classification, decision effectiveness level, conditional characteristics, and associated decision sequences. For example, the labeling result of node T3 is: the intent label is "mobilization and adjustment under cover", the action style classification is "combination of coverage and mobilization", the decision effectiveness level is "efficient decision", the conditional characteristics cover four dimensions: power comparison, environmental conditions, intelligence completeness, and time pressure, and the associated decision sequences point to three nodes: T1 information acquisition and advance, T2 intelligence acquisition, and T4 detour and relocation.

[0037] S104, Receive the constraints of the new scenario, retrieve candidate historical trajectory fragments from the labeled decision trajectory, and synthesize decision trajectory data through trajectory logic transfer and triple consistency verification;

[0038] First, the system receives the constraints of the newly proposed scenario, including operational objectives, force composition, environmental conditions, time constraints, and spatial constraints. Taking the new scenario "Sand Operation-01" as an example, its constraints are: the operational objective is to seize and control the transportation hub of oasis area O1; the force composition includes a combined mobile unit, a support company, and an information acquisition platoon for the opposing side, while the opposing side consists of a motorized unit; the environmental conditions are sandy desert with low visibility due to sandstorms; the time constraint is that the total mission duration does not exceed 4 hours; and the spatial constraint is the key points O1, S2, and N3 within a radius of 20 kilometers. Based on these constraints, the system retrieves matching candidate historical trajectory segments from the tagged decision trajectories generated in step three.

[0039] After the retrieval is completed, the candidate historical trajectory segments are quantitatively evaluated based on a multidimensional similarity scoring model. This model includes five dimensions: terrain features, power balance, action objectives, time pressure, and intelligence conditions. Each dimension is assigned a weight: terrain features (0.25) are calculated using a semantic similarity matrix based on terrain taxonomy; power balance (0.20) is calculated using log-normalized power ratios; action objectives (0.20) are calculated using a target type matching matrix; time pressure (0.15) is calculated using normalized decision window ratios; and intelligence conditions (0.20) are calculated using the product of intelligence timeliness and credibility. Taking the trajectory of node T3 marked in step three as an example, its similarity with the newly defined dimensions is calculated: In the terrain feature dimension, the original condition is hilly terrain while the target condition is sandy desert, and the semantic similarity score is 0.30; in the force comparison dimension, the score after normalization is 0.65; in the action target dimension, the correlation score between "coverage and blockade" and "seizure and control of the hub" is 0.50; in the time pressure dimension, the score after normalization of the decision window ratio is 0.80; in the intelligence condition dimension, the original condition is real-time high-definition acquisition by unmanned equipment while the target condition is ground information acquisition and close-range acquisition under sandstorm weather, and the ratio of the product of timeliness and credibility after normalization is 0.45. The weighted comprehensive fit calculation result is 0.515.

[0040] The system presets a migration threshold of 0.60. When the overall fit is not lower than this threshold, parameterized adjustments and logical reorganization are performed on candidate historical trajectory segments. When the overall fit is lower than the threshold, a low-fit warning is triggered, and the system performs at least one of the following actions: prompting the operator with the fit value and suggesting manual review; automatically retrieving other candidate historical trajectory segments with an overall fit higher than the threshold; recommending the fusion of multiple complementary candidate historical trajectory segments; and displaying the contribution of each dimension to the low fit to guide the adjustment direction. In this example, the overall fit of 0.515 is lower than the threshold of 0.60, and the system indicates that terrain and intelligence conditions are the main weaknesses and recommends a combination strategy.

[0041] For candidate trajectories whose overall adaptability reaches the threshold, the system performs parameterized adjustments, replacing elements such as the execution unit, target area, and intelligence source in the original decision with corresponding elements in the new scenario. It then logically reorganizes these elements based on the new scenario conditions. For example, due to the impact of sandstorms on accuracy, a meteorological condition assessment and effect correction sub-decision is added before the coverage decision. Because the opposing force is a motorized unit, the coverage scheme is adjusted from a high-efficiency approach targeting clusters to an adaptable approach targeting individual units. After adjustment, boundary verification is performed based on the tactical applicable boundary constraint engine. This engine predefines applicable boundary conditions for each action mode, including applicable terrain type, minimum visibility, force ratio range, minimum intelligence confidence level, and applicable mission phase. It also predefines the adjustment tolerance range for each parameter. The system checks each adjusted parameter to ensure it does not exceed the boundaries. For example, if the terrain type is sandy desert, not in the original action mode's list of allowed hills, mountains, and plains; if the visibility is 200 meters, below the minimum constraint of 500 meters; or if the intelligence confidence level is 0.56, below the minimum constraint of 0.70, all three are considered failures.

[0042] When boundary verification fails, the system executes an intelligent reconstruction strategy, including at least one of the following strategies: A tactical fragment combination strategy decomposes the original action into multiple sub-fragments, retrieves suitable alternatives for each, and recombines them. For example, the "coverage and obstruction" sub-action is adapted to "indirect coverage" under sandstorm conditions, and the "mobilization and adjustment" sub-action is adapted to "navigation-guided detour" under low visibility conditions. A tactical generalization and instantiation strategy generalizes the original action to higher-level principles and then re-instantiates it according to new scenario conditions. For example, "mobilization and adjustment under coverage" is generalized to "mobilization that creates a window through coverage," and instantiated under sandy conditions as a combination of "smoke cover plus support suppression plus maneuver detour." A similar case fusion strategy retrieves multiple candidate historical trajectory fragments with moderate overall adaptability but complementary characteristics and performs intelligent fusion. For example, the coverage trajectory in a sandy environment is fused with the mobilization trajectory under low visibility to generate a new synthetic trajectory.

[0043] After the above migration and reconstruction, the system performs a triple consistency check on the synthetic trajectory and iteratively corrects it until all checks are passed. The triple consistency check includes: tactical logic consistency check, which verifies that the migrated decision sequence conforms to the operational doctrine and that the mission phase transitions are reasonable. For example, "launching a mobile assault under 200-meter visibility conditions in sandstorm weather" is judged as failing the doctrine check and is corrected to "conducting a low-speed directional assault under smoke cover and information acquisition guidance"; resource continuity check, which verifies that the evolution of force status, equipment wear and tear, and supply consumption in the trajectory conforms to physical laws. For example, "the synthetic mobile unit plans to continuously move and interact for 4 hours" is judged as a boundary risk because sandy terrain consumption is 1.8 times the standard and a 20% safety margin needs to be maintained. The decision to "establish a temporary supply point" is inserted at the middle moment; and spatiotemporal constraint compliance check, which verifies that the spatiotemporal relationships of all mobilization paths, information acquisition ranges, and coverage areas can be realized under the new scenario. For example, "requiring support forces to complete the mobilization and deployment that requires 90 minutes within 30 minutes" is judged as failing, and the coverage preparation decision is advanced and the subsequent action time is postponed. For trajectory segments that fail the verification, rule-driven iterative correction is performed until synthetic decision trajectory data that fully meets the triple verification requirements is generated. This data is accompanied by an adaptation evaluation report and boundary verification records, which can be used for decision knowledge push in step five.

[0044] S105, based on the commander's historical decision-making style profile and real-time situation simulation, performs personalized and exploratory decision-making knowledge push on the synthesized decision trajectory data.

[0045] First, based on labeled decision-making trajectories, the system analyzes commanders' decision-making preferences, risk tolerance, and innovativeness in past simulations, constructing a quantitative profile of their historical decision-making styles. The system extracts decision-making tendency characteristics by statistically analyzing the frequency and effectiveness of various action patterns used by specific commanders in multiple simulations. For example, analyzing the historical decision-making trajectory of Commander Zhang, it was found that he frequently used coverage-strike action patterns in past simulations and proactively initiated actions under high-risk situations, while occasionally attempting unconventional solutions. Based on this, the generated decision-making style profile is: primary style "coverage-strike type," risk tolerance "high," and innovativeness "moderate."

[0046] During the simulation, the system receives situational data in real time and matches scenario cases similar to the current situation from the synthetic decision trajectory data generated in step four. For example, when the simulation of "Operation Sandy Land-01" reached the 80th minute after its start, the system received a situational update that "the enemy's reinforced facilities were discovered in area S2". It then retrieved and matched the previously synthesized trajectory case "ST-2024-Operation Sandy Land-01" in the synthetic trajectory library. This case corresponds to the scenario of dealing with reinforced facilities under sandy conditions.

[0047] Based on matched scenario cases, a multi-dimensional knowledge push strategy is implemented, including personalized push, exploratory push, and risk warning push. Personalized push prioritizes decision trajectory segments that match the current commander's historical decision-making style profile and have a high level of decision-making effectiveness. For example, for Brigade Commander Zhang's "coverage strike" style, the system prioritizes pushing synthetic trajectory cases that use a concentrated coverage strike method to deal with reinforced facilities and whose effectiveness is evaluated as highly efficient. Exploratory push provides comparative cases that use different decision-making styles but achieve similar action goals to broaden the commander's decision-making perspective. For example, the system simultaneously pushes the "steady advance and point-by-point elimination" plan and the "electromagnetic interference combined with special forces sabotage" plan. The former is more conservative in style, while the latter is more innovative. Both aim to seize control of transportation hubs but take completely different action paths. Risk warning push notifications are based on typical error patterns in tagged decision-making trajectories to provide risk warnings for current decisions. For example, if the system detects a historical trajectory showing an error pattern of "directly attacking fortified facilities without sufficient information," leading to significant losses, this pattern is highly similar to the current situation where information acquisition in the S2 area is insufficient. The system then pushes a warning to the commander: "Historical data shows that directly attacking fortified facilities without sufficient information has a high failure rate; it is recommended to enhance information acquisition methods first." Through the coordinated operation of these three push strategies, a closed-loop support system is achieved, encompassing data processing and knowledge-based decision-making.

[0048] Please see Figure 2 The second embodiment of the present invention provides a multimodal cross-domain wargaming decision data construction device based on trajectory logic transfer, comprising:

[0049] The standardization processing unit 201 is used to define a structured data model of decision trajectory that includes timestamps, command positions, operation types, multimodal information sources and semantic context. Based on the data model, it performs structured parsing and semantic alignment on text instructions, voice communication records and graphic annotation data in the distributed wargaming system to generate standardized decision trajectory fragments.

[0050] The spatiotemporal fusion unit 202 is used to perform spatiotemporal alignment and logical fusion of the standardized decision trajectory segments from different simulation sessions and command positions to form a panoramic record of the decision-making process.

[0051] Pattern labeling unit 203 is used to perform pattern mining and semantic labeling on the panoramic decision-making process record based on the military rule base, and generate a labeled decision trajectory containing tactical intention labels, decision effectiveness levels and trigger condition features.

[0052] The trajectory synthesis unit 204 is used to receive the constraints of the new inference scenario, retrieve candidate historical trajectory fragments from the labeled decision trajectory, and synthesize decision trajectory data through trajectory logic transfer and triple consistency verification.

[0053] The knowledge push unit 205 is used to push personalized and exploratory decision knowledge to the synthetic decision trajectory data based on the commander's historical decision-making style profile and real-time situation simulation.

[0054] The third embodiment of the present invention provides a multimodal cross-domain wargame decision data construction device based on trajectory logic transfer, including a memory and a processor. The memory stores a computer program, which can be executed by the processor to implement the multimodal cross-domain wargame decision data construction method based on trajectory logic transfer as described in any of the above embodiments.

[0055] The fourth embodiment of the present invention provides a computer-readable storage medium storing a computer program, which can be executed by the processor of the device where the computer-readable storage medium is located, to implement the method for constructing multimodal cross-domain wargaming decision data based on trajectory logic migration as described in any of the above embodiments.

[0056] Exemplary examples show that the computer program described in the third and fourth embodiments of the present invention can be divided into one or more modules, which are stored in the memory and executed by the processor to complete the present invention. The one or more modules can be a series of computer program instruction segments capable of performing specific functions, which describe the execution process of the computer program in the device for constructing a multimodal cross-domain wargaming decision data based on trajectory logic migration. For example, the apparatus described in the second embodiment of the present invention.

[0057] The processor referred to can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor. This processor is the control center of the aforementioned method for constructing multimodal cross-domain wargaming decision data based on trajectory logic transfer, connecting various parts of the method through various interfaces and lines.

[0058] The memory can be used to store the computer program and / or modules. The processor, by running or executing the computer program and / or modules stored in the memory, and by calling the data stored in the memory, implements various functions of a multimodal cross-domain wargaming decision data construction method based on trajectory logic migration. The memory may mainly include a program storage area and a data storage area. The program storage area may store the operating system, at least one application program required for a function (such as sound playback function, text conversion function, etc.), etc.; the data storage area may store data created according to the use of the mobile phone (such as audio data, text message data, etc.). In addition, the memory may include high-speed random access memory, and may also include non-volatile memory, such as hard disk, memory, plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0059] If the implemented module is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the above embodiments of the present invention can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content included in the computer-readable medium can be appropriately added or removed according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electrical carrier signals and telecommunication signals.

[0060] It should be noted that the device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Furthermore, in the accompanying drawings of the device embodiments provided by this invention, the connection relationships between modules indicate that they have communication connections, which can be specifically implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement this without any creative effort.

[0061] The above description is merely a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for constructing multimodal cross-domain wargame decision data based on trajectory logic transfer, characterized in that, include: Define a structured data model for decision trajectories that includes timestamps, command positions, operation types, multimodal information sources, and semantic context. Based on the data model, perform structured parsing and semantic alignment on text instructions, voice communication records, and graphic annotation data in a distributed wargaming system to generate standardized decision trajectory fragments. For the standardized decision trajectory segments from different simulation sessions and command positions, a cross-session spatiotemporal alignment algorithm that integrates dynamic time warping algorithm and military rule constraints is adopted to generate a unified logical timeline. The consistency of trajectories from multiple positions within the same time period is verified by using a military rule base. For detected conflicts, a weighted arbitration mechanism based at least on the command level weight is used for logical fusion to form a panoramic record of the decision process. Based on the military rule base, pattern mining and semantic annotation are performed on the panoramic decision-making process records to generate tagged decision trajectories containing tactical intention labels, decision effectiveness levels, and trigger condition features. The system receives the constraints of the newly proposed scenario, retrieves candidate historical trajectory fragments from the labeled decision trajectory, and synthesizes decision trajectory data through trajectory logic transfer and triple consistency verification. Based on the commander's historical decision-making style profile and real-time situation simulation, personalized and exploratory decision-making knowledge is pushed to the synthesized decision-making trajectory data.

2. The method according to claim 1, characterized in that, The process involves performing structured parsing and semantic alignment of text commands, voice communication records, and graphical annotation data in the distributed wargaming system based on the data model, generating standardized decision trajectory fragments, specifically: Based on the structured data model of the decision trajectory, the historical command records, voice communication archives, electronic sand table annotation files and interface interaction logs from different simulation systems in the distributed wargaming simulation system are identified and extracted. The time information with different formats in each system is uniformly converted into the ISO 8601 standard format. The different names of the same command position in different systems are uniformly mapped to the standard position name. The multi-source data at the same decision moment are associated through the unified timestamp to form a unified data view. The voice communication records in the unified data view are processed to convert speech to text, the graphic annotation data is parsed semantically, the text instructions are analyzed in a structured manner, and the parsing results are associated with the corresponding timestamps, command positions and semantic contexts to generate the standardized decision trajectory fragments.

3. The method according to claim 1, characterized in that, The method of employing a fusion of dynamic time warping algorithm and military rule constraints for cross-situational spatiotemporal alignment to generate a unified logical timeline is as follows: For each simulation, a multidimensional feature sequence is constructed from the trajectory data, including operation density, state change rate, event interval distribution, and resource consumption rate. Calculate the eigenvector distance between any two time points in the two simulation trajectories to construct a cost matrix, and search for the minimum cumulative cost path from the starting point to the ending point in the cost matrix as the optimal matching relationship; The optimal matching relationship is then subjected to operational phase constraint verification, causal relationship constraint verification, and rhythm boundary constraint verification in sequence. When the verification fails, iterative optimization based on constraint relaxation is performed, including adding a constraint relaxation term to the cost matrix to re-search for the suboptimal path, performing dynamic time warping independently in segments according to tactical phases, and calling predefined processing rules in the expert rule base; The alignment results that pass the verification are converted into the unified logical timeline with rhythm labels.

4. The method according to claim 1, characterized in that, The weights of the weighted arbitration mechanism include command level weight, data completeness, and time accuracy.

5. The method according to claim 1, characterized in that, The process involves pattern mining and semantic annotation of the panoramic decision-making process records based on a military rule base, generating labeled decision trajectories that include tactical intent tags, decision effectiveness levels, and triggering condition features. Specifically: Extract the key decision node sequence and the semantic context corresponding to each decision node from the panoramic decision process record, and establish a decision, state, and time sequence correlation analysis matrix based on the extracted decision node sequence and semantic context. Based on the military rule base, pattern analysis is performed on the decision node sequence in the correlation analysis matrix to identify typical tactical behavior patterns, and each decision node is labeled with a tactical intention tag and combat style classification. Based on the feedback data from the simulation and adjudication system, the decision effectiveness level of each decision node is marked according to the degree of achievement of combat objectives, the battle loss exchange ratio, and the situation control capability index. The enemy and friendly forces comparison, battlefield environment, and intelligence completeness corresponding to the decision are extracted as conditional features to form the labeled decision trajectory.

6. The method according to claim 1, characterized in that, The process of receiving the constraints of the newly deduced scenario, retrieving candidate historical trajectory fragments from the labeled decision trajectory, and synthesizing decision trajectory data through trajectory logic transfer and triple consistency verification, specifically involves: Receive constraints of a new simulation scenario, including operational objectives, force composition, battlefield environment, time constraints, and space constraints, and retrieve candidate historical trajectory segments that match the constraints from the labeled decision trajectories; The candidate historical trajectory segments are quantitatively evaluated for their suitability based on a multidimensional similarity scoring model. The multidimensional similarity scoring model includes weighted similarity calculations for five dimensions: terrain features, troop strength comparison, combat objectives, time pressure, and intelligence conditions, to obtain the comprehensive suitability of each candidate historical trajectory segment. When the overall fit is not lower than the preset migration threshold, parameter adjustment and logical reorganization are performed on the candidate historical trajectory segments; When the overall adaptability is lower than the preset migration threshold, a low adaptability warning is triggered. The tactical applicable boundary constraint engine performs boundary verification on the parameterized adjusted trajectory, and executes an intelligent reconstruction strategy when the verification fails. The synthetic trajectory is subjected to the triple consistency check and iterative correction until all checks are passed, thereby generating the synthetic decision trajectory data.

7. The method according to claim 6, characterized in that, The triple consistency check includes: Tactical logic consistency verification: Verify that the decision sequence after migration conforms to military operational doctrine and that the transition of operational phases is reasonable; Resource continuity verification: Verify that the evolution of troop status, equipment losses, and supply consumption in the trajectory conforms to physical laws and military operational constraints; Spatiotemporal constraint compliance verification: Verify that the spatiotemporal relationships of all maneuver paths, reconnaissance ranges, and fire coverage can be achieved under the constraints of the new simulation scenario.

8. The method according to claim 6, characterized in that, The intelligent reconfiguration strategy includes at least one of the following strategies: Tactical fragment combination strategy: Decompose the original tactic into multiple sub-tactical fragments, search for suitable alternatives for each, and recombine them. Tactical generalization and instantiation strategy: generalize the original tactics into higher-level tactical principles, and then instantiate them into specific tactical schemes based on the newly deduced constraints. Similar case fusion strategy: intelligently fuse multiple candidate historical trajectory segments with moderate overall fit but complementary characteristics.

9. The method according to claim 1, characterized in that, The process of applying personalized and exploratory decision-making knowledge to the synthesized decision trajectory data based on the commander's historical decision-making style profile and real-time situation simulation involves: Based on the labeled decision-making trajectory, the commander's decision-making pattern preferences, risk tolerance tendencies, and innovative characteristics in past simulations are analyzed to construct a quantitative profile of the commander's historical decision-making style. During the simulation, situational data is received in real time, and scenario cases similar to the current situation are matched from the synthetic decision trajectory data. A multi-dimensional knowledge push strategy is implemented based on matched scenario cases. The multi-dimensional knowledge push strategy includes personalized push, exploratory push, and risk warning push. Among them, the personalized push prioritizes pushing decision trajectory segments that match the current commander's historical decision-making style profile and have a high level of decision-making effectiveness; the exploratory push pushes comparative cases that adopt different decision-making styles but achieve similar combat objectives; and the risk warning push provides risk warnings for the current decision based on typical error patterns in the labeled decision trajectory.

10. A device for constructing multimodal cross-domain wargaming decision data based on trajectory logic transfer, characterized in that, include: The standardization processing unit is used to define a structured data model of decision trajectory that includes timestamps, command positions, operation types, multimodal information sources, and semantic context. Based on the data model, it performs structured parsing and semantic alignment on text instructions, voice communication records, and graphic annotation data in the distributed wargaming system to generate standardized decision trajectory fragments. The spatiotemporal fusion unit is used to align the standardized decision trajectory segments from different simulation sessions and command positions across sessions using a fusion dynamic time warping algorithm and military rule constraints to generate a unified logical timeline. It also uses a military rule base to verify the consistency of trajectories from multiple positions within the same time period. For detected conflicts, it uses a weighted arbitration mechanism based at least on the command level weight to perform logical fusion, forming a panoramic record of the decision-making process. The pattern labeling unit is used to perform pattern mining and semantic labeling on the panoramic decision-making process record based on the military rule base, and generate a labeled decision trajectory containing tactical intention labels, decision effectiveness levels and trigger condition features. The trajectory synthesis unit is used to receive the constraints of the new inference scenario, retrieve candidate historical trajectory fragments from the labeled decision trajectory, and synthesize decision trajectory data through trajectory logic transfer and triple consistency verification. The knowledge push unit is used to push personalized and exploratory decision-making knowledge to the synthetic decision trajectory data based on the commander's historical decision-making style profile and real-time situation simulation.

Images