An anti-armor weapon simulation training method and system based on a large model

Abstract

The application discloses a large model-based anti-armor weapon simulation training method and system, relates to the technical field of simulation training, and comprises the following steps: constructing an anti-armor weapon special large model and a battlefield environment element library; receiving semantic requirements of training personnel and setting training task parameters; analyzing the semantic requirements through the special large model to determine constraint conditions for battlefield environment generation; calling corresponding environment resources from the battlefield environment element library according to the constraint conditions to generate an initial battlefield environment; adapting and optimizing the initial battlefield environment and outputting the initial battlefield environment to a simulation training system for immersive simulation training and real-time collection of training data; evaluating a training process based on the training data and iteratively optimizing the special large model and the battlefield environment element library according to an evaluation result. The application realizes intelligent customization and closed-loop iteration of an anti-armor weapon training scene.

Description

Technical Field

[0001] This invention relates to the field of simulation training technology, and specifically to a method and system for simulation training of anti-armor weapons based on a large model. Background Technology

[0002] In modern intelligent battlefield systems, anti-armor weapons are core equipment for countering enemy armored forces and breaking through mechanized combat formations. The operator's control precision, battlefield adaptability, and tactical handling skills directly affect the equipment's combat effectiveness. Constrained by objective conditions such as high costs of live-fire training, large ammunition consumption, limited field training grounds, and the difficulty in realistically replicating high-risk and complex battlefield environments, routine, high-frequency, and multi-condition live-fire training is difficult to conduct. Therefore, 3D simulation and virtual immersive simulation training technologies have become core methods for practical training, subject assessment, and personnel capability development of anti-armor weapons. With the increasingly complex modern battlefield environment, and the common occurrence of various weather conditions, complex terrain, electromagnetic interference, and mobile target confrontation scenarios, the industry has placed higher demands on the scenario richness, simulation accuracy, training flexibility, and scientific training capabilities of anti-armor simulation training systems.

[0003] Currently, anti-armor weapon simulation training systems used in the industry generally adopt a static resource pre-fabrication and fixed script-driven technical model. Related equipment uses pre-set manual modeling and parameter pre-setting to solidify basic battlefield scenarios, target models, and environmental parameters, forming a fixed simulation resource library. Before training begins, technicians need to manually configure training subjects, battlefield environment, target status, and other relevant parameters, and pre-determine the entire training process. During training, the system can only run according to the preset script and fixed parameters, displaying fixed battlefield situations and combat conditions. Trainees can only choose standardized training subjects built into the system for training; training scenarios, environmental conditions, and combat modes cannot be flexibly adjusted, resulting in a rigid and monotonous overall training mode.

[0004] However, constrained by a fixed script and static pre-built resource architecture, existing systems cannot flexibly generate differentiated and complex real-world battlefield scenarios based on actual training needs. They can only rely on pre-set standardized subjects to conduct homogeneous training, greatly reducing the relevance of simulated training to real complex battlefields. Furthermore, this rigid architecture prevents the system from optimizing scenario resources and training rules based on training process data. Training modes, simulation parameters, and adversarial situations remain unchanged for extended periods, lacking a data-driven scientific evaluation and optimization loop. Ultimately, this results in a severe deficiency in training relevance, realism, and scientific rigor, failing to meet the diverse, advanced, and combat-oriented training needs of current anti-armor weapons. Summary of the Invention

[0005] In view of the shortcomings of the prior art, the purpose of this invention is to provide a method and system for anti-armor weapon simulation training based on a large model, which aims to solve the above-mentioned problems described in the prior art.

[0006] A first aspect of the present invention is to provide a method for simulating and training anti-armor weapons based on a large model, the method comprising: Construct a large-scale model for anti-armor weapons and a database of battlefield environment elements; Receive the semantic requirements from the trainers and set the training task parameters; The semantic requirements are analyzed using the dedicated large model to determine the constraints for generating the battlefield environment; Based on the constraints, the corresponding environmental resources are retrieved from the battlefield environment element library to generate the initial battlefield environment; The initial battlefield environment is adapted and optimized, and then output to the simulation training system for immersive simulation training, while collecting training data in real time. The training process is evaluated based on the training data, and the dedicated large model and battlefield environment element library are iteratively optimized based on the evaluation results.

[0007] According to one aspect of the above technical solution, the step of constructing a large-scale model specifically for anti-armor weapons includes: Acquire common weapon parameters, operation manuals, guidance principles, tactical specifications, failure case databases, and historical training data of anti-armor weapons as multi-source training sample data; Based on a general large model architecture, entity extraction and relation extraction algorithms are used to extract semantic information from the multi-source training sample data, construct knowledge triples, and form a semantic association network. The general large model is fine-tuned and trained by combining reinforcement learning methods with an expert scoring mechanism to obtain the special large model. The dedicated large model includes an environment generation module, a tactical decision-making module, an interactive understanding module, and a debriefing analysis module.

[0008] According to one aspect of the above technical solution, the battlefield environment element library includes a three-dimensional model library, a parameter configuration library, and a dynamic special effects library; The 3D model library is used to store exclusive models of anti-armor weapons and their supporting equipment, as well as battlefield element models. The parameter configuration library is used to store the parameter ranges of various environmental elements and the default performance parameters of anti-armor weapons; The dynamic effects library is used to store battlefield dynamic effects and, combined with the six-degree-of-freedom dynamic model of anti-armor weapons, to recreate the dynamic scenes of weapon flight and impact.

[0009] According to one aspect of the above technical solution, the step of parsing the semantic requirements using the dedicated large model to determine the constraints for generating the battlefield environment includes: Through the interactive understanding module of the dedicated large model, the semantic requirements are segmented, part-of-speech tagging is performed, and entity recognition is performed to extract core semantic features; The knowledge graph reasoning engine is used to analyze the relationships between entities in the core semantic features; By combining the general performance parameters and tactical rules of anti-armor weapons, the constraints for generating the battlefield environment are determined.

[0010] According to one aspect of the above technical solution, the step of parsing the relationships between entities in the core semantic features using a knowledge graph reasoning engine specifically includes: The extracted core semantic features are mapped to a pre-defined military domain knowledge graph to match the corresponding entity nodes; Based on the edge relationships in the knowledge graph, the causal relationships and tactical collaboration logic between the entity nodes are deduced, and a structured parameter set for constraining scenario generation is output.

[0011] According to one aspect of the above technical solution, the step of adapting and optimizing the initial battlefield environment includes: Adjust battlefield environment parameters according to the current weapon type to match the core performance indicators and guidance methods of that weapon type, so as to optimize weapon adaptation. Adjust the complexity of the scenario and the speed of the target's movement according to the training subject and difficulty to optimize training adaptation; The rendering frame rate is increased to no less than a preset threshold by using LOD algorithm and texture compression algorithm in order to optimize rendering adaptation.

[0012] According to one aspect of the above technical solution, the step of weapon adaptation optimization specifically includes: Parse the weapon type identifier bound to the current training task and retrieve the corresponding dynamic parameter package from the parameter configuration library; The dynamic parameters are imported into the physics engine of the initial battlefield environment, and the electromagnetic interference parameters and weather attenuation coefficients in the environment are dynamically adjusted according to the weapon's guidance method.

[0013] A second aspect of the present invention is to provide an anti-armor weapon simulation training system based on a large model, applicable to the method described in the above-mentioned technical solution, the system comprising: The model building module is used to build large-scale models of anti-armor weapons and a library of battlefield environment elements. The parameter setting module is used to receive the semantic requirements of the trainers and set the training task parameters; The semantic parsing module is used to parse the semantic requirements through the dedicated large model and determine the constraints for generating the battlefield environment. The environment generation module is used to call the corresponding environmental resources from the battlefield environment element library according to the constraints to generate the initial battlefield environment. The training execution module is used to adapt and optimize the initial battlefield environment, output it to the simulation training system for immersive simulation training, and collect training data in real time. The iterative optimization module is used to evaluate the training process based on the training data and to iteratively optimize the dedicated large model and the battlefield environment element library according to the evaluation results.

[0014] A third aspect of the present invention is to provide a readable storage medium having computer instructions stored thereon, which, when executed by a processor, implement the steps of the method described in the above-described technical solution.

[0015] A fourth aspect of the present invention is to provide an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of the method described in the above technical solutions.

[0016] Compared with existing technologies, the advantages of using the anti-armor weapon simulation training method and system based on a large model as shown in this invention are as follows: This invention can receive personalized semantic training requests from trainees and generate standardized battlefield constraints through intelligent parsing using a dedicated large-scale model. It rapidly generates and adaptively optimizes exclusive battlefield scenarios based on a resource library, overcoming the limitations of traditional technologies that rely on manually preset scripts and fixed resources. This allows for flexible adaptation to diverse and complex real-world training conditions, significantly improving the dynamism and battlefield realism of simulated training scenarios. Simultaneously, this invention can collect training data in real time and perform intelligent quantitative evaluation. Based on the evaluation results, iteratively optimizes the large-scale model's parsing logic and battlefield resource library, constructing a complete training optimization loop. This continuously improves the relevance and scientific rigor of training, effectively compensating for the shortcomings of traditional training, such as its inability to dynamically iterate and its limited effectiveness. It precisely meets the current high-quality, personalized training needs for anti-armor weapons in real-world combat scenarios. Attached Figure Description

[0017] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the description of the embodiments taken in conjunction with the following drawings, in which: Figure 1 A flowchart illustrating the anti-armor weapon simulation training method based on a large model provided in an embodiment of the present invention; Figure 2The diagram shows the structure of an anti-armor weapon simulation training system based on a large model, as provided in an embodiment of the present invention. Detailed Implementation

[0018] To make the objectives, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Several embodiments of the present invention are shown in the drawings. However, the present invention can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided so that the disclosure of the present invention will be more thorough and complete.

[0019] It should be noted that when a component is said to be "fixed to" another component, it can be directly on the other component or there may be an intervening component. When a component is said to be "connected to" another component, it can be directly connected to the other component or there may be an intervening component. The terms "vertical," "horizontal," "left," "right," and similar expressions used in this document are for illustrative purposes only.

[0020] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0021] Example 1 Please see Figure 1 The first embodiment of the present invention provides a method for simulating and training anti-armor weapons based on a large model, the method comprising steps S10-S60: Step S10: Construct a large-scale model for anti-armor weapons and a database of battlefield elements.

[0022] Specifically, when constructing a large-scale model dedicated to anti-armor weapons, an open-source or commercially available general-purpose pre-trained model is used as the foundation. The model's original feature extraction, semantic encoding, and text reasoning capabilities are retained. Based on this, specialized modifications and fine-tuning training for the military vertical domain are carried out. In this embodiment, multi-source heterogeneous data corresponding to anti-armor weapons is acquired, standardized, and then used to form a dedicated training dataset. A military knowledge graph is constructed based on this dataset. Then, through reinforcement learning combined with expert-verified training, the general-purpose model is iteratively optimized into a large-scale vertical model fully adapted to anti-armor weapon training scenarios. Ultimately, this model is endowed with four core intelligent capabilities: military-specific semantic understanding, intelligent battlefield scenario generation, tactical rule compliance verification, and training data review and analysis.

[0023] The project also involves building a battlefield element database. Based on 3D modeling software, dynamic simulation tools, and parameter calibration, three standardized resource databases are constructed, categorized and hierarchically. These databases consist of a high-precision 3D model library, a quantitative parameter configuration library, and a dynamic battlefield effects library. During the database construction process, the file format, resolution standard, model patch specifications, parameter accuracy range, and engine call interface protocol are standardized for all resources. This ensures that all resources can be quickly identified, accurately retrieved, batch-assigned, and dynamically modified by the simulation engine. After the resources are added to the database, interface integration, data mapping, permission binding, and read / write authorization configuration between the large model and the three resource libraries will be performed. This will enable the large model to read resource library content in real time, accurately retrieve and match resources, dynamically modify scene parameters, and iteratively update resource materials.

[0024] Step S20: Receive the semantic requirements from the trainers and set the training task parameters.

[0025] Specifically, ordinary trainees can directly use colloquial and everyday natural language to describe their personalized training needs. For example, they can input custom requirements such as "nighttime mountain low-light mobile armored target tracking and interception training" or "high-precision strike assessment training under strong electromagnetic interference conditions in sandstorm weather." The front end can capture complete original text data in real time and automatically retain traceability information such as operation logs, entry time, and user identification to ensure that the requirements are traceable and reviewable.

[0026] Coaches or assessment managers can use a dedicated configuration portal in the backend to meticulously input a complete set of structured training controllable parameters, including training subject type, duration of a single training session, number of training cycles, overall difficulty level, designated training weapon model, armored target category, initial target coordinates, target speed range, movement trajectory pattern, terrain type, meteorological parameters, visibility range, electromagnetic interference intensity and frequency band, training scoring weighting, and other comprehensive and detailed parameters.

[0027] After acquiring two types of input data, the backend server automatically performs data fusion and encapsulation operations, integrating the scattered semantic text and structured parameters into a single standardized training task data package. Simultaneously, it completes data format verification, initial null value completion, parameter logic pre-check, format unification and standardization, eliminates obvious input errors, and finally pushes the encapsulated complete task data package to the dedicated large model parsing port, thereby completing the initial definition of the training task.

[0028] Step S30: The semantic requirements are parsed using the dedicated large model to determine the constraints for battlefield generation.

[0029] Specifically, the first step is semantic preprocessing and core feature extraction. The large model first performs refined natural language processing on the complete training requirement text input, successively completing text segmentation, part-of-speech tagging, stop word filtering, redundant sentence removal, and accurate identification of military-specific entities. It filters out interfering information such as colloquial modifiers, invalid interjections, and repetitive expressions, accurately identifies and extracts corresponding core military semantic features such as weapon models, training areas, meteorological and optical conditions, electromagnetic interference status, target types and motion characteristics, training difficulty levels, tactical training subjects, and assessment rules, and generates corresponding structured semantic feature vectors, completing the transformation from natural language text to machine-recognizable feature data.

[0030] Then, knowledge graph reasoning and tactical compliance verification are performed. A pre-built knowledge graph specific to anti-armor weapons and a built-in tactical rule base are invoked to conduct multi-dimensional correlation inference and compliance judgment on the extracted core semantic features. The model automatically compares weapon performance limit parameters, battlefield physics laws, and standard tactical usage rules, intelligently identifying and eliminating contradictory parameters and invalid training requirements that exceed weapon range, exceed adaptation limits, violate physical laws, or do not conform to tactical logic. Simultaneously, slightly unreasonable parameters are automatically corrected and optimized to ensure that all requirements are feasible for simulation implementation.

[0031] Finally, the constraints are structured and standardized. All effective elements after preprocessing, reasoning, and verification are classified and organized according to the simulation engine's recognition specifications. This results in a standardized parameter set that covers terrain constraints, meteorological constraints, optical visibility constraints, electromagnetic interference constraints, target motion constraints, weapon performance constraints, tactical action constraints, and training difficulty constraints. All parameters have uniform precision, format, and value range, and can be directly used as the sole rule basis and parameter standard for subsequent initial battlefield generation.

[0032] Step S40: Based on the constraints, retrieve the corresponding resources from the battlefield element library to generate the initial battlefield.

[0033] Specifically, using the structured constraints output by the large model as the sole search index, a multi-dimensional weighted matching algorithm is built. Prioritizing terrain adaptation, followed by weapon and equipment adaptation, then target characteristic adaptation, interference condition adaptation, and finally dynamic effects adaptation, the algorithm sequentially searches three resource libraries: the 3D model library, the parameter configuration library, and the dynamic effects library. It automatically compares resource tags, parameter ranges, and adaptation scene identifiers to accurately match terrain models, fortification obstacle models, weapon and equipment models, armored target models, meteorological and physical parameters, electromagnetic interference parameters, ballistic dynamics parameters, and various dynamic simulation effects resources that perfectly fit the current training constraints.

[0034] After all the required resources are matched and retrieved, the scene initialization and stitching program is started. It automatically completes the world coordinate positioning of all 3D models, spatial hierarchy adaptation, scene boundary calibration, and model scale normalization. At the same time, it batch completes the initialization and assignment of physical parameters, weapon dynamics calculation parameters, and dynamic effect triggering parameters. It associates, binds, and integrates all scattered and independent models, parameters, and effect resources, and finally generates a complete initial battlefield simulation scene that can be rendered normally and loaded for calculation.

[0035] Step S50: The initial battlefield is adapted and optimized, and output to the simulation training for immersive simulation training, and training data is collected in real time.

[0036] Specifically, the first step is to optimize the scene adaptation in multiple dimensions. In response to problems such as weapon adaptation deviation, difficulty matching imbalance, and insufficient rendering performance in the initial battlefield, the weapon performance adaptation, training subject difficulty adaptation, and device rendering performance adaptation are completed in sequence. Various defects such as initial scene parameter mismatch, unreasonable scene complexity, excessive model polygon count, and screen stuttering and delay are corrected, and the basic scene is calibrated into a standardized combat scene that fits the training task.

[0037] Then, the immersive simulation training is implemented. The optimized and calibrated high-quality standardized battlefield scene is pushed to hardware devices such as VR immersive simulation terminal, desktop 3D visualization training, and practical simulation training device through a dedicated interface. After the scene is loaded into the simulation physics engine and the parameters are initialized, the preset training task is automatically started, supporting the trainees to complete the entire process of practical simulation training, including weapon setup, target search, dynamic aiming, continuous tracking, timely firing, ballistic observation, emergency response, and fault diagnosis.

[0038] In addition, real-time data collection is conducted throughout the entire training process. Simultaneously with the start of the training task, background data collection is initiated at millisecond-level uninterrupted acquisition. This process comprehensively captures multi-dimensional, refined training data, including the timing of trainees' operations, the standardization of operational steps, aiming angle deviation data, target tracking stability data, weapon launch timing parameters, ballistic flight data, target maneuver response data, the impact of interference on operations, impact point error, key operational error nodes, and the completeness of fault handling procedures. All data is automatically classified, archived, encrypted, stored, and traceable to ensure the authenticity, completeness, and reproducibility of the training data, providing solid data support for subsequent intelligent evaluation and iterative optimization.

[0039] Step S60: Evaluate the training process based on the training data, and iteratively optimize the dedicated large model and battlefield element library according to the evaluation results.

[0040] Specifically, the training quality is first assessed using intelligent quantitative methods. This involves retrieving comprehensive and detailed practical data from the training archives and, based on pre-set military training quantitative assessment algorithms, official anti-armor weapon operating procedures, and combat tactical assessment standards, performing quantitative scoring calculations across multiple dimensions, including operational standardization, aiming accuracy, timing of launch, target tracking stability, anti-interference handling capabilities, scientific tactical decision-making, and proficiency in handling malfunctions. This automatically generates comprehensive training scores, individual subject scores, and capability weakness distribution maps for trainees. Simultaneously, it accurately identifies and addresses personnel-related issues such as operational errors, scenario adaptation deviations, parameter configuration errors, dynamic effect simulation distortions, and inaccurate interference fitting. Finally, it outputs a standardized, data-driven, and traceable training review and assessment report.

[0041] Then, iterative optimization is carried out. Based on the various problems identified in the review report, two major iterative upgrade operations are performed simultaneously: On the one hand, for problems such as semantic parsing deviation, scene generation logic error, tactical judgment standard deviation, and insufficient accuracy of evaluation algorithm, the semantic parsing weight, knowledge reasoning logic, scene generation rules, tactical judgment threshold, and review evaluation model parameters of the dedicated large model are automatically fine-tuned to complete the iterative upgrade of the large model algorithm and continuously improve the model's intelligent accuracy and military professionalism; On the other hand, for resource library defects such as missing scene models, physical parameter deviation, dynamic effect distortion, and incomplete adaptation, the system automatically adds new combat scene models, corrects deviation dynamics and parameters, optimizes distorted dynamic effects, and improves special working condition adaptation resources to continuously enrich and improve the resource coverage and simulation accuracy of the battlefield element library.

[0042] Through continuous accumulation and iterative optimization of training data, the entire system can dynamically adapt to the needs of actual combat training, achieving an intelligent upgrade effect that becomes more accurate with use and more closely aligned with actual combat with training.

[0043] Compared with existing technologies, the anti-armor weapon simulation training method based on a large model as shown in this invention has the following advantages: This embodiment can receive personalized semantic training requests from trainees and generate standardized battlefield constraints through intelligent parsing using a dedicated large-scale model. It rapidly generates and adaptively optimizes exclusive battlefield scenarios based on a resource library, overcoming the limitations of traditional technologies that rely on manually preset scripts and fixed resources. This allows for flexible adaptation to diverse and complex real-world training conditions, significantly improving the dynamism and battlefield realism of simulated training scenarios. Simultaneously, this embodiment can collect training data in real time and perform intelligent quantitative evaluation. Based on the evaluation results, iterative optimization of the large-scale model's parsing logic and battlefield resource library is achieved, constructing a complete training optimization loop. This continuously improves the relevance and scientific rigor of training, effectively compensating for the shortcomings of traditional training, such as its inability to dynamically iterate and its limited effectiveness. It precisely meets the current high-quality, personalized training needs for anti-armor weapons in a realistic and personalized manner.

[0044] Example 2 The second embodiment of the present invention also provides an anti-armor weapon simulation training method. The method shown in this embodiment is basically similar to the method shown in the first embodiment, except that: In this embodiment, the step of constructing a large-scale model specifically for anti-armor weapons includes: Acquire common weapon parameters, operation manuals, guidance principles, tactical specifications, failure case databases, and historical training data of anti-armor weapons as multi-source training sample data; Based on a general large model architecture, entity extraction and relation extraction algorithms are used to extract semantic information from the multi-source training sample data, construct knowledge triples, and form a semantic association network. The general large model is fine-tuned and trained by combining reinforcement learning methods with an expert scoring mechanism to obtain the special large model. The dedicated large model includes an environment generation module, a tactical decision-making module, an interactive understanding module, and a debriefing analysis module.

[0045] Specifically, the first step is to conduct targeted collection and standardization of multi-source training sample data, including general performance parameters of various anti-armor weapons, military standardized operation manuals, core working principles of weapon guidance, standard tactical application specifications, a database of typical equipment failure cases, and data from the review of combat training conducted by officers and soldiers over the years.

[0046] For structured weapon parameter data, a standardized structured parameter dataset was created by systematically analyzing fixed performance indicators such as range, rate of fire, flight speed, guidance system type, anti-jamming threshold, aerodynamic coefficient, accuracy error range, and extreme high and low temperature conditions for various types of anti-tank missiles, man-portable rocket launchers, and vehicle-mounted anti-armor artillery. For unstructured text data, official military operating procedures, operational prohibitions, tactical application scenarios, formation coordination rules, and equipment usage specifications under different terrain and weather conditions were comprehensively collected and, after cleaning, deduplication, and standardization, formed a standardized tactical text dataset. For fault and training scenario data, typical fault cases such as weapon hardware failure, guidance signal loss, external interference failure, and abnormal trajectory deviation were collected. Simultaneously, daily training error records of officers and soldiers, practical data from live-fire exercises, and annual assessment review materials were compiled and categorized to form a scenario case dataset.

[0047] After all data has been collected, standardized governance operations such as data anonymization, confidentiality removal, error correction, redundancy removal, and format standardization are uniformly performed to generate a high-quality sample set that is uniform in format, accurate in content, covers all scenarios, and can be directly used for model fine-tuning and training.

[0048] Then, a semantic association network specific to anti-armor weapons is built through a dual-algorithm linkage mechanism of entity extraction and relation extraction. It should be noted that since the original training data consists mostly of scattered text, independent parameters, and fragmented cases without fixed logical connections, it cannot be directly used by the model for learning. It must be transformed into a systematic knowledge system through knowledge structuring. Therefore, in this embodiment, by retaining the original basic feature extraction, text encoding, and semantic processing architecture of the general large model, entity extraction and relation extraction algorithms from the military domain are superimposed on this basis to achieve precise structuring of military knowledge.

[0049] More specifically, the entire trained sample data is first scanned, analyzed line by line, and fully identified to accurately extract nine categories of standardized military entities: weapon models, equipment accessories, terrain environment, meteorological conditions, electromagnetic environment, target attributes, tactical actions, fault types, and training subjects. These entities are then tagged and categorized. Next, a relation extraction algorithm is used to deeply mine the inherent logical relationships between these entities, accurately identifying seven categories of military-specific logical relationships: adaptation and matching, conditional constraints, environmental influences, state triggers, equipment failures, external interference, and tactical coordination. Finally, following the standard knowledge triplet structure of "entity-relationship-entity," all scattered knowledge units are uniformly organized, categorized, and stored in the database. All triplet data are integrated to construct a lightweight anti-armor weapon military knowledge graph, forming a complete closed-loop military semantic association network. This allows the model system to grasp the inherent logical relationships between weapons, environment, tactics, faults, and training, fully equipping it with professional military reasoning capabilities and completing the technical transformation from raw, scattered data to a structured military knowledge system.

[0050] Finally, through a combination of dynamic iterative training and manual professional correction, a high-precision large-scale model specifically for anti-armor weapon simulation training is obtained. It should be noted that traditional single-supervised learning fine-tuning methods suffer from fixed model outputs and poor adaptability, making them unsuitable for dynamically changing military training scenarios. Furthermore, pure algorithm training is prone to tactical logic biases and professional judgment errors. This solution employs a dual training mechanism to compensate for these shortcomings. Specifically, firstly, a dedicated reinforcement learning simulation environment adapted to military training scenarios is built. Four core quantitative reward functions are specifically set: scenario generation accuracy, semantic parsing accuracy, tactical rule compliance rate, and training evaluation matching rate. A dynamic iterative reward and penalty update mechanism for the model is constructed. Each time the model completes user requirement parsing, battlefield scenario generation, tactical compliance verification, and training review evaluation, the system immediately provides corresponding reward or penalty scores based on the accuracy, compliance, and fit of the output results. The internal weight parameters of the model are continuously updated through backpropagation algorithms, constantly optimizing the model's output accuracy.

[0051] In this embodiment, the battlefield environment element library includes a 3D model library, a parameter configuration library, and a dynamic effects library; The 3D model library is used to store exclusive models of anti-armor weapons and their supporting equipment, as well as battlefield element models. The parameter configuration library is used to store the parameter ranges of various environmental elements and the default performance parameters of anti-armor weapons; The dynamic effects library is used to store battlefield dynamic effects and, combined with the six-degree-of-freedom dynamic model of anti-armor weapons, to recreate the dynamic scenes of weapon flight and impact.

[0052] Specifically, high-precision 3D modeling technology is used to create 3D models of various anti-armor weapons, supporting equipment, main battle armored targets, all types of battlefield terrain, field fortifications, obstacles, and vegetation and landforms according to the actual equipment scale, appearance details, and size parameters. The number of model facets, texture resolution, rendering and shading standards, and model lightweighting parameters are standardized. All models are baked and optimized and then standardized and stored in the library to provide a realistic, detailed, and renderable visual scene carrier for immersive simulation training.

[0053] It also comprehensively collects and organizes all weapon and equipment performance parameters and battlefield environment physical parameters, accurately defines the normal adjustable range, default standard threshold, and extreme condition boundary of each parameter, establishes a one-to-one correspondence index mechanism between weapon models and parameter packages, and automatic matching rules for scene parameters, and provides accurate, authoritative, and traceable data support for simulation physics engine calculation, ballistic trajectory calculation, environmental interference fitting, and weapon dynamics simulation.

[0054] It also features a full range of battlefield visualization dynamic effects, including meticulously crafted weapon launch tail flames, high-speed airflow, impact explosion impact, shrapnel flying, battlefield smoke dispersion, rain, snow, sand and dust dynamics, and electromagnetic interference flashes. The effects playback frame rate, trigger conditions, duration, and overlay logic are unified. At the same time, all dynamic effects are deeply bound to the weapon's multi-degree-of-freedom dynamic model. Through dynamic algorithms, the pitch, yaw, roll, and translation attitude changes of the weapon during flight are calculated in real time. The effects are dynamically fitted to the actual ballistic trajectory, impact angle, explosion range, and other combat effects, comprehensively improving the battlefield atmosphere and simulation accuracy of the simulation training.

[0055] In this embodiment, the step of parsing the semantic requirements using the dedicated large model to determine the constraints for generating the battlefield environment includes: Through the interactive understanding module of the dedicated large model, the semantic requirements are segmented, part-of-speech tagging is performed, and entity recognition is performed to extract core semantic features; The knowledge graph reasoning engine is used to analyze the relationships between entities in the core semantic features; By combining the general performance parameters and tactical rules of anti-armor weapons, the constraints for generating the battlefield environment are determined.

[0056] Specifically, the system first calls the interactive understanding module built into the dedicated large model to perform standardized, full-process natural language processing operations on the user's input raw semantic text. It sequentially completes accurate word segmentation, part-of-speech tagging, stop word filtering, redundant semantic removal, and military-specific entity recognition. It accurately extracts core military semantic features such as weapon models, training terrain, weather conditions, interference intensity, target features, training difficulty, and tactical subjects, and finally generates standardized feature vectors that can be recognized by machines, completing the digital transformation of spoken language requirements.

[0057] Then, the knowledge graph reasoning engine is automatically started, and the extracted core feature vectors are imported into the semantic association network dedicated to anti-armor weapons. Based on the entity association relationship within the graph, the mutual constraints, mutual influences, tactical adaptation, and scenario matching logic among the core elements are automatically analyzed to form a complete, rigorous, and combat-oriented scenario generation logic chain.

[0058] Finally, by linking the weapon performance parameter library and the military tactical rule library, the preliminary scenario parameters generated by reasoning are fully validated for compliance. Invalid parameters that exceed the weapon performance limits, violate actual combat tactical rules, or cannot be simulated are automatically removed. Inconsistent parameter content is corrected, and necessary missing parameters are added to the scenario. The final output is a standardized set of battlefield environment constraint parameters with a unified format, consistent accuracy standards, rigorous and compliant logic, and can be directly called by the 3D simulation engine. This fully realizes the fully automatic, intelligent, and precise transformation of user-personalized needs into simulation scenario rules.

[0059] In this embodiment, the step of parsing the relationships between entities in the core semantic features using a knowledge graph reasoning engine specifically includes: The extracted core semantic features are mapped to a pre-defined military domain knowledge graph to match the corresponding entity nodes; Based on the edge relationships in the knowledge graph, the causal relationships and tactical collaboration logic between the entity nodes are deduced, and a structured parameter set for constraining scenario generation is output.

[0060] Specifically, the system first uses a built-in dictionary that maps spoken language requirements to professional knowledge to accurately match the fragmented and colloquial core semantic features extracted in the early stage with the standardized and professional military entity nodes within the knowledge graph. This completes the accurate conversion of colloquial descriptions into standard military terms and standardized scenario entities, completely avoiding the problems of semantic comprehension bias and misinterpretation of requirements.

[0061] Then, tactical causal logic quantitative deduction is performed, traversing all the related edge relationships between entity nodes in the knowledge graph. Based on the preset military tactical logic and physical operation rules, the actual combat linkage law between the battlefield environment, weapons and equipment, and armored targets is automatically deduced. Key quantitative parameters such as optical attenuation coefficient in rainy and foggy weather, electromagnetic interference suppression coefficient in different frequency bands, terrain obstruction limiting coefficient of target mobility, and shooting distance correction coefficient of ballistic error are accurately quantified and calculated. The user's fuzzy and qualitative natural language training needs are completely transformed into a precise, quantitative, and directly driving scenario generation structured parameter set.

[0062] In this embodiment, the step of adapting and optimizing the initial battlefield environment includes: Adjust battlefield environment parameters according to the current weapon type to match the core performance indicators and guidance methods of that weapon type, so as to optimize weapon adaptation. Adjust the complexity of the scenario and the speed of the target's movement according to the training subject and difficulty to optimize training adaptation; The rendering frame rate is increased to no less than a preset threshold by using LOD algorithm and texture compression algorithm in order to optimize rendering adaptation.

[0063] Specifically, the system uses a built-in rule table that links weapon models with battlefield environment parameters. During the scenario optimization phase, the system automatically reads the weapon model and guidance system currently bound to the training and adjusts the physical calculation parameters, external interference parameters, and meteorological and optical attenuation parameters of the battlefield environment in batches according to the adaptation rules, so that all environmental characteristics are fully matched with the actual combat working mechanism and performance indicators of the weapon.

[0064] Then, based on the training subject type and difficulty level set by the user, the system automatically retrieves the corresponding difficulty-adaptive template and dynamically adjusts the density of obstacle arrangement, the complexity of terrain obstruction, the target maneuver speed, and the frequency of trajectory maneuver changes. For basic training, the system simplifies the scenario complexity and reduces the difficulty of target maneuvering. For advanced assessment scenarios, the system adds multiple interferences, improves the target maneuver flexibility, and adds unexpected tactical situations, thus realizing a scientific training mode of graded training and step-by-step improvement.

[0065] Finally, the LOD multi-level detail rendering algorithm and intelligent texture compression algorithm are invoked to dynamically adjust the model detail accuracy according to the real-time viewing distance of the simulation lens. In distant scenes, the model surface structure is automatically simplified and the texture accuracy is reduced. In close-range training scenes, the high-definition model details and texture effects are fully preserved. At the same time, lossless compression processing is performed on all scene texture resources to effectively reduce GPU computing power and memory consumption, and stably improve the rendering frame rate to the standard threshold.

[0066] In this embodiment, the step of performing weapon adaptation optimization specifically includes: Parse the weapon type identifier bound to the current training task and retrieve the corresponding dynamic parameter package from the parameter configuration library; The dynamic parameters are imported into the physics engine of the initial battlefield environment, and the electromagnetic interference parameters and weather attenuation coefficients in the environment are dynamically adjusted according to the weapon's guidance method.

[0067] Specifically, the system first reads the unique weapon model identifier bound to the current training mission in real time, and then accurately locates the corresponding parameter package in the parameter configuration library. With one click, it retrieves the exclusive, complete, and unmodified dynamic parameter set of the weapon model, which includes a full set of physical calculation parameters such as flight drag coefficient, aerodynamic adaptation coefficient, fuselage inertial torque, maneuver response delay, attitude dynamic adjustment parameters, and ballistic reference error parameters, ensuring that the parameters are unique, accurate, and closely match the actual performance.

[0068] Then, the retrieved proprietary dynamic parameters are fully injected into the battlefield simulation physics engine, replacing the system's default general parameters with a single click. The overall scene physics calculation rules are updated, ensuring that all weapon movements, trajectory changes, and attitude adjustments on the battlefield are calculated according to actual performance. Simultaneously, the system automatically matches differentiated environment adaptation strategies based on weapon guidance type. Radar-guided weapons automatically adjust corresponding electromagnetic interference intensity, interference frequency band, and interference mode parameters; laser-guided weapons automatically adjust optical transmittance and signal attenuation coefficients in rain, fog, and sandstorm scenarios; and infrared-guided weapons automatically adjust environmental background temperature, artificial heat source interference, and infrared radiation attenuation parameters. Through precise coupling and adaptation of software and hardware parameters, the system maximizes the reproduction of the interaction mechanism between weapons and the environment in a real battlefield, ultimately achieving high-precision, combat-oriented, and practical anti-armor weapon simulation training effects.

[0069] Example 3 Please see Figure 2 The third embodiment of the present invention provides an anti-armor weapon simulation training system based on a large model, applicable to the methods described in any of the above embodiments, the system comprising: Model building module 10 is used to build a large-scale model for anti-armor weapons and a database of battlefield environment elements; The parameter setting module 20 is used to receive the semantic requirements of the trainers and set the training task parameters; Semantic parsing module 30 is used to parse the semantic requirements through the dedicated large model and determine the constraints for generating the battlefield environment; The environment generation module 40 is used to call the corresponding environmental resources from the battlefield environment element library according to the constraints to generate the initial battlefield environment. The training execution module 50 is used to adapt and optimize the initial battlefield environment, output it to the simulation training system for immersive simulation training, and collect training data in real time. The iterative optimization module 60 is used to evaluate the training process based on the training data and to iteratively optimize the dedicated large model and the battlefield environment element library according to the evaluation results.

[0070] Compared with existing technologies, the advantages of using the large-model-based anti-armor weapon simulation training system shown in this invention are as follows: This embodiment can receive personalized semantic training requests from trainees and generate standardized battlefield constraints through intelligent parsing using a dedicated large-scale model. It rapidly generates and adaptively optimizes exclusive battlefield scenarios based on a resource library, overcoming the limitations of traditional technologies that rely on manually preset scripts and fixed resources. This allows for flexible adaptation to diverse and complex real-world training conditions, significantly improving the dynamism and battlefield realism of simulated training scenarios. Simultaneously, this embodiment can collect training data in real time and perform intelligent quantitative evaluation. Based on the evaluation results, iterative optimization of the large-scale model's parsing logic and battlefield resource library is achieved, constructing a complete training optimization loop. This continuously improves the relevance and scientific rigor of training, effectively compensating for the shortcomings of traditional training, such as its inability to dynamically iterate and its limited effectiveness. It precisely meets the current high-quality, personalized training needs for anti-armor weapons in a realistic and personalized manner.

[0071] Example 4 A fourth embodiment of the present invention provides a readable storage medium having computer instructions stored thereon, which, when executed by a processor, implement the steps of the methods described in the above embodiments.

[0072] Example 5 A fifth embodiment of the present invention provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of the methods described in the above embodiments.

[0073] In the description of this specification, references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0074] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these all fall within the protection scope of the present invention. Therefore, the protection scope of this patent should be determined by the appended claims.

Claims

1. A method for simulating and training anti-armor weapons based on a large model, characterized in that, The method includes: Construct a large-scale model for anti-armor weapons and a database of battlefield environment elements; Receive the semantic requirements from the trainers and set the training task parameters; The semantic requirements are analyzed using the dedicated large model to determine the constraints for generating the battlefield environment; Based on the constraints, the corresponding environmental resources are retrieved from the battlefield environment element library to generate the initial battlefield environment; The initial battlefield environment is adapted and optimized, and then output to the simulation training system for immersive simulation training, while collecting training data in real time. The training process is evaluated based on the training data, and the dedicated large model and battlefield environment element library are iteratively optimized based on the evaluation results.

2. The anti-armor weapon simulation training method according to claim 1, characterized in that, The steps for constructing a large-scale model specifically for anti-armor weapons include: Acquire common weapon parameters, operation manuals, guidance principles, tactical specifications, failure case databases, and historical training data of anti-armor weapons as multi-source training sample data; Based on a general large model architecture, entity extraction and relation extraction algorithms are used to extract semantic information from the multi-source training sample data, construct knowledge triples, and form a semantic association network. The general large model is fine-tuned and trained by combining reinforcement learning methods with an expert scoring mechanism to obtain the special large model. The dedicated large model includes an environment generation module, a tactical decision-making module, an interactive understanding module, and a debriefing analysis module.

3. The anti-armor weapon simulation training method according to claim 2, characterized in that, The battlefield environment element library includes a 3D model library, a parameter configuration library, and a dynamic effects library; The 3D model library is used to store exclusive models of anti-armor weapons and their supporting equipment, as well as battlefield element models. The parameter configuration library is used to store the parameter ranges of various environmental elements and the default performance parameters of anti-armor weapons; The dynamic effects library is used to store battlefield dynamic effects and, combined with the six-degree-of-freedom dynamic model of anti-armor weapons, to recreate the dynamic scenes of weapon flight and impact.

4. The anti-armor weapon simulation training method according to claim 1, characterized in that, The step of parsing the semantic requirements using the dedicated large model to determine the constraints for generating the battlefield environment includes: Through the interactive understanding module of the dedicated large model, the semantic requirements are segmented, part-of-speech tagging is performed, and entity recognition is performed to extract core semantic features; The knowledge graph reasoning engine is used to analyze the relationships between entities in the core semantic features; By combining the general performance parameters and tactical rules of anti-armor weapons, the constraints for generating the battlefield environment are determined.

5. The anti-armor weapon simulation training method according to claim 4, characterized in that, The step of analyzing the relationships between entities in the core semantic features using a knowledge graph reasoning engine specifically includes: The extracted core semantic features are mapped to a pre-defined military domain knowledge graph to match the corresponding entity nodes; Based on the edge relationships in the knowledge graph, the causal relationships and tactical collaboration logic between the entity nodes are deduced, and a structured parameter set for constraining scenario generation is output.

6. The anti-armor weapon simulation training method according to claim 1, characterized in that, The step of adapting and optimizing the initial battlefield environment includes: Adjust battlefield environment parameters according to the current weapon type to match the core performance indicators and guidance methods of that weapon type, so as to optimize weapon adaptation. Adjust the complexity of the scenario and the speed of the target's movement according to the training subject and difficulty to optimize training adaptation; The rendering frame rate is increased to no less than a preset threshold by using LOD algorithm and texture compression algorithm in order to optimize rendering adaptation.

7. The anti-armor weapon simulation training method according to claim 6, characterized in that, The steps for weapon adaptation and optimization specifically include: Parse the weapon type identifier bound to the current training task and retrieve the corresponding dynamic parameter package from the parameter configuration library; The dynamic parameters are imported into the physics engine of the initial battlefield environment, and the electromagnetic interference parameters and weather attenuation coefficients in the environment are dynamically adjusted according to the weapon's guidance method.

8. A large-scale anti-armor weapon simulation training system, characterized in that, The system, applicable to the method of any one of claims 1-7, comprises: The model building module is used to build large-scale models of anti-armor weapons and a library of battlefield environment elements. The parameter setting module is used to receive the semantic requirements of the trainers and set the training task parameters; The semantic parsing module is used to parse the semantic requirements through the dedicated large model and determine the constraints for generating the battlefield environment. The environment generation module is used to call the corresponding environmental resources from the battlefield environment element library according to the constraints to generate the initial battlefield environment. The training execution module is used to adapt and optimize the initial battlefield environment, output it to the simulation training system for immersive simulation training, and collect training data in real time. The iterative optimization module is used to evaluate the training process based on the training data and to iteratively optimize the dedicated large model and the battlefield environment element library according to the evaluation results.

9. A readable storage medium having computer instructions stored thereon, characterized in that, When executed by the processor, this instruction implements the steps of the method as described in any one of claims 1-7.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps of the method as described in any one of claims 1-7.

Images