An AF SIM scenario editing method, device and equipment based on a large model and a storage medium

Abstract

The application provides a large model-based AFSIM scenario editing method and device, equipment and storage medium, through a large model, semantic analysis is performed on a natural language task description input by a user, task targets, force allocation, space-time constraints and other scenario key elements are extracted, structured task requirement data is generated, and a semantic bridge between natural language and structured data is established. Based on the large model, the user's intention is understood, and the task is allocated to the corresponding function intelligent agent, the scheme recommendation intelligent agent links the scenario case knowledge base, the weapon equipment knowledge base and the combat skill and method knowledge base to generate force deployment scheme data. The force deployment scheme data is converted into structured JSON data, and an AFSIM scenario script is automatically generated through a template matching mechanism, the format mapping between the scheme data and the simulation script is realized. The generated AFSIM scenario script is loaded to execute simulation deduction and output a structured result report, and an end-to-end automatic process from natural language input to deduction result output is completed.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of artificial intelligence and military simulation, in particular to an AFSIM scenario editing method and device based on a large model, equipment and a storage medium. BACKGROUND

[0002] In the field of military simulation and intelligent decision support, scenario generation is the core link connecting operational requirements and simulation deduction. AFSIM (Advanced Framework for Simulation, Integration and Modeling) is an advanced military simulation platform widely used in operational scheme verification and effectiveness evaluation. However, in the current full process from operational requirements to AFSIM scenario deduction, there is a core technical problem that needs to be solved: how to realize the automatic end-to-end conversion from unstructured natural language task description to executable AFSIM scenario script, and complete multi-scheme parallel deduction and intelligent result analysis.

[0003] In the requirement understanding stage, commanders usually express operational intent in natural language, such as "adopting multiple batches of unmanned aerial vehicles to implement saturation attack on the target from the southeast direction, with a 10-minute interval between batches". This kind of unstructured description contains multi-dimensional information such as task target, force deployment, space-time constraints, etc., but existing systems lack effective semantic analysis mechanisms and cannot automatically extract and structure these key elements.

[0004] In the scheme generation stage, even if the requirement understanding is completed, how to automatically convert structured scheme data (including force deployment, route planning, timing arrangement, etc.) into AFSIM platform executable scenario script still requires a lot of manual writing and format adaptation work. There is a lack of standardized interface and linkage mechanism between military base knowledge base, weapon equipment knowledge base and scheme generation process.

[0005] In the deduction analysis stage, multiple sets of alternative schemes need to be executed separately and compared horizontally, but the existing process relies on manual operation and subjective judgment, lacking automatic deduction scheduling mechanism and intelligent multi-scheme comparison and analysis capability.

[0006] The above problems lead to the defects of long cycle, low efficiency and high dependence on manual work in the traditional scenario generation process. The construction of a single scenario scheme in a complex operational scenario often takes several weeks, which is difficult to meet the timeliness requirements of rapid response in a dynamic battlefield environment. Therefore, an intelligent scenario generation method that can realize full-process automation is needed, which deeply integrates the natural language understanding and knowledge reasoning capabilities of large models with the professional deduction capabilities of AFSIM simulation platform, and builds an end-to-end solution from task description to deduction analysis.

[0007] In view of this, the present application is proposed. SUMMARY

[0008] The application discloses an AF SIM scenario editing method and device based on a large model, equipment and a storage medium, and aims to solve the technical problem that a combat task demand expressed in natural language by a user cannot be automatically converted into a scenario script executable by an AF SIM simulation platform and automatic deduction cannot be completed.

[0009] The first embodiment of the application provides an AF SIM scenario editing method based on a large model, comprising:

[0010] Receiving a natural language task description input by a user, performing semantic analysis on the natural language task description by a large model to extract context information and scenario key elements, and generating structured task demand data;

[0011] Understanding a user intention in the structured task demand data based on a large model, and dynamically assigning a task to a corresponding functional agent according to the user intention;

[0012] A scheme recommendation agent in the functional agent receives the structured task demand data, retrieves a scenario case knowledge base through a scenario case assistant, calls a weapon equipment knowledge base through a force recommendation assistant to recommend force configuration, and calls a battle skill and battle method knowledge base through a deployment planning assistant to generate deployment planning, and comprehensively generates force deployment scheme data;

[0013] A scenario generation agent in the functional agent receives the force deployment scheme data, converts the force deployment scheme data into structured JSON data, and converts the structured JSON data into an AF SIM scenario script through a template matching mechanism;

[0014] Loading the AF SIM scenario script, performing simulation deduction on each scenario scheme respectively, generating deduction result data and outputting as a structured result report file.

[0015] Preferably, the functional agent further comprises a result analysis agent that reads the structured result report file, compares and analyzes the deduction result data of multiple schemes through a large model, and generates an evaluation analysis report.

[0016] Preferably, the scenario key elements comprise a task target, an attack direction, a force type, a force quantity, a time constraint and a space constraint; and the structured task demand data is a data structure containing the scenario key elements.

[0017] Preferably, the method further comprises:

[0018] The scheme recommendation agent receives the structured task requirement data, extracts attack direction, time interval, and resource quantity parameters therefrom by a large model, and checks logical consistency among the parameters to generate checked task parameter data;

[0019] According to the checked task parameter data, target coordinate information is obtained by calling a military geographic knowledge base, and equipment performance parameters are obtained by calling a weapon equipment knowledge base to generate scene basic data;

[0020] Based on the checked task parameter data and the scene basic data, a plurality of sets of differentiated force deployment schemes are generated in combination with constraint conditions, the differentiated force deployment schemes including different attack route combinations and resource configuration combinations;

[0021] The plurality of sets of differentiated force deployment schemes are integrated into force deployment scheme data, the force deployment scheme data including execution steps, a resource list, and risk prompt information.

[0022] Preferably, a scenario generation agent in the functional agent receives the force deployment scheme data, converts the force deployment scheme data into structured JSON data, and converts the structured JSON data into an AFSIM scenario script through a template matching mechanism, specifically:

[0023] The scenario generation agent receives the force deployment scheme data, analyzes the force deployment scheme data through natural language processing technology, automatically identifies core parameters therein and maps them to a preset data model to generate scheme element data;

[0024] The scheme element data is subjected to format checking and logical checking to generate checked scheme element data, wherein the format checking includes numerical range checking, time format checking, and enumeration value legality checking, and the logical checking includes multi-parameter correlation verification;

[0025] The checked scheme element data is converted into structured JSON data, the structured JSON data including scene identification, camp information, path data, and equipment data;

[0026] A template-driven architecture is adopted to match model library data according to model identification in the structured JSON data, and to generate an AFSIM scenario script in combination with camp name, path point information, and spatial coordinate data.

[0027] Preferably, it further includes:

[0028] According to deployment location information in the force deployment scheme data, starting point coordinates and target point coordinates are obtained in combination with a military geographic knowledge base;

[0029] generate three-dimensional track data based on the starting point coordinates and the target point coordinates by using Dijkstra algorithm;

[0030] convert the three-dimensional track data into an array of waypoints and write into the structured JSON data, each waypoint containing longitude, latitude, height coordinates and speed parameters.

[0031] Preferably, the AFSIM scenario script is loaded, simulation deduction is performed for each scenario, deduction result data is generated and output as a structured result report file, specifically:

[0032] The AFSIM Warlock engine is called to load the AFSIM scenario script, parallel simulation deduction is performed on multiple sets of AFSIM scenario scripts, key parameter changes in the deduction process are captured in real time, and deduction result data is generated;

[0033] The time series data, performance index data and abnormal event data in the deduction result data are written into a structured result report file in CSV format.

[0034] The second embodiment of the application provides an AFSIM scenario editing device based on a large model, comprising:

[0035] A user session unit is configured to receive a natural language task description input by a user, perform semantic analysis on the natural language task description by a large model to extract context information and scenario key elements, and generate structured task requirement data;

[0036] An agent routing unit is configured to understand a user intention in the structured task requirement data based on a large model, and dynamically allocate a task to a corresponding functional agent according to the user intention;

[0037] A scenario recommendation unit is configured to receive the structured task requirement data by a scenario recommendation agent in the functional agent, retrieve a scenario case knowledge base by a scenario case assistant, recommend force configuration by a force recommendation assistant calling a weapon equipment knowledge base, generate a deployment plan by a deployment planning assistant calling a combat skill and tactics knowledge base, and comprehensively generate force deployment scheme data;

[0038] A scenario generation unit is configured to receive the force deployment scheme data by a scenario generation agent in the functional agent, convert the force deployment scheme data into structured JSON data, and convert the structured JSON data into an AFSIM scenario script by a template matching mechanism;

[0039] A deduction scheduling unit is configured to load the AFSIM scenario script, perform simulation deduction for each scenario, generate deduction result data and output as a structured result report file.

[0040] The third embodiment of the present application provides a large model-based AFSIM scenario editing device, comprising a memory and a processor, the memory stores a computer program, and the computer program can be executed by the processor to implement the large model-based AFSIM scenario editing method according to any one of claims 1 to 7.

[0041] The fourth embodiment of the present application provides a computer readable storage medium storing a computer program, and the computer program can be executed by the processor of the device where the computer readable storage medium is located to implement the large model-based AFSIM scenario editing method according to any one of the above.

[0042] Based on the large model-based AFSIM scenario editing method, device, equipment and storage medium provided by the present application, the natural language task description input by the user is semantically analyzed by the large model, the task target, force configuration, space-time constraint and other scenario key elements are extracted, and structured task requirement data is generated, thereby establishing a semantic bridge between natural language and structured data. Based on the large model, the user's intention is understood and the task is allocated to the corresponding functional agent, and the scheme recommendation agent links the scenario case knowledge base, the weapon equipment knowledge base and the battle skill and tactics knowledge base to generate force deployment scheme data. The force deployment scheme data is converted into structured JSON data, and the AFSIM scenario script is automatically generated through the template matching mechanism, realizing the format mapping between the scheme data and the simulation script. The generated AFSIM scenario script is loaded to execute simulation deduction and output structured result report, completing the end-to-end automation process from natural language input to deduction result output. BRIEF DESCRIPTION OF DRAWINGS

[0043] Figure 1 is a flowchart of the large model-based AFSIM scenario editing method provided by the first embodiment of the present application;

[0044] Figure 2 is a module schematic diagram of the large model-based AFSIM scenario editing device provided by the second embodiment of the present application. DETAILED DESCRIPTION

[0045] The technical solutions in the embodiments of the present application will be described in detail below with reference to the drawings of the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative labor are within the scope of protection of the present application.

[0046] In order to better understand the technical solutions of the present application, the embodiments of the present application will be described in detail below with reference to the drawings.

[0047] The application discloses an AF SIM scenario editing method and device based on a large model, equipment and a storage medium, and aims to solve the technical problem that a combat task demand expressed in natural language by a user cannot be automatically converted into a scenario script executable by an AF SIM simulation platform and automatic deduction cannot be completed.

[0048] Please refer to Figure 1 The first embodiment of the application provides an AF SIM scenario editing method based on a large model, which can be executed by an AF SIM scenario editing device based on a large model (hereinafter referred to as a generation device or system), in particular, by one or more processors in the generation device or system, to at least implement the following steps:

[0049] S101, receiving a natural language task description input by a user, performing semantic analysis on the natural language task description by a large model to extract context information and scenario key elements, and generating structured task demand data;

[0050] In this embodiment, the generation device or system can be a desktop computer, a notebook computer, a server, a workstation or the like terminal with data processing capability, and a corresponding operating system and application software can be installed in the generation device or system, and the functions required in this embodiment can be realized by the combination of the operating system and the application software.

[0051] Specifically, in this embodiment, the user can input a combat task description in the form of natural language through a man-machine interaction interface, for example, "adopting multiple batches of unmanned aerial vehicles to implement saturated attack on the CC base from the southeast direction, deploying 3 attack unmanned aerial vehicles in each batch, with a batch interval of 10 minutes, and requiring to complete all attack tasks within 2 hours". After the system receives the above natural language task description, a large model based on a Transformer architecture is called to perform semantic analysis processing. The large model performs deep semantic understanding on the input text through a multi-layer self-attention mechanism, and automatically identifies and extracts scenario key elements contained therein.

[0052] Specifically, the assumed key elements extracted by the large model from the above example description include: the task goal is "to carry out a saturation attack on the CC base", the attack direction is "southeast", the force type is "attack-type unmanned aerial vehicle", the force quantity is "3 per batch", the time constraint is "10 minutes between batches" and "completed within 2 hours", and the space constraint is the target location "CC base" and its associated geographic coordinate information. While extracting the above assumed key elements, the large model also conducts logical consistency verification in combination with the dialogue context information, such as verifying whether the semantic association between "multiple batch attacks" and "10-minute batch intervals" is reasonable, and whether there is a conflict between the time constraint of "completed within 2 hours" and the number of batches and interval time. After completing semantic analysis and logical verification, the system structures and encapsulates the extracted assumed key elements according to the preset data model, generating structured task requirement data. The structured task requirement data is organized in the form of key-value pairs, including the task goal field, the attack direction field, the force type field, the force quantity field, the time constraint field, and the space constraint field, providing standardized data input for subsequent agent routing and scheme generation.

[0053] S102, based on the large model understanding the user's intention in the structured task requirement data, dynamically assigning the task to the corresponding functional agent according to the user's intention;

[0054] In this embodiment, after the system receives the structured task requirement data generated in the previous step, it is processed by the agent routing module. Its core component is an agent router built based on an open-source large model (such as qwen, etc.), which has the ability to deeply understand the semantic and intent of the structured task requirement data. The agent router first parses the field information in the structured task requirement data, including the task goal, attack direction, force type, force quantity, time constraint, and space constraint, and comprehensively analyzes the above field information to understand the user's true operational intent.

[0055] For example, when the task goal field in the structured task requirement data is "to carry out a saturation attack on the CC base" and the force type field is "attack-type unmanned aerial vehicle", the agent router determines that the user's intention is to perform a joint operation task of unmanned aerial vehicles, and needs to complete the complete process from scheme recommendation to scenario generation to execution. Based on the above intention understanding result, the agent router dynamically assigns the task to the corresponding functional agent according to the preset task allocation strategy.

[0056] The function agent in the embodiment adopts a multi-agent collaborative architecture, including three types of core agents, namely, a scheme recommendation agent, a scenario generation agent, and a result analysis agent, and each type of agent is provided with a plurality of special assistant modules. An agent router determines a task execution path according to the specific type of user intention. For an intention that requires generation of a combat scheme, the router allocates structured task demand data to the scheme recommendation agent. For an intention that requires conversion of an existing scheme into a simulation script, the router allocates the task to the scenario generation agent. For an intention that requires analysis of a deduction result, the router allocates the task to the result analysis agent. In the unmanned aerial vehicle joint combat scene in the embodiment, the agent router identifies that the current task requires generation of a complete combat scheme from scratch and execution of deduction, and therefore sequentially dispatches the function agents in the order of "scheme recommendation agent, scenario generation agent, deduction scheduling, result analysis agent", to realize multi-agent collaboration to complete the user's combat task demand.

[0057] In S103, the scheme recommendation agent in the function agent receives the structured task demand data, retrieves a scenario case knowledge base through a scenario case assistant, calls a weapon equipment knowledge base to recommend force configuration through a force recommendation assistant, generates a deployment plan through a deployment planning assistant calling a combat skill and tactics knowledge base, and comprehensively generates force deployment scheme data.

[0058] In the embodiment, after the scheme recommendation agent receives the structured task demand data allocated by the agent router, a multi-dimensional scheme generation process is started. The scheme recommendation agent first performs deep analysis on the structured task demand data through a built-in large model, extracts core task parameters therefrom, including an attack direction parameter (such as "southeast direction"), a time interval parameter (such as "batch interval 10 minutes"), and a resource quantity parameter (such as "3 unmanned aerial vehicles per batch"). The large model verifies the logical consistency among the parameters while extracting the parameters, for example, verifies whether the tactical combination of "single direction attack" and "multi-batch unmanned aerial vehicle cluster" is reasonable, and whether there is a contradiction between the batch interval time and the total task time limit, and generates the verified task parameter data after the verification is passed.

[0059] Subsequently, the scheme recommendation agent calls its three types of special assistant modules to work collaboratively. The scenario case assistant first searches the scenario case knowledge base, matches the historical scenario cases similar to the current task scenario in the knowledge base, and if a scenario case that meets the conditions is matched, extracts the scheme elements in the case as a reference template; if no similar case is matched, the force recommendation assistant and the deployment planning assistant generate a scheme from scratch based on the knowledge base data. The force recommendation assistant calls the weapon equipment knowledge base to obtain the performance parameters of the corresponding equipment according to the verified force type and quantity requirements in the task parameter data, including the flight speed, combat radius, payload, sensor configuration and other technical indicators of the unmanned aerial vehicle, and calls the military geographic knowledge base to obtain the accurate coordinate information of the target area "CC base" and the surrounding terrain data, and comprehensively generates the scene basic data. The deployment planning assistant receives the scene basic data, calls the tactics and techniques knowledge base to select the tactics suitable for the current combat scenario, such as selecting the "multi-batch wave attack" tactics for the "saturation attack" task target, and generates the specific deployment position and force quantity configuration based on the tactics rule and the tactical planning algorithm.

[0060] After the above knowledge base linkage is completed, the scheme recommendation agent generates multiple sets of differentiated force deployment schemes based on the verified task parameter data and scene basic data, combined with time window, space range, resource total amount and other constraint conditions, through the built-in scheme generation algorithm. Taking the unmanned aerial vehicle joint combat task in the embodiment as an example, the scheme recommendation agent generates at least three sets of differentiated schemes: scheme one adopts a single route attack in the southeast direction, deploys 4 batches of a total of 12 unmanned aerial vehicles, and the batch interval is 10 minutes; scheme two adopts a double direction cooperative attack in the southeast and east directions, deploys 2 batches of a total of 6 unmanned aerial vehicles in each direction, and the batch interval is 8 minutes; scheme three adopts a combined attack route of the southeast direction as the main attack and the south direction as the feint attack, deploys 3 batches of a total of 9 unmanned aerial vehicles in the main attack direction, and deploys 1 batch of a total of 3 unmanned aerial vehicles in the feint attack direction. The above differentiated schemes reflect different attack route combinations and resource configuration combinations, providing users with diversified tactical options.

[0061] Finally, the scheme recommendation agent integrates the above multiple sets of differentiated schemes into a unified format of force deployment scheme data. The force deployment scheme data is organized in a structured form, including the detailed execution steps of each scheme (such as the take-off time, flight path planning, attack timing of each batch of unmanned aerial vehicles), resource list (such as the model, quantity, and mounted weapon type of each batch of unmanned aerial vehicles), and risk prompt information (such as potential risks that each scheme may face, such as air defense threat, weather influence, and insufficient time margin), providing complete scheme input data for the subsequent scenario generation link.

[0062] S104, a scenario generation agent in the functional intelligent agent receives the force deployment scheme data, converts the force deployment scheme data into structured JSON data, and converts the structured JSON data into an AFSIM scenario script through a template matching mechanism;

[0063] In this embodiment, after the scenario generation agent receives the force deployment scheme data output by the scheme recommendation agent, a multi-stage data processing procedure is started to convert the unstructured or semi-structured scheme document into an AFSIM scenario script executable by the AFSIM simulation platform.

[0064] The scenario generation agent first calls its built-in natural language processing module to perform in-depth analysis on the force deployment scheme data. The module automatically identifies the core parameters in the scheme data through semantic analysis technology, including the attack direction, batch interval, number of unmanned aerial vehicles, weapon type, deployment location, and launch time, and maps the above parameters to the preset data model structure to generate scheme element data. Taking the unmanned aerial vehicle joint operation task in this embodiment as an example, the scenario generation agent extracts the core parameters from the scheme data, such as the starting point and target point coordinates of the "southeast direction attack" route, the timestamp sequence corresponding to the "batch interval 10 minutes", and the platform type and number information corresponding to "3 attack type unmanned aerial vehicles per batch".

[0065] Subsequently, the scenario generation agent performs double-checking processing on the generated scheme element data. At the format checking level, the system checks the rationality of the numerical range, such as verifying whether the number of unmanned aerial vehicles is a positive integer and whether the flight altitude is within the allowable altitude range of the equipment; checks the standardization of the time format, such as verifying whether the batch interval conforms to the time representation specification of "HH:MM:SS"; and checks the legality of the enumeration value, such as verifying whether the unmanned aerial vehicle model exists in the pre-defined platform type library. At the logic checking level, the system verifies the associativity constraints between multiple parameters, such as the time constraint of "batch interval 10 minutes" needs to match the generation rule of the timestamp sequence, and the launch time of each batch of unmanned aerial vehicles needs to meet the interval requirements of the previous and subsequent batches. After the above double-checking, the system generates the checked scheme element data.

[0066] After the verification process is completed, the scenario generation agent converts the verified scenario element data into structured JSON data. The JSON data is organized using a hierarchical data structure, including four levels: scenario layer, camp layer, path layer, and equipment layer. The scenario layer includes a scenario identification field to distinguish different scenarios; the camp layer includes a camp name field (such as "red side"), a path array field, and an equipment array field; the path layer includes a path identification field and a waypoint array field, each waypoint including position coordinates (longitude, latitude, altitude), speed, and label name information; the equipment layer includes a model identification field, a quantity field, an associated path field, a deployment location field, and a deployment time field. In this example, the structured JSON data generated includes complete flight path information from the takeoff point to the target area via multiple waypoints, and equipment data includes the model identification "zbwrj" of the attack-type unmanned aerial vehicle, the quantity "1", the associated path identification "route_0_2", the initial deployment location coordinates, and the deployment time "1200" seconds, etc.

[0067] Finally, the scenario generation agent uses a template-driven architecture to convert structured JSON data into AFSIM scenario scripts. The system locates and extracts the camp name, path identifier, waypoint information under the "camps" node, and the model identifier, path information, spatial coordinates under the "equipments" node, and creates structured data objects through a JSON recursive parsing algorithm. Subsequently, the system matches the platform type data in the AFSIM model library according to the model identifier field to obtain the corresponding platform attribute configuration. On this basis, the system reads the standard templates in the AFSIM scenario template library and performs template replacement operations combined with the camp name, path point information, and spatial coordinate data to generate scenario script code that meets the AFSIM syntax specifications. The generated path definition script contains path identifier, point name, position coordinates, height, and speed information, formatted as "route {path identifier} label {point name} position {latitude} {longitude} altitude {altitude} mspeed {speed} end_route"; the generated platform definition script contains platform entity name, model identifier, camp, position coordinates, launch time, and path following instructions, formatted as "platform {platform name} {model identifier} side {camp} position {latitude} {longitude} altitude {altitude} m execute at_time {launch time} secabsolute PLATFORM.FollowRoute({path name}) end_execute end_platform". In the platform type definition, the system automatically associates the pre-set task planning logic in the scenario template, including automatic target detection and automatic target attack tactical behavior scripts, to ensure that the generated AFSIM scenario script has complete combat task execution capabilities.

[0068] S105, load the AFSIM scenario script, execute simulation deduction for each scenario respectively, generate deduction result data and output as a structured result report file.

[0069] In this embodiment, after the scenario generation agent completes the generation of the AFSIM scenario script, the system automatically calls the AFSIM deduction scheduling module to execute the simulation deduction task. The AFSIM deduction scheduling module interacts with the AFSIM simulation engine through a standardized interface, uses a closed-loop execution logic of "parameter injection, process monitoring, and exception intervention", and realizes full-process automation control from script loading to result output.

[0070] The deduction scheduling module first calls the Warlock engine of the AFSIM platform to load the AFSIM scenario script output by the scenario generation intelligent agent. As the core deduction execution component of the AFSIM platform, the Warlock engine is responsible for parsing the simulation elements defined in the scenario script, such as platform entities, path planning, and task logic, and constructing the corresponding simulation scene in the virtual battlefield environment. For the multiple differentiated force deployment schemes generated in the scheme recommendation link, the deduction scheduling module adopts a parallel deduction mechanism to start independent simulation deduction processes for the AFSIM scenario scripts corresponding to each scheme. Taking the UAV joint operation task in the embodiment as an example, the system simultaneously loads multiple scenario scripts such as scheme one (single route attack in the southeast direction), scheme two (coordinated attack in the southeast and east directions), and scheme three (combined attack of main attack and feint attack), and performs simulation calculation in parallel in the Warlock engine, which significantly improves the overall efficiency of multi-scheme deduction.

[0071] During the simulation deduction execution process, the deduction scheduling module captures the changes of key parameters in the deduction process in real time and generates deduction result data. The captured data content includes three types of time series data, performance index data, and abnormal event data. The time series data records the state change trajectory of each platform entity during the simulation clock advancement process, including the real-time position coordinates, flight speed, heading angle, and remaining fuel of the UAV; the performance index data records statistical information related to combat effectiveness, including the number of force losses of the red and blue sides, weapon hit rate, task completion rate, and target damage degree; the abnormal event data records special events occurring during the deduction process, including the occurrence time and specific parameters of tactical events such as platform destruction, weapon launch, target detection, and communication interruption.

[0072] After the deduction execution is completed, the deduction scheduling module writes the collected deduction result data into a structured result report file in a specified directory. The result report file is stored in CSV format to ensure data standardization and cross-tool compatibility. Each scenario scheme corresponds to an independent CSV file, and the file records data entries in the entire deduction process in chronological order, and each data entry contains simulation timestamp, event type, associated platform, and parameter value fields. Through the above processing, the system realizes the full-link automation from scenario script loading, simulation deduction execution to result data output, and provides complete and standardized deduction data support for the subsequent result analysis link.

[0073] In one possible implementation of the present application, the result analysis intelligent agent in the functional intelligent agent reads the structured result report file and compares and analyzes the deduction result data of multiple schemes through a large model to generate an evaluation analysis report.

[0074] The result analysis agent first accesses the system designated result storage directory, reads the CSV format structured result report file corresponding to each scheme. Taking the unmanned aerial vehicle joint operation task in this embodiment as an example, the result analysis agent reads the deduction result files of scheme one, scheme two, scheme three, and scheme four in turn, extracts time series data, performance index data, and abnormal event data and other original deduction records from each file, and constructs a unified multi-scheme data set for subsequent comparative analysis.

[0075] After completing data reading, the result analysis agent uses data visualization and statistical analysis methods to make horizontal comparison of the multi-scheme deduction results. Based on the deduction result data in the structured result report file, the system extracts the performance values of each scheme on the core performance indicators, including task completion rate (such as target damage ratio, attack task achievement degree), resource consumption (such as red and blue forces loss quantity, ammunition consumption), and time efficiency (such as task completion time, each stage execution time) and other key indicators. The system summarizes and arranges the above index data according to the scheme dimension, calculates the numerical difference and ranking result of each scheme on the same index, and generates index comparison data. The index comparison data is organized in a matrix form, with the row dimension being the scenario scheme identifier, the column dimension being the performance evaluation index, and the matrix element being the quantitative value of the corresponding scheme on the corresponding index.

[0076] Subsequently, the result analysis agent calls the large model to perform deep intelligent analysis on the index comparison data. The large model constructs a "performance-cost-risk" three-dimensional evaluation matrix based on the index comparison data, and performs multi-dimensional quantitative evaluation on the comprehensive combat performance of each scheme. The large model performs deep understanding of the deduction data through semantic analysis technology, automatically identifies the advantages and potential defects of each scheme, and analyzes the influence law of different tactical configurations on the combat effect. Taking this embodiment as an example, when analyzing the deduction results of scheme four, the large model accurately identifies that this scheme realizes the balanced configuration of the tactical capabilities of the red and blue sides, and the blue side adopts a multi-batch attack strategy to improve the attack success rate by about thirty percent compared with other schemes, while the red side defense system maintains effective interception capability in continuous confrontation. The large model comprehensively analyzes the above analysis results, automatically identifies the key characteristics of the optimal scheme, including force configuration ratio, batch interval setting, attack route selection, and other core elements that affect the combat effect.

[0077] Finally, the result analysis intelligent agent automatically generates an evaluation analysis report based on the multi-dimensional analysis results of the large model. The report is presented in a structured document form, including trend charts of multi-scheme performance indicators (such as a comparison bar chart of task completion rates of each scheme, resource consumption change curves), significant difference analysis of each indicator dimension (such as the impact degree analysis of different attack routes on task achievement rate), optimal scheme recommendation and its advantage feature explanation, and improvement suggestions for non-optimal schemes, etc. Through the above processing, the system realizes the full-process automation from data collection, multi-dimensional comparative analysis to intelligent evaluation report generation, providing scientific and objective quantitative basis support for users' operational decision-making.

[0078] In a possible implementation manner of the present application, the method further comprises:

[0079] According to the deployment location information in the force deployment scheme data, the starting point coordinates and the target point coordinates are obtained in combination with a military geographic knowledge base;

[0080] The Dijkstra algorithm is adopted to generate three-dimensional flight path data based on the starting point coordinates and the target point coordinates;

[0081] The three-dimensional flight path data is converted into a flight point array and written into the structured JSON data, and each flight point contains longitude, latitude, height coordinates and speed parameters.

[0082] Specifically, in the present embodiment, the scenario generation intelligent agent needs to perform path planning processing to generate the flight path of each combat platform in the process of converting the force deployment scheme data into structured JSON data. The path planning processing automatically generates three-dimensional flight path data that meets the combat task requirements based on the deployment location information in the force deployment scheme data, in combination with a military geographic knowledge base and a path planning algorithm.

[0083] The scenario generation intelligent agent first parses the deployment location information in the force deployment scheme data, extracts the initial deployment location and attack target location of each batch of unmanned aerial vehicles. Then, the system calls the military geographic knowledge base to obtain the accurate geographic coordinate data corresponding to the above locations. The military geographic knowledge base stores the spatial information of various geographic elements in the combat area, including military base locations, important target coordinates, terrain elevation data, no-fly zone boundaries, etc. Taking the unmanned aerial vehicle joint combat task in the present embodiment as an example, the system obtains the starting point coordinates of the unmanned aerial vehicle take-off base and the target point coordinates of the attack target "CC base" from the military geographic knowledge base, and extracts the terrain elevation data along the flight route and the potential threat area distribution information, providing geographic environment constraint conditions for subsequent flight path planning.

[0084] After the coordinate data acquisition is completed, the scenario generation agent adopts Dijkstra algorithm to perform three-dimensional flight path planning calculation. As a classic shortest path search algorithm, Dijkstra algorithm can solve the optimal path from the starting point to the target point under the consideration of various constraint conditions. In the flight path planning application of the present application, the system first constructs a three-dimensional space grid model based on the geographical range of the combat area, and discretizes the continuous flight space into a calculable node network. Then, the system maps the constraint factors such as terrain obstacles, air defense threat areas, no-fly zones, etc. into the passing cost weight of the grid nodes, and the higher the threat degree of the area, the greater the passing cost. On this basis, Dijkstra algorithm takes the starting point coordinates as the source node and the target point coordinates as the terminal node, and calculates the three-dimensional flight path with the minimum comprehensive passing cost through iterative search, to generate three-dimensional flight path data. The three-dimensional flight path data contains a complete space trajectory from the starting point to the target point via several intermediate waypoints, and each waypoint has a clear three-dimensional space coordinate and a recommended flight speed.

[0085] Finally, the scenario generation agent converts the generated three-dimensional flight path data into a standardized waypoint array format and writes it into the path layer field of the structured JSON data. Each waypoint element in the waypoint array contains four core parameters: the longitude coordinate represents the east-west geographical position of the waypoint, the latitude coordinate represents the south-north geographical position of the waypoint, the height coordinate represents the flight altitude of the waypoint, and the speed parameter represents the recommended flight speed of the platform passing through the waypoint. Taking the present embodiment as an example, the generated waypoint array contains the takeoff point waypoint (longitude 92.914400, latitude 36.810500, height 6000 meters, speed 60 meters per second), the intermediate waypoint (longitude 92.901788, latitude 36.804388, height 6000 meters, speed 60 meters per second), and the end waypoint near the target area (longitude 92.896000, latitude 36.797663, height 6000 meters, speed 60 meters per second), etc. After the above waypoint array data is written into the structured JSON data, it forms a complete scenario data structure together with the camp information, equipment data, etc. field, providing accurate flight path planning input for subsequent AFSIM scenario script generation.

[0086] Please refer to Figure 2 The second embodiment of the present application provides an AFSIM scenario editing device based on a large model, comprising:

[0087] The user session unit 201 is used to receive the natural language task description input by the user, perform semantic analysis on the natural language task description through the large model to extract context information and scenario key elements, and generate structured task requirement data;

[0088] The agent routing unit 202 is configured to understand a user intention in the structured task demand data based on a large model, and dynamically allocate a task to a corresponding functional agent according to the user intention.

[0089] The scheme recommendation unit 203 is configured to receive the structured task demand data by a scheme recommendation agent in the functional agent, retrieve a scenario case knowledge base by a scenario case assistant, call a weapon equipment knowledge base by a force recommendation assistant to recommend a force configuration, and call a battle skill and tactics knowledge base by a deployment planning assistant to generate a deployment planning, and comprehensively generate force deployment scheme data.

[0090] The scenario generation unit 204 is configured to receive the force deployment scheme data by a scenario generation agent in the functional agent, convert the force deployment scheme data into structured JSON data, and convert the structured JSON data into an AFSIM scenario script by a template matching mechanism.

[0091] The deduction and scheduling unit 205 is configured to load the AFSIM scenario script, perform simulation deduction on each scenario scheme respectively, generate deduction result data and output the deduction result data as a structured result report file.

[0092] The third embodiment of the present application provides an AFSIM scenario editing device based on a large model, which comprises a memory and a processor, the memory stores a computer program, and the computer program can be executed by the processor to realize the AFSIM scenario editing method based on a large model according to any one of claims 1 to 7.

[0093] The fourth embodiment of the present application provides a computer readable storage medium, which stores a computer program, and the computer program can be executed by a processor of a device where the computer readable storage medium is located to realize the AFSIM scenario editing method based on a large model according to any one of the above.

[0094] Based on the AF SIM scenario editing method, device, equipment and storage medium provided by the application, the natural language task description input by the user is semantically analyzed by a large model, task target, force configuration, space-time constraint and other scenario key elements are extracted, structured task requirement data is generated, and a semantic bridge between natural language and structured data is established. Based on the large model, the user's intention is understood and the task is assigned to the corresponding function agent, the scheme recommendation agent links the scenario case knowledge base, the weapon equipment knowledge base and the combat skill and method knowledge base to generate force deployment scheme data. The force deployment scheme data is converted into structured JSON data, and the AFSIM scenario script is automatically generated through the template matching mechanism, the format mapping between the scheme data and the simulation script is realized. The generated AFSIM scenario script is loaded to execute simulation deduction and output structured result report, and the end-to-end automation process from natural language input to deduction result output is completed.

[0095] Exemplarily, the computer program in the third and fourth embodiments of the application can be divided into one or more modules stored in the memory and executed by the processor to complete the application. The one or more modules can be a series of computer program instruction segments capable of completing a specific function, which are used to describe the execution process of the computer program in the implementation of the large model-based AFSIM scenario editing device. For example, the device in the second embodiment of the application.

[0096] The processor can be a central processing unit (CPU), and can also be other general-purpose processors, digital signal processors (DSP), application specific integrated circuits (ASIC), ready programmable gate arrays (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor, etc. The processor is the control center of the large model-based AFSIM scenario editing method, and connects all parts of the implementation of the large model-based AFSIM scenario editing method through various interfaces and lines.

[0097] The memory can be used to store the computer program and / or modules, and the processor realizes various functions of the large model-based AF SIM scenario editing method by running or executing the computer program and / or modules stored in the memory, and calling the data stored in the memory. The memory can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, at least one application required by a function (such as a sound playing function, a text conversion function, etc.), and the like; and the data storage area can store data (such as audio data, text message data, etc.) created according to the use of the mobile phone, and the like. In addition, the memory can include a high-speed random access memory, and can also include a non-volatile memory, for example, a hard disk, a memory, a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, at least one disk storage device, a flash memory device, or other volatile solid-state storage devices.

[0098] The implemented modules, if realized in the form of software function units and sold or used as independent products, can be stored in a computer readable storage medium. Based on this understanding, all or part of the processes in the above-mentioned embodiment methods can also be completed by a computer program instructing related hardware, and the computer program can be stored in a computer readable storage medium. The computer program can realize the steps of the above-mentioned various method embodiments when executed by a processor. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or some intermediate forms, etc. The computer readable medium can include any entity or device capable of carrying the computer program code, recording medium, U disk, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signal, telecommunication signal, and software distribution medium, etc. It should be noted that the contents included in the computer readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction, for example, in some jurisdictions, according to legislation and patent practice, the computer readable medium does not include electrical carrier signals and telecommunication signals.

[0099] It should be noted that the apparatus embodiments described above are merely illustrative, and the units described as separate units can or can not be physically separate, and the units shown as units can or can not be physical units, i.e., can be located in one place, or can be distributed to multiple network units. Part or all of the modules can be selected to achieve the purpose of the embodiment according to actual needs. In addition, the connection relationship between the modules in the apparatus embodiment provided by the present application indicates that there is a communication connection between them, which can be implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement it without creative labor.

[0100] The above is only a preferred specific embodiment of the present application, but the protection scope of the present application is not limited thereto, and any person skilled in the art can easily think of changes or replacements within the technical range disclosed by the present application, which should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. An AFSIM scenario editing method based on a large model, characterized in that, include: The system receives a natural language task description input by the user, performs semantic parsing on the natural language task description through a large model to extract contextual information and key elements, and generates structured task requirement data. Based on the large model, the user intent in the structured task requirement data is understood, and the task is dynamically assigned to the corresponding functional agent according to the user intent; The scheme recommendation agent in the functional agent receives the structured task requirement data, and generates troop deployment scheme data by comprehensively generating the troop deployment scheme data through the scenario case assistant searching the scenario case knowledge base, the troop recommendation assistant calling the weapon and equipment knowledge base to recommend troop configuration, and the deployment planning assistant calling the combat skills and tactics knowledge base to generate deployment plans. The scenario generation agent in the functional agent receives the troop deployment plan data, converts the troop deployment plan data into structured JSON data, and converts the structured JSON data into AFSIM scenario scripts through a template matching mechanism. Specifically, the scenario generation agent receives the troop deployment plan data, parses the troop deployment plan data through natural language processing technology, automatically identifies the core parameters and maps them to a preset data model to generate plan element data. The scheme element data undergoes format and logic validation to generate validated scheme element data. The format validation includes numerical range checks, time format checks, and enumeration value validity checks. The logic validation includes multi-parameter correlation validation. The validated scheme element data is then converted into structured JSON data, which includes scene identifiers, faction information, path data, and equipment data. Using a template-driven architecture, model library data is matched based on the model identifiers in the structured JSON data, and an AFSIM scenario script is generated by combining faction names, path location information, and spatial coordinate data. Based on the deployment location information in the troop deployment plan data, the starting point coordinates and target point coordinates are obtained by combining the military geography knowledge base; the Dijkstra algorithm is used to generate three-dimensional track data based on the starting point coordinates and target point coordinates; the three-dimensional track data is converted into a waypoint array and written into the structured JSON data, each waypoint containing longitude, latitude, altitude coordinates and speed parameters; Load the AFSIM scenario script, perform simulations for each scenario, generate simulation result data, and output it as a structured result report file.

2. The AFSIM scenario editing method based on a large model according to claim 1, characterized in that, Also includes: The result analysis agent in the functional agent reads the structured result report file, compares and analyzes the simulation results data of multiple schemes through a large model, and generates an evaluation analysis report.

3. The AFSIM scenario editing method based on a large model according to claim 1, characterized in that, The key elements of the scenario include: mission objective, attack direction, troop type, troop quantity, time constraints, and space constraints; the structured mission requirement data is a data structure that includes the key elements of the scenario.

4. The AFSIM scenario editing method based on a large model according to claim 1, characterized in that, Also includes: The proposed solution recommends that the intelligent agent receive the structured task requirement data, extract the attack direction, time interval, and resource quantity parameters from it through a large model, verify the logical consistency between the parameters, and generate verified task parameter data. Based on the verified task parameter data, the target coordinate information is obtained by calling the military geography knowledge base, and the equipment performance parameters are obtained by calling the weapon and equipment knowledge base to generate basic scene data. Based on the verified task parameter data and the scenario basic data, multiple differentiated troop deployment schemes are generated in combination with constraints. The differentiated troop deployment schemes include different combinations of attack routes and resource configurations. The multiple differentiated troop deployment schemes are integrated into troop deployment scheme data, which includes execution steps, resource lists, and risk warning information.

5. The AFSIM scenario editing method based on a large model according to claim 1, characterized in that, The process of loading the AFSIM scenario script, performing simulations for each scenario, generating simulation result data, and outputting it as a structured result report file is as follows: The AFSIM Warlock engine is invoked to load the AFSIM scenario script, and parallel simulation is performed on multiple AFSIM scenario scripts. The changes in key parameters during the simulation process are captured in real time, and simulation result data is generated. Write the time series data, performance index data, and abnormal event data from the simulation results into a structured results report file in CSV format.

6. An AFSIM scenario editing device based on a large model, characterized in that, include: The user session unit is used to receive natural language task descriptions input by users, perform semantic parsing on the natural language task descriptions through a large model to extract contextual information and key elements, and generate structured task requirement data. The intelligent agent routing unit is used to understand the user intent in the structured task requirement data based on the large model, and dynamically allocate tasks to the corresponding functional intelligent agents according to the user intent; The scheme recommendation unit is used by the scheme recommendation agent in the functional agent to receive the structured task requirement data, retrieve the scenario case knowledge base through the scenario case assistant, recommend the troop configuration by calling the weapon and equipment knowledge base through the troop recommendation assistant, and generate the deployment plan by calling the combat skills and tactics knowledge base through the deployment planning assistant, and comprehensively generate troop deployment scheme data. The scenario generation unit is used by the scenario generation agent in the functional agent to receive the troop deployment plan data, convert the troop deployment plan data into structured JSON data, and convert the structured JSON data into AFSIM scenario scripts through a template matching mechanism. Specifically, the scenario generation agent receives the troop deployment plan data, parses the troop deployment plan data through natural language processing technology, automatically identifies the core parameters and maps them to a preset data model, and generates plan element data. The scheme element data undergoes format and logic validation to generate validated scheme element data. The format validation includes numerical range checks, time format checks, and enumeration value validity checks. The logic validation includes multi-parameter correlation validation. The validated scheme element data is then converted into structured JSON data, which includes scene identifiers, faction information, path data, and equipment data. Using a template-driven architecture, model library data is matched based on the model identifiers in the structured JSON data, and an AFSIM scenario script is generated by combining faction names, path location information, and spatial coordinate data. Based on the deployment location information in the troop deployment plan data, the starting point coordinates and target point coordinates are obtained by combining the military geography knowledge base; the Dijkstra algorithm is used to generate three-dimensional track data based on the starting point coordinates and target point coordinates; the three-dimensional track data is converted into a waypoint array and written into the structured JSON data, each waypoint containing longitude, latitude, altitude coordinates and speed parameters; The simulation scheduling unit is used to load the AFSIM scenario script, perform simulation simulations on each scenario, generate simulation result data, and output it as a structured result report file.

7. An AFSIM-based hypothetical editing device based on a large model, characterized in that, It includes a memory and a processor, wherein the memory stores a computer program that can be executed by the processor to implement an AFSIM scenario editing method based on a large model as described in any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that, The device contains a computer program that can be executed by a processor of the device in which the computer-readable storage medium is located, to implement the AFSIM scenario editing method based on a large model as described in any one of claims 1 to 5.

Images