A radar countermeasure simulation training method and system based on a multi-modal large model

Abstract

This invention discloses a radar countermeasure simulation training method and system based on a multimodal large model, belonging to the field of radar simulation training technology. The method includes: receiving training requirements input by trainees in natural language, parsing and extracting core elements, and initializing radar and target parameters. Based on these requirements, it generates complete electromagnetic environment scenarios, multi-system radar signals conforming to tactical situations, synthesizes waveforms, simulates complex environmental effects, and outputs standardized files. A real-time simulation engine unit simulates radar power-on and search scanning, generating target echoes. An intelligent countermeasure strategy unit perceives the situation, intercepts signals, identifies threats, and generates adaptive countermeasure strategies. An intelligent evaluation and feedback unit collects data throughout the process, quantifies and scores data through a large model, checks compliance, and generates training evaluation results in real time. This invention solves the problem in existing technologies where radar countermeasure training is difficult to adapt to complex and ever-changing electromagnetic countermeasure requirements.

Description

Technical Field

[0001] This invention relates to the field of radar simulation training technology, and in particular to a radar adversarial simulation training method and system based on a multimodal large model. Background Technology

[0002] As a core intelligence tool in electronic warfare, radar plays a crucial role in battlefield electromagnetic reconnaissance, electronic countermeasures, equipment effectiveness verification, and operator skills training. With the increasingly complex electromagnetic battlefield environment and the ever-increasing technical requirements for radar countermeasures, high-fidelity and highly intelligent simulation training systems are urgently needed in the research, development, upgrading, application, and maintenance of radar systems to test equipment countermeasure performance and improve operators' combat readiness.

[0003] While field-based radar countermeasures training can directly test the capabilities of equipment and personnel, it suffers from problems such as stringent training site requirements, high training costs, long testing cycles, and high safety risks. Furthermore, field-based training cannot reproduce real-world scenarios of multi-system radar and multi-target collaborative countermeasures in complex electromagnetic environments, and the limited types of countermeasures and signals restrict the training effectiveness.

[0004] In recent years, the development of computer simulation technology has promoted the application of radar countermeasure simulation training systems. Traditional radar countermeasure simulation training schemes are mainly implemented through hardware configuration, fixed rule bases, or manual script writing. The core is to receive radar transmitted signals and modulate and forward simulated echoes, or to generate fixed-mode radar signals through hardware such as FPGAs and vector signal sources, and complete simple countermeasure simulations with preset strategy libraries. However, such schemes still have many key defects in practical applications and cannot meet the training needs of modern radar countermeasures.

[0005] Therefore, existing radar countermeasure training still relies on the traditional paradigm of "manual configuration + fixed rules + hardware dependence," which cannot achieve intelligent and dynamic generation of scenarios, signals, and strategies; and is difficult to adapt to the complex and ever-changing electromagnetic countermeasure requirements. Summary of the Invention

[0006] In view of this, the purpose of this invention is to provide a radar countermeasure simulation training method and system based on a multimodal large model, which aims to solve the problem that existing technologies are difficult to adapt to the complex and ever-changing electromagnetic countermeasure requirements when conducting radar countermeasure training.

[0007] The embodiments of the present invention are implemented as follows: On one hand, this invention proposes a radar countermeasure simulation training method based on a multimodal large model, applied to a radar countermeasure simulation training platform based on a multimodal large model. The multimodal large model-based radar countermeasure simulation training platform includes interconnected large model intelligent interaction units, intelligent scene generation units, intelligent radar signal generation units, real-time simulation engine units, intelligent countermeasure strategy units, intelligent evaluation and feedback units, and a model library. The method includes: When the large model intelligent interaction unit receives training requests input by the trainer via natural language, such as voice or text, it parses the training requests accordingly, extracts the core elements, and completes the initialization settings of radar parameters and target parameters. Based on training requirements, the intelligent scene generation unit generates a complete electromagnetic environment scene, and the intelligent radar signal generation unit generates multi-mode radar signal parameters that conform to the tactical situation, synthesizes signal waveforms and simulates complex environmental effects, and outputs standardized scene and signal files. The real-time simulation engine unit is used to simulate radar startup and to simulate radar search and scanning beams to generate target echo signals. The intelligent countermeasure strategy unit perceives the countermeasure situation in real time, completes radar signal interception, parameter measurement, and threat identification, and generates adaptive countermeasure strategies through large model reasoning to simulate jamming transmission and countermeasure effects. The intelligent evaluation and feedback unit collects training operation logs, decision-making timelines, and adversarial effect data throughout the entire process. It then uses a large model to perform multi-dimensional quantitative scoring, tactical compliance checks, and causal reasoning for strategy rationality, generating training evaluation results in real time.

[0008] Furthermore, the aforementioned radar countermeasure simulation training method based on a multimodal large model includes the following steps: the intelligent scene generation unit generates a complete electromagnetic environment scene based on training requirements, and the intelligent radar signal generation unit generates multi-mode radar signal parameters that conform to the tactical situation; synthesizes signal waveforms and simulates complex environmental effects; and outputs standardized scenes and signal files. Based on training requirements, the intelligent scene generation unit calls terrain, meteorology, and electromagnetic propagation models to generate a complete electromagnetic environment scene including radar radiation source deployment, target flight path, and threat level assessment. The intelligent radar signal generation unit generates radar signal parameters for multiple systems and tactical scenarios based on large model inference, synthesizes signal waveforms and simulates complex environmental effects, and outputs standardized scenarios and signal files.

[0009] Furthermore, in the aforementioned radar countermeasures simulation training method based on a multimodal large model, the step of using a real-time simulation engine unit to simulate radar startup includes: The real-time simulation engine unit is used to simulate the working status of the transmitter, servo control subsystem, and anti-jamming subsystem of the actual radar to complete the radar azimuth calibration. After selecting the training mode, the trainees simulate the antenna's rotation and elevation until it is aligned with the target, thus completing the mode setting and simulating the radar entering a ready-to-operate state.

[0010] Furthermore, the aforementioned radar confrontation simulation training method based on a multimodal large model, after the steps of performing multi-dimensional quantitative scoring, tactical compliance checks, and causal reasoning on the rationality of strategies through the large model to generate training evaluation results in real time, also includes: The training data and evaluation results are fed back to the model library to fine-tune the large model for domain-specific purposes, enabling iterative training of the large model through reinforcement learning.

[0011] Furthermore, in the aforementioned radar adversarial simulation training method based on a multimodal large model, the radar parameters include at least the radar initial coordinates, azimuth orientation, antenna scanning boundary, detection range, and elevation angle; the target parameters include at least the target initial coordinates, elevation, initial velocity, flight trajectory, and ballistic coefficients. The steps of analyzing the training requirements, extracting core elements, and initializing the radar and target parameters include: Using a large model intelligent interaction unit to perform multi-granular semantic analysis on natural language training requirements, identify explicit tactical instructions and implicit combat backgrounds, and construct a fuzzy set of tactical intentions containing uncertainty. The fuzzy set includes undefined combat nodes and their spatiotemporal relationships. The fuzzy set of tactical intent is input into the embedded electromagnetic propagation and kinematic constraint model. Within the feasible solution space that satisfies the radar equation limit, antenna servo rate limit, and target maneuver overload limit in the electromagnetic propagation and kinematic constraint model, the fuzzy tactical intent is projected into a specific parameter candidate interval, and parameter combinations that do not conform to physical laws are eliminated. Based on the parameter candidate interval, a zero-sum game matrix for both red and blue sides is constructed. A large model is used to simulate the adversarial scenario for a preset number of rounds. The objective function is to maximize the training coverage and adversarial intensity. Adaptive optimization is performed within the parameter candidate interval to lock the optimal initial values ​​of radar parameters and target parameters. Based on the locked optimal initial value and combined with the clutter distribution characteristics of the current electromagnetic environment, the radar detection range and target RCS model are dynamically adjusted to complete the initialization settings of radar parameters and target parameters.

[0012] Furthermore, in the aforementioned radar countermeasure simulation training method based on a multimodal large model, the step of inputting the fuzzy set of tactical intentions into an embedded electromagnetic propagation and kinematic constraint model, and within the feasible solution space satisfying the radar equation limit, antenna servo rate limit, and target maneuver overload limit in the electromagnetic propagation and kinematic constraint model, projecting the fuzzy tactical intentions into specific parameter candidate intervals and eliminating parameter combinations that do not conform to physical laws includes: The fuzzy set of tactical intentions is converted into a fuzzy logic proposition containing spatiotemporal constraints and input into an embedded electromagnetic propagation and kinematic constraint model. The constraint model includes boundary conditions such as radar equation limits, antenna servo rate limits, and target maneuver overload limits. A bidirectional inference network between fuzzy logic propositions and physical constraint models is constructed using a large model. An initial parameter set is generated in the feasible solution space through Monte Carlo random sampling. The initial parameter set satisfies the physical laws of radar equation limits, antenna servo rate limits, and target maneuver overload limits. The initial parameter set is physically consistent, and parameter combinations that cause radar detection blind spots, sudden changes in target trajectory, or instability of the servo system are eliminated. The remaining parameters are then projected into specific parameter candidate intervals.

[0013] Furthermore, the aforementioned radar adversarial simulation training method based on a multimodal large model, wherein the steps of constructing a zero-sum game matrix for both red and blue sides based on parameter candidate intervals, simulating a preset round of adversarial deduction using a large model, and taking maximizing training coverage and adversarial intensity as the objective function, adaptively optimizing within the parameter candidate intervals to lock the optimal initial values ​​of radar parameters and target parameters include: Based on the parameter candidate interval, the strategy space of the red and blue sides is generated by the large model, and an extended zero-sum game tree containing incomplete information and mixed strategies is constructed. By combining Monte Carlo tree search with deep reinforcement learning algorithms, we simulated adversarial scenarios for a preset number of rounds and calculated the joint utility value of training coverage and adversarial intensity under each strategy combination. Based on the joint utility value, the Pareto front search is performed at the leaf nodes of the game tree using a multi-objective particle swarm optimization algorithm to lock the optimal initial values ​​of radar and target parameters that enable training coverage and adversarial intensity to reach Nash equilibrium.

[0014] On the other hand, this invention proposes a radar countermeasure simulation training system based on a multimodal large model, applied in a radar countermeasure simulation training platform based on a multimodal large model. The multimodal large model-based radar countermeasure simulation training platform includes interconnected large model intelligent interaction units, intelligent scene generation units, intelligent radar signal generation units, real-time simulation engine units, intelligent countermeasure strategy units, intelligent evaluation and feedback units, and a model library. The system includes: The distribution module is used to parse the training requirements when the large model intelligent interaction unit receives training requirements input by the trainer via natural language, such as voice or text, extract the core elements, and complete the initialization settings of radar parameters and target parameters. The generation module is used to generate complete electromagnetic environment scenarios based on training requirements, and generate multi-mode radar signal parameters that conform to tactical situations based on intelligent scene generation units and intelligent radar signal generation units. It synthesizes signal waveforms and simulates complex environmental effects, and outputs standardized scenarios and signal files. The simulation module is used to simulate radar startup using the real-time simulation engine unit, and to simulate radar search and scanning beams to generate target echo signals. The training module is used by the intelligent countermeasure strategy unit to perceive the countermeasure situation in real time, complete radar signal interception, parameter measurement, threat identification, generate adaptive countermeasure strategies through large model inference, and simulate jamming transmission and countermeasure effects. The evaluation module is used by the intelligent evaluation feedback unit to collect training operation logs, decision time series, and adversarial effect data throughout the process. It uses a large model to perform multi-dimensional quantitative scoring, tactical compliance checks, and causal reasoning on the rationality of strategies, and generates training evaluation results in real time.

[0015] In another aspect, embodiments of the present invention provide a readable storage medium on which a computer program is stored, which, when executed by a processor, implements the steps of the above-described method.

[0016] In another aspect, embodiments of the present invention 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 above.

[0017] This invention deeply integrates the natural language understanding, knowledge reasoning, and generative capabilities of Large Language Models (LLM) with radar countermeasure simulation technology by setting up a large-scale intelligent interaction unit, an intelligent scene generation unit, an intelligent radar signal generation unit, a real-time simulation engine unit, an intelligent countermeasure strategy unit, an intelligent evaluation and feedback unit, and a model library. This achieves natural language-driven scene construction, intelligent radar signal generation, adaptive countermeasure strategy reasoning, high-fidelity real-time simulation, and personalized intelligent evaluation. It can simulate the complex electromagnetic environment of a real battlefield, multi-system radar signals, and adaptive countermeasure behavior. It can also simulate the signal processing and servo control subsystems of actual radar. Through interconnection with radar display and control software and real radar equipment, it provides operators with a highly realistic radar countermeasure control environment. Based on the battlefield environment influencing factors (terrain, weather, electromagnetic propagation, etc.) determined by the training subjects, the solution dynamically generates simulated data that closely matches actual combat. By simulating radar echoes, jamming effects, and target ballistic trajectories, it precisely controls the echo strength, stability, signal characteristics, and countermeasure strategy changes, achieving highly realistic radar countermeasure simulation training. This solves the problem that existing technologies struggle to adapt to the complex and ever-changing electromagnetic countermeasures requirements when conducting radar countermeasures training.

[0018] In addition, the embodiments of the present invention also have at least the following beneficial effects: 1. Significantly enhances the intelligence and realism of training: By integrating the reasoning and generation capabilities of large models, it can reproduce the complex electromagnetic environment and adaptive adversarial behavior of real battlefields, solving the problems of single scenarios and rigid adversarial behavior in traditional solutions, making training closer to actual combat.

[0019] 2. Improve training efficiency and interactive experience: Natural language-driven scene construction and interaction reduce the time for building complex scenes from hours to minutes, and support real-time adjustments during training, greatly improving training efficiency.

[0020] 3. Achieve multi-target, high-fidelity adversarial simulation: It can simultaneously simulate multiple batches and different types of radars and targets, accurately simulate target angles, trajectories, radar signal characteristics and complex environmental effects, and solve the problems of difficult multi-target simulation and low fidelity in traditional solutions.

[0021] 4. Provide personalized training guidance and closed-loop optimization: Real-time quantitative intelligent assessment and natural language review can accurately locate the problems of trainees and provide targeted suggestions; closed-loop optimization of training data enables the system to continuously improve its capabilities. Attached Figure Description

[0022] Figure 1 This is an architecture diagram of the radar confrontation simulation training platform in the radar confrontation simulation training method based on a multimodal large model in the first embodiment of the present invention. Figure 2This is a flowchart of the radar adversarial simulation training method based on a multimodal large model in the first embodiment of the present invention; Figure 3 This is a structural block diagram of a radar countermeasure simulation training system based on a multimodal large model, as shown in the second embodiment of the present invention.

[0023] The following detailed description, in conjunction with the accompanying drawings, will further illustrate the present invention. Detailed Implementation

[0024] To facilitate understanding of the present invention, a more complete description will be given below with reference to the accompanying drawings. Several embodiments of the invention are illustrated in the drawings. However, the invention can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete.

[0025] 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.

[0026] 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.

[0027] To address the shortcomings of traditional radar countermeasure simulation training schemes, such as insufficient intelligence, limited scenarios, complex interactions, and hardware dependence, this invention proposes a radar countermeasure simulation training method based on a multimodal large model. This method uses pure software as its core and is compatible with existing radar hardware. It deeply integrates the natural language understanding, knowledge reasoning, and generative capabilities of large language models (LLM) with radar countermeasure simulation technology, enabling natural language-driven scenario construction, intelligent radar signal generation, adaptive countermeasure strategy reasoning, high-fidelity real-time simulation, and personalized intelligent evaluation.

[0028] This invention can simulate the complex electromagnetic environment, multi-system radar signals, and adaptive countermeasures of a real battlefield. It can also simulate the signal processing and servo control subsystems of a real radar system. Through interconnection with radar display and control software and real radar equipment, it provides operators with a highly realistic radar countermeasures control environment. Based on the battlefield environment influencing factors (terrain, weather, electromagnetic propagation, etc.) determined for the training subjects, it dynamically generates simulated data that closely matches actual combat. By simulating radar echoes, jamming effects, and target ballistic trajectories, it precisely controls the echo strength, stability, signal characteristics, and changes in countermeasure strategies, achieving highly realistic radar countermeasures simulation training.

[0029] Specifically, such as Figure 1 As shown, the core of the radar countermeasure simulation training platform in this embodiment of the invention is its software architecture design, which consists of core units such as a large-scale intelligent interaction unit, an intelligent scene generation unit, an intelligent radar signal generation unit, a real-time simulation engine unit, an intelligent countermeasure strategy unit, and an intelligent evaluation and feedback unit. These units work collaboratively to form a data closed-loop optimization. Among them: The large-scale intelligent interaction unit is the core intelligent interface of the system, realizing voice / text natural language input processing. It includes speech recognition optimized with industry terminology, text cleaning, intent recognition, knowledge retrieval enhanced generation (RAG), and large-scale model tactical reasoning functions. It can analyze the natural language training needs of trainers and also realize natural language question answering for training review. It is a bridge connecting users and other units.

[0030] Based on the training requirements of the large model analysis, the intelligent scene generation unit generates a complete electromagnetic environment scenario that includes battlefield terrain, weather conditions, electromagnetic propagation environment, radar radiation source deployment, and target movement trajectory. It realizes radiation source parameter generation, intelligent target flight path planning, dynamic threat level assessment, and outputs standardized scene files.

[0031] The intelligent radar signal generation unit is the system's "signal factory." Based on tactical parameters generated from a large model, it dynamically constructs highly realistic radar signal waveforms with multiple modes, supporting various radar modes such as pulse Doppler, continuous wave, phased array, and low probability of intercept. It simulates complex environmental effects such as multipath, clutter, and signal fading, and can be connected to software simulation modules or hardware signal sources to output analog signals.

[0032] The real-time simulation engine unit can achieve high-fidelity real-time simulation, including high-precision spatiotemporal calculation, electromagnetic propagation and radar equation calculation, radar signal processing link simulation, interference effect modeling, and weapon ballistic simulation functions. It simulates the working status of the transmitter, servo control, signal processing and other subsystems of the actual radar, and presents the confrontation situation and simulation effect in real time.

[0033] The intelligent countermeasure strategy unit is the system's "tactical advisor," which perceives the battlefield confrontation situation in real time and generates adaptive electronic countermeasure strategies through a large-scale model tactical reasoning chain. It supports various jamming styles such as noise suppression, deception jamming, and combined jamming, and achieves optimized allocation of jamming resources for multiple targets. At the same time, it combines reinforcement learning to achieve continuous evolution of strategies.

[0034] The intelligent evaluation and feedback unit is used to collect operational data, decision-making data, and adversarial effect data throughout the training process. Through large models, it performs quantitative analysis, tactical compliance checks, and causal inference, and automatically generates personalized evaluation reports, including scores, problem analysis, improvement suggestions, and classic case references. It also supports natural language debriefing Q&A.

[0035] The model library provides underlying support for each unit, including radar system knowledge, typical equipment parameters, electronic warfare tactics manuals, classic battle case library, fine-tuning data for large model domains, reinforcement learning models, and receives training data to achieve continuous optimization.

[0036] For example, the large-scale intelligent interaction unit can be based on current open-source models, such as Qwen-14B-Chat or Llama 2-13B-Chat, to achieve natural language understanding; the intelligent scene generation / intelligent radar signal generation / intelligent countermeasure strategy unit can be based on Llama 2-7B or Qwen-7B as a foundation, and the model obtained by fine-tuning with domain corpus can realize electromagnetic scene generation, radar signal generation, jamming strategy reasoning, situational awareness, etc.; the intelligent evaluation and feedback unit can be based on BERT-Large / RoBERTa, and realize operation scoring, tactical compliance checking, causal reasoning, and report generation through incremental pre-training of domain corpus, instruction fine-tuning and reinforcement learning iteration optimization.

[0037] The external system interface unit is used to interconnect with radar display and control software, real radar equipment, tactical data links, and guidance and control systems, and to complete command reception, data transmission, and status synchronization. It is compatible with existing radar hardware equipment and does not require replacement of existing resources.

[0038] The following will describe in detail, with reference to specific embodiments and accompanying drawings, how to meet the complex and ever-changing electromagnetic countermeasures requirements of current radar countermeasures training.

[0039] Example 1 Please see Figure 2The image shows a radar countermeasure simulation training method based on a multimodal large model according to the first embodiment of the present invention. It is applied to a radar countermeasure simulation training platform based on a multimodal large model. The radar countermeasure simulation training platform based on a multimodal large model includes a large model intelligent interaction unit, an intelligent scene generation unit, an intelligent radar signal generation unit, a real-time simulation engine unit, an intelligent countermeasure strategy unit, an intelligent evaluation feedback unit, and a model library that are interconnected. The method includes steps S10 to S13.

[0040] Step S10: When the large model intelligent interaction unit receives the training requirements input by the trainer through natural language via voice or text, it parses the training requirements accordingly, extracts the core elements, and completes the initialization settings of radar parameters and target parameters.

[0041] Trainees can input training requirements such as training subjects, confrontation scenarios, radar types, number of targets, and battlefield environment through natural language methods such as voice or text. The system also supports professional industry terminology related to radar confrontation, such as "Constructing a mountain defense scenario where two of our aircraft penetrate the defenses and confront three early warning radars and one fire control radar under rain and fog conditions."

[0042] The large-scale intelligent interactive unit analyzes training requirements, extracts core elements, and completes parameter initialization settings. Core elements include key information such as training subjects, adversarial scenario types, radar models and quantities, target types and quantities, battlefield environmental conditions, and training difficulty levels. Radar parameters include at least radar initial coordinates, azimuth orientation, antenna scanning boundaries, detection range, and elevation angle. Target parameters include at least target initial coordinates, elevation, initial velocity, flight trajectory, and ballistic coefficients. Parameters can be transmitted via guidance and control communication or manually fine-tuned.

[0043] In addition, in some optional embodiments of the present invention, the steps of analyzing the training requirements, extracting core elements, and initializing the radar parameters and target parameters include: Using a large model intelligent interaction unit to perform multi-granular semantic analysis on natural language training requirements, identify explicit tactical instructions and implicit combat backgrounds, and construct a fuzzy set of tactical intentions containing uncertainty. The fuzzy set includes undefined combat nodes and their spatiotemporal relationships. The fuzzy set of tactical intent is input into the embedded electromagnetic propagation and kinematic constraint model. Within the feasible solution space that satisfies the radar equation limit, antenna servo rate limit, and target maneuver overload limit in the electromagnetic propagation and kinematic constraint model, the fuzzy tactical intent is projected into a specific parameter candidate interval, and parameter combinations that do not conform to physical laws are eliminated. Based on the parameter candidate interval, a zero-sum game matrix for both red and blue sides is constructed. A large model is used to simulate the adversarial scenario for a preset number of rounds. The objective function is to maximize the training coverage and adversarial intensity. Adaptive optimization is performed within the parameter candidate interval to lock the optimal initial values ​​of radar parameters and target parameters. Based on the locked optimal initial value and combined with the clutter distribution characteristics of the current electromagnetic environment, the radar detection range and target RCS model are dynamically adjusted to complete the initialization settings of radar parameters and target parameters.

[0044] Among them, the large-scale intelligent interaction unit, as the intelligent interaction core of the radar confrontation simulation training platform, first receives and processes the natural language training requirements input by trainees in the form of voice or text. Through multi-granularity semantic analysis technology, it performs multi-level analysis of natural language text from vocabulary, syntax to contextual semantics. In this process, it accurately identifies explicit tactical instructions and implicit combat background. Explicit tactical instructions refer to the training requirements that are clearly stated in the training requirements and can be directly quantified, such as instructions that can be directly extracted, such as "constructing a mountain air defense scenario, 2 phased array radars, 4 penetrating targets, rain and fog environment". Implicit combat background refers to the combat conditions and tactical tendencies that are not directly stated in the requirements text but can be inferred from the context, such as inferring background information such as low target flight altitude, radar detection being greatly affected by clutter, and fast confrontation pace from "low-altitude penetration". After extracting explicit and implicit information, a fuzzy set of tactical intentions containing uncertainty is constructed based on the extracted information. This fuzzy set, expressed in fuzzy mathematics, can be denoted as: ,in This refers to operational nodes that are not fully defined, including node elements such as radar deployment locations, target appearance areas, confrontation triggering timing, and threat level ranges that do not have specific numerical values. The membership degree of the corresponding node is used to characterize the probability that the node belongs to a certain type of combat status. For the spatiotemporal relationships between combat nodes, including spatial distance constraints, temporal sequence, and node interaction logic, such as "radar deployment area A is 30km~50km away from target route area B" and "the target enters the detection range 10s~20s after the radar is turned on", the overall fuzzy set of tactical intentions can be compatible with fuzzy expressions and uncertain information in natural language requirements, providing a standardized semantic and logical foundation for the subsequent transformation of fuzzy tactical intentions into candidate intervals of radar and target parameters that conform to physical constraints.

[0045] Next, the constructed tactical intent fuzzy set is input into the system's embedded electromagnetic propagation and kinematic constraint model. This model, built into the radar countermeasures simulation training platform, is a set of core constraint rules used to ensure that all simulation parameters and combat processes strictly conform to physical laws and engineering realities. It consists of two parts: electromagnetic propagation constraints and kinematic constraints. The electromagnetic propagation constraints, based on the radar equations, describe the energy relationship between radar transmitted signals, spatial propagation, target reflection, and received echoes. The formula is: ; in, For radar receiving power, For transmission power, For the transmit antenna gain, For receiving antenna gain, For the operating wavelength, The target's radar cross-section. This refers to the distance between the radar and the target. To account for the overall system loss, this equation strictly limits the radar's maximum detection range, minimum detectable signal power, signal propagation loss, and other insurmountable physical boundaries, which determine whether the radar can detect targets normally under specific terrain, weather, and clutter environments. Kinematic constraints consist of two parts: radar antenna motion constraints and target motion constraints. Radar antenna motion constraints are essentially antenna servo rate limits, restricting the maximum angular velocity and maximum angular acceleration of the antenna's azimuth / elevation rotation to prevent scanning parameters from exceeding the mechanical structure's capabilities. Target motion constraints are essentially target maneuver overload limits, following the formula... ,in The target maximum allowable overload factor, Gravitational acceleration is used to limit the target's acceleration, turning rate, and climb / dive rate to ensure that the target trajectory conforms to the motion characteristics of entities such as aircraft and missiles. These three constraints together constitute the feasible solution space that all parameters must satisfy. Only parameters within this space are physically feasible. Subsequently, relying on the powerful reasoning and mapping capabilities of the multimodal large model, within the feasible solution space that simultaneously satisfies the above three physical constraints, the originally fuzzy and non-quantitative tactical intentions are transformed into quantifiable and executable specific parameter candidate intervals. The multimodal large model first performs bidirectional matching between the unstructured information such as fuzzy combat nodes and spatiotemporal correlations in the fuzzy set of tactical intentions and the electromagnetic propagation and kinematic constraint model, mapping fuzzy expressions such as "long-range detection," "high-speed penetration," and "complex electromagnetic environment" into calculable numerical intervals such as radar coordinates, antenna rotation speed, detection range, target speed, trajectory curvature, and RCS range. At the same time, the entire process is bounded by the limits of radar equations, antenna servo rate limits, and target maneuver overload limits, without exceeding the feasible solution space, and finally generating parameter candidate intervals. Furthermore, in some optional embodiments of the present invention, a large number of initial parameter sets can be randomly generated within the feasible solution space that satisfies all the above physical constraints by using the Monte Carlo random sampling method. Each initial parameter set contains preliminary values ​​of radar parameters (radar initial coordinates, azimuth orientation, etc.) and target parameters (target initial coordinates, initial velocity, etc.), and all values ​​strictly conform to the physical laws of radar detection and target motion.

[0046] Finally, a comprehensive physical consistency check is performed on the generated initial parameter set. Physical consistency check refers to checking the compliance of each item in the initial parameter set generated by Monte Carlo random sampling, and determining whether the parameter combination simultaneously satisfies the laws of electromagnetic propagation and kinematics. Specifically, this includes checking whether the radar parameters exceed the detection limit determined by the radar equation, whether the antenna rotation speed exceeds the maximum rate of the servo system, and whether the target acceleration and trajectory changes exceed the maneuver overload limit. The check process will directly eliminate illegal parameter combinations that cause the radar to be unable to detect, the target trajectory to change abruptly, the antenna to go out of control, or the signal propagation to violate the laws of physics, and only retain parameters that are physically valid, engineering feasible, and reasonable in actual combat.

[0047] Next, after obtaining the candidate parameter range, based on the typical tactical norms and operational rules of radar countermeasures, complete strategy spaces are generated for the red side (radar detection and defense) and the blue side (target penetration and jamming). The red side's strategy includes radar scanning mode selection, detection range adjustment, tracking beam switching, and activation of anti-jamming measures. The blue side's strategy includes target flight trajectory planning, maneuver overload control, jamming pattern selection, and jamming power adjustment. This leads to the construction of a zero-sum game matrix suitable for scenarios with incomplete information and mixed strategies. This zero-sum game matrix is ​​a two-dimensional payoff matrix, which can be represented as: ; The row represents the red side's strategy. The column represents the blue team's strategy. Matrix elements The red team's gain and the blue team's gain are It satisfies the zero-sum characteristic that an increase in one side's gains inevitably leads to a decrease in the other side's gains, accurately reflecting the mutual constraints and ebb and flow of the opposing forces in radar confrontation. Then, it executes a closed-loop confrontation simulation with a preset number of rounds. Each round of simulation fully simulates the entire process of radar power-on, signal transmission, target echo generation, signal interception, threat identification, jamming transmission, anti-jamming handling, and situational changes. In the simulation, it calculates the training coverage and confrontation intensity corresponding to the current parameter combination in real time. Training coverage measures the comprehensiveness of training elements, such as tactical scenarios, radar systems, target types, jamming patterns, and environmental conditions, that can be covered by the parameter combination. The specific calculation method involves first determining the full set of training elements supported by the system. Total number Identify the current parameter combination Activable effective training element set Statistical quantity Finally, according to the formula The calculation shows that, for example, if the total number of training elements in the system is 12 and the current parameters can cover 10 of them, then the training coverage is approximately 10 / 12 ≈ 0.83. The intensity of the confrontation is used to characterize the intensity of the red-blue confrontation. The specific calculation steps are to extract the target maneuver overload intensity. Radar detection and tracking pressure Interference pattern complexity The three core sub-indicators were each normalized to the [0,1] interval. The ratio of the target actual overload to the maximum permissible overload. This is the ratio of the target's velocity to the radar's maximum trackable velocity. This is the ratio of the number of currently enabled interference patterns to the total number of interference patterns supported by the system, calculated using the weighted formula: ; in , , These are the normalized weighting coefficients; With the core optimization objective of simultaneously maximizing training coverage and adversarial intensity, the objective function is constructed as follows: ; in , Using weighted coefficients, Monte Carlo tree search combined with deep reinforcement learning algorithm is used for adaptive optimization within the parameter candidate interval. Through multiple rounds of iteration, better parameter combinations are continuously screened. Then, Pareto front search is performed at the leaf nodes of the game tree using multi-objective particle swarm optimization algorithm to find the parameter combination that makes the training coverage and adversarial intensity reach a Nash equilibrium state. In this equilibrium state, neither the red nor the blue side can improve their own gains by unilaterally adjusting the parameters. The parameter combination reaches a stable optimal state. Finally, the optimal initial values ​​of radar parameters and target parameters that simultaneously meet physical constraints, tactical requirements, high training coverage and high adversarial intensity are accurately locked from all feasible parameters.

[0048] After locking in the optimal initial values ​​for radar and target parameters, the final parameters are not directly determined. Instead, the parameters are dynamically fine-tuned based on the clutter distribution characteristics of the current electromagnetic environment, ultimately completing the initialization settings for both radar and target parameters. The clutter distribution characteristics of the electromagnetic environment (such as clutter intensity, clutter type, and clutter distribution range) directly affect the radar's detection performance. For example, in a strong clutter environment, the radar detection range is suppressed; without adjustment, the radar may be unable to effectively acquire the target. The target RCS (radar cross-section) model determines the intensity of the target's reflection of the radar signal, directly affecting the radar's detection and tracking accuracy. The target RCS will vary under different clutter environments. Therefore, the system dynamically adjusts the radar's detection range based on the clutter distribution of the current electromagnetic environment (e.g., appropriately shortening the detection range in strong clutter environments to ensure detection accuracy; and appropriately extending the detection range in weak clutter environments to expand the detection range). At the same time, it adjusts the target RCS model parameters to match the target RCS performance with the current clutter environment, ensuring that the radar and target parameters are not only optimally set but also adaptable to the electromagnetic environment of the current training scenario. This further enhances the realism and tactical rationality of the training scenario, ultimately completing the initialization settings of the entire radar and target parameters.

[0049] Furthermore, in some optional embodiments of the present invention, the step of constructing a zero-sum game matrix for both red and blue sides based on the parameter candidate interval, simulating a preset round of adversarial deduction using a large model, with the objective function of maximizing training coverage and adversarial intensity, and adaptively optimizing within the parameter candidate interval to lock the optimal initial values ​​of radar parameters and target parameters includes: Based on the parameter candidate interval, the strategy space of the red and blue sides is generated by the large model, and an extended zero-sum game tree containing incomplete information and mixed strategies is constructed. By combining Monte Carlo tree search with deep reinforcement learning algorithms, we simulated adversarial scenarios for a preset number of rounds and calculated the joint utility value of training coverage and adversarial intensity under each strategy combination. Based on the joint utility value, the Pareto front search is performed at the leaf nodes of the game tree using a multi-objective particle swarm optimization algorithm to lock the optimal initial values ​​of radar and target parameters that enable training coverage and adversarial intensity to reach Nash equilibrium.

[0050] First, based on legal and compliant parameter candidate ranges, and according to radar countermeasures tactical manuals, equipment performance databases, battlefield constraint rules, and typical confrontation modes, complete strategy spaces for both the red and blue sides are automatically generated. The red side, acting as the radar defender and detector, has a strategy space containing a set of executable decisions such as radar scanning mode selection, detection range adjustment, antenna beam pointing control, activation of anti-jamming measures, and tracking threshold setting. The blue side, acting as the target penetration and jammer, has a strategy space containing a set of executable decisions such as target flight trajectory planning, selection of maneuver overload magnitude, jamming pattern switching, jamming power adjustment, and penetration timing control. After obtaining the strategy spaces of both sides, an extended zero-sum game tree containing incomplete information and mixed strategies is further constructed. This extended zero-sum game tree is a game model structured around time-series decisions, capable of fully realizing... The entire expression represents the dynamic confrontation process in which the red and blue sides make decisions alternately in sequence. Zero-sum means that an increase in the gains of one side will inevitably lead to an equal decrease in the gains of the other side, and the total gains are always zero. This is in line with the opposition and constraint relationship between detection and counter-detection, and interference and anti-interference in radar confrontation. Incomplete information means that the red and blue sides cannot fully obtain the other side's real-time parameters, strategy choices and tactical intentions during the confrontation process, which is closer to the non-transparent environment of the real battlefield. Mixed strategy means that the red and blue sides do not use a fixed strategy, but randomly select multiple strategy combinations according to a certain probability, so as to avoid the rigidity of the confrontation mode. The game tree uses the root node to represent the initial state of the confrontation, the intermediate layer nodes to represent the strategy selection branches at each step, and the leaf nodes to represent the final confrontation state after a set of strategy combinations is executed. It can completely depict the multi-round, multi-branch, and dynamically changing confrontation decision chain. After constructing the game tree, a Monte Carlo Tree Search (MCTS) algorithm combined with a Deep Reinforcement Learning (DRL) algorithm is used to simulate a complete adversarial exercise with a preset fixed number of rounds within the game tree structure. Each round of the exercise starts from the root node and proceeds along the policy branches, gradually selecting decisions, simulating radar signal propagation, target movement, jamming transmission, echo generation, threat identification, and presentation of adversarial effects, until reaching the leaf node and obtaining the adversarial result corresponding to that set of policies and parameters. Simultaneously, during the exercise, the training coverage and adversarial intensity corresponding to the combination of parameters and policies are calculated in real time. The training coverage and adversarial intensity are then fused to obtain the joint utility value corresponding to the combination of parameters and policies. , and The weights for coverage and intensity are used to comprehensively evaluate the training value of this set of parameters. After completing all preset rounds of simulation and obtaining the joint utility values ​​of all leaf nodes, the Multi-Objective Particle Swarm Optimization (MOPSO) algorithm is used to search for the Pareto front in all leaf nodes of the game tree. The Pareto front refers to a set of optimal solutions in which no solution can improve one metric without decreasing the other. That is, there is no room for improvement that can simultaneously improve training coverage and adversarial intensity. Finally, the parameter combination that can make training coverage and adversarial intensity reach a Nash equilibrium state is locked in the Pareto optimal solution. Nash equilibrium means that under the current parameter and strategy configuration, neither side can improve its own gain or reduce the opponent's gain by changing its own strategy or parameters alone. The adversarial state reaches a stable and optimal equilibrium state. At this time, the corresponding radar parameters and target parameters are the optimal initial values ​​that simultaneously satisfy physical constraints, tactical rationality, high training coverage, and high adversarial intensity.

[0051] Step S11: Based on training requirements, the intelligent scene generation unit generates a complete electromagnetic environment scene, and the intelligent radar signal generation unit generates multi-mode radar signal parameters that conform to the tactical situation, synthesizes signal waveforms and simulates complex environmental effects, and outputs standardized scene and signal files.

[0052] Specifically, the intelligent scene generation unit calls terrain, meteorological, and electromagnetic propagation models to generate a complete electromagnetic environment scene including radar radiation source deployment, target flight path, and threat level assessment; the intelligent radar signal generation unit, based on large model inference, generates multi-system radar signal parameters that conform to tactical scenarios, synthesizes signal waveforms and simulates complex environmental effects, and outputs standardized scenes and signal files.

[0053] Step S12: The real-time simulation engine unit is used to simulate the radar power-on and to simulate the radar search and scanning beam to generate target echo signals.

[0054] The real-time simulation engine unit simulates the working status of the transmitter, servo control subsystem, and anti-jamming subsystem of the actual radar, and completes the radar azimuth calibration. The trainees select the training mode (such as reconnaissance confrontation, fire correction confrontation, low-altitude penetration confrontation), and the system simulates the antenna rotation and elevation until it is aligned with the target azimuth, completes the working mode setting, and the radar enters the working ready state.

[0055] In step S13, the intelligent countermeasure strategy unit perceives the countermeasure situation in real time, completes radar signal interception, parameter measurement, and threat identification, and generates an adaptive countermeasure strategy through large model inference to simulate the jamming transmission and countermeasure effects.

[0056] After the radar is powered on, the real-time simulation engine unit simulates the radar's search and scanning beams to generate target echo signals. The intelligent countermeasure strategy unit perceives the countermeasure situation in real time, completes radar signal interception, parameter measurement, and threat identification, and generates adaptive countermeasure strategies through large-scale model inference, simulating jamming transmission and countermeasure effects, presenting a three-dimensional countermeasure situation in real time, and supporting trainees to adjust the scene, signal, or strategy in real time using natural language. For example, the large model can adopt a large language model (LLM) or be based on Qwen-7B, integrating a spatiotemporal situation encoder and a radar signal encoder, adopting a Transformer cross-modal fusion architecture, and fine-tuning and training with radar countermeasure tactical data commands. It can generate adaptive jamming patterns, jamming timing, resource allocation, and other countermeasure strategies online, completing the simulation of jamming transmission and countermeasure effects.

[0057] In step S14, the intelligent evaluation and feedback unit collects training operation logs, decision sequence, and adversarial effect data throughout the process. It then uses a large model to perform multi-dimensional quantitative scoring, tactical compliance checks, and causal reasoning on the rationality of strategies, generating training evaluation results in real time.

[0058] The intelligent evaluation and feedback unit collects training operation logs, decision-making timelines, and adversarial effect data throughout the process. It then uses a large model to perform multi-dimensional quantitative scoring, tactical compliance checks, and causal reasoning for strategy rationality, generating training evaluation results in real time to provide operational guidance for trainers. This model can be built on the BERT-Large architecture, using temporal coding and a multi-task learning structure. After fine-tuning with training specifications, tactical rules, and evaluation cases, it can achieve multi-dimensional quantitative scoring, tactical compliance checks, and causal reasoning for strategy rationality, and generate training evaluation results in natural language in real time.

[0059] In addition, in some optional embodiments of the present invention, after the step of performing multi-dimensional quantitative scoring, tactical compliance checks, and causal reasoning on the rationality of strategies through a large model to generate training evaluation results in real time, the method further includes: The training data and evaluation results are fed back to the model library to fine-tune the large model for domain-specific purposes, enabling iterative training of the large model through reinforcement learning.

[0060] After the training, trainees can ask questions about the training process using natural language. The system will then provide targeted answers, generate personalized assessment reports, and recommend specific training subjects.

[0061] In summary, the radar countermeasure simulation training method based on a multimodal large model in the above embodiments of the present invention, by setting up a large model intelligent interaction unit, an intelligent scene generation unit, an intelligent radar signal generation unit, a real-time simulation engine unit, an intelligent countermeasure strategy unit, an intelligent evaluation feedback unit, and a model library, deeply integrates the natural language understanding, knowledge reasoning, and generative capabilities of large language models (LLM) with radar countermeasure simulation technology. This achieves natural language-driven scene construction, intelligent radar signal generation, adaptive countermeasure strategy reasoning, high-fidelity real-time simulation, and personalized intelligent evaluation. It can simulate the complex electromagnetic environment of a real battlefield, multi-system radar signals, and adaptive countermeasure behavior. It can also simulate the signal processing, servo control, and transmitter subsystem functions of a real radar system. Through interconnection with radar display and control software and real radar equipment, it provides operators with a highly realistic radar countermeasure control environment. The solution dynamically generates realistic simulation data based on battlefield environmental factors (terrain, weather, electromagnetic propagation, etc.) determined for the training subjects. By simulating radar echoes, jamming effects, and target ballistic trajectories, it precisely controls echo strength, stability, signal characteristics, and the changing patterns of countermeasure strategies, achieving highly realistic radar countermeasure simulation training. This solves the problem in existing technologies that struggle to adapt to the complex and ever-changing demands of electromagnetic countermeasures during radar countermeasure training.

[0062] Example 2 Please see Figure 3 The image shows a radar countermeasure simulation training system based on a multimodal large model proposed in the second embodiment of the present invention. This system is applied to a radar countermeasure simulation training platform based on a multimodal large model. The multimodal large model-based radar countermeasure simulation training platform includes interconnected large model intelligent interaction units, intelligent scene generation units, intelligent radar signal generation units, real-time simulation engine units, intelligent countermeasure strategy units, intelligent evaluation and feedback units, and a model library. The system includes: The distribution module 100 is used to parse the training requirements when the large model intelligent interaction unit receives the training requirements input by the trainer in natural language via voice or text, extract the core elements, and complete the initialization settings of radar parameters and target parameters. The generation module 200 is used to generate a complete electromagnetic environment scene based on training requirements, and the intelligent scene generation unit generates radar signal parameters of multiple systems that conform to tactical situations based on training requirements, synthesize signal waveforms and simulate complex environmental effects, and output standardized scene and signal files. The simulation module 300 is used to simulate the radar power-on using the real-time simulation engine unit, and to simulate the radar searching and scanning beam to generate target echo signals using the real-time simulation engine unit. The training module 400 is used by the intelligent countermeasure strategy unit to perceive the countermeasure situation in real time, complete radar signal interception, parameter measurement, threat identification, generate adaptive countermeasure strategies through large model inference, and simulate jamming transmission and countermeasure effects. The evaluation module 500 is used by the intelligent evaluation feedback unit to collect training operation logs, decision time series, and adversarial effect data throughout the process. Through a large model, it performs multi-dimensional quantitative scoring, tactical compliance checks, and causal reasoning on the rationality of strategies, and generates training evaluation results in real time.

[0063] The functions or operation steps implemented by the above modules are largely the same as those in the above method embodiments, and will not be repeated here.

[0064] Example 3 In another aspect, the present invention provides a readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method described in Embodiment 1 above.

[0065] Example 4 In another aspect, the present invention provides an electronic device, the 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 Embodiment 1 above.

[0066] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0067] Those skilled in the art will understand that the logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequential list of executable instructions for implementing logical functions, and can be embodied in any computer-readable storage medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable storage medium" can mean any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0068] More specific examples (a non-exhaustive list) of computer-readable storage media include: electrical connections (electronic devices) having one or more wires, portable computer disk drives (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable storage media can even be paper or other suitable media on which the program can be printed, since the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.

[0069] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0070] 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.

[0071] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are 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 modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of this patent should be determined by the appended claims.

Claims

1. A radar adversarial simulation training method based on a multimodal large model, characterized in that, This method is applied to a radar countermeasure simulation training platform based on a multimodal large model. The platform includes interconnected intelligent interaction units for the large model, intelligent scene generation units, intelligent radar signal generation units, real-time simulation engine units, intelligent countermeasure strategy units, intelligent evaluation and feedback units, and a model library. The method includes: When the large model intelligent interaction unit receives training requests input by the trainer via natural language, such as voice or text, it parses the training requests accordingly, extracts the core elements, and completes the initialization settings of radar parameters and target parameters. Based on training requirements, the intelligent scene generation unit generates a complete electromagnetic environment scene, and the intelligent radar signal generation unit generates multi-mode radar signal parameters that conform to the tactical situation, synthesizes signal waveforms and simulates complex environmental effects, and outputs standardized scene and signal files. The real-time simulation engine unit is used to simulate radar startup and to simulate radar search and scanning beams to generate target echo signals. The intelligent countermeasure strategy unit perceives the countermeasure situation in real time, completes radar signal interception, parameter measurement, and threat identification, and generates adaptive countermeasure strategies through large model reasoning to simulate jamming transmission and countermeasure effects. The intelligent evaluation and feedback unit collects training operation logs, decision-making timelines, and adversarial effect data throughout the entire process. It then uses a large model to perform multi-dimensional quantitative scoring, tactical compliance checks, and causal reasoning for strategy rationality, generating training evaluation results in real time.

2. The radar adversarial simulation training method based on a multimodal large model according to claim 1, characterized in that, The steps of generating a complete electromagnetic environment scenario based on training requirements, and generating multi-mode radar signal parameters that conform to tactical situations based on intelligent radar signal generation unit, synthesizing signal waveforms and simulating complex environmental effects, and outputting standardized scenarios and signal files include: Based on training requirements, the intelligent scene generation unit calls terrain, meteorology, and electromagnetic propagation models to generate a complete electromagnetic environment scene including radar radiation source deployment, target flight path, and threat level assessment. The intelligent radar signal generation unit generates radar signal parameters for multiple systems and tactical scenarios based on large model inference, synthesizes signal waveforms and simulates complex environmental effects, and outputs standardized scenarios and signal files.

3. The radar adversarial simulation training method based on a multimodal large model according to claim 1, characterized in that, The steps for simulating radar startup using a real-time simulation engine unit include: The real-time simulation engine unit is used to simulate the working status of the transmitter, servo control subsystem, and anti-jamming subsystem of the actual radar to complete the radar azimuth calibration. After selecting the training mode, the trainees simulate the antenna's rotation and elevation until it is aligned with the target, thus completing the mode setting and simulating the radar entering a ready-to-operate state.

4. The radar adversarial simulation training method based on a multimodal large model according to claim 1, characterized in that, The steps of performing multi-dimensional quantitative scoring, tactical compliance checks, and causal reasoning on the rationality of strategies using a large model to generate training evaluation results in real time also include: The training data and evaluation results are fed back to the model library to fine-tune the large model for domain-specific purposes, enabling iterative training of the large model through reinforcement learning.

5. The radar adversarial simulation training method based on a multimodal large model according to claim 1, characterized in that, Radar parameters include at least radar initial coordinates, azimuth orientation, antenna scan boundaries, detection range, and elevation angle; target parameters include at least target initial coordinates, elevation, initial velocity, flight trajectory, and ballistic coefficients. The steps of analyzing the training requirements, extracting core elements, and initializing the radar and target parameters include: Using a large model intelligent interaction unit to perform multi-granular semantic analysis on natural language training requirements, identify explicit tactical instructions and implicit combat backgrounds, and construct a fuzzy set of tactical intentions containing uncertainty. The fuzzy set includes undefined combat nodes and their spatiotemporal relationships. The fuzzy set of tactical intent is input into the embedded electromagnetic propagation and kinematic constraint model. Within the feasible solution space that satisfies the radar equation limit, antenna servo rate limit, and target maneuver overload limit in the electromagnetic propagation and kinematic constraint model, the fuzzy tactical intent is projected into a specific parameter candidate interval, and parameter combinations that do not conform to physical laws are eliminated. Based on the parameter candidate interval, a zero-sum game matrix for both red and blue sides is constructed. A large model is used to simulate the adversarial scenario for a preset number of rounds. The objective function is to maximize the training coverage and adversarial intensity. Adaptive optimization is performed within the parameter candidate interval to lock the optimal initial values ​​of radar parameters and target parameters. Based on the locked optimal initial value and combined with the clutter distribution characteristics of the current electromagnetic environment, the radar detection range and target RCS model are dynamically adjusted to complete the initialization settings of radar parameters and target parameters.

6. The radar adversarial simulation training method based on a multimodal large model according to claim 5, characterized in that, The step of inputting the fuzzy set of tactical intent into the embedded electromagnetic propagation and kinematic constraint model, and projecting the fuzzy tactical intent into specific parameter candidate intervals within the feasible solution space that satisfies the radar equation limit, antenna servo rate limit, and target maneuver overload limit in the electromagnetic propagation and kinematic constraint model, and eliminating parameter combinations that do not conform to physical laws, includes: The fuzzy set of tactical intentions is converted into a fuzzy logic proposition containing spatiotemporal constraints and input into an embedded electromagnetic propagation and kinematic constraint model. The constraint model includes boundary conditions such as radar equation limits, antenna servo rate limits, and target maneuver overload limits. A bidirectional inference network between fuzzy logic propositions and physical constraint models is constructed using a large model. An initial parameter set is generated in the feasible solution space through Monte Carlo random sampling. The initial parameter set satisfies the physical laws of radar equation limits, antenna servo rate limits, and target maneuver overload limits. The initial parameter set is physically consistent, and parameter combinations that cause radar detection blind spots, sudden changes in target trajectory, or instability of the servo system are eliminated. The remaining parameters are then projected into specific parameter candidate intervals.

7. The radar adversarial simulation training method based on a multimodal large model according to claim 6, characterized in that, The steps of constructing a zero-sum game matrix for both red and blue sides based on parameter candidate intervals, simulating a preset round of adversarial simulation using a large model, and taking maximizing training coverage and adversarial intensity as the objective function, adaptively optimizing within the parameter candidate intervals to lock the optimal initial values ​​of radar and target parameters include: Based on the parameter candidate interval, the strategy space of the red and blue sides is generated by the large model, and an extended zero-sum game tree containing incomplete information and mixed strategies is constructed. By combining Monte Carlo tree search with deep reinforcement learning algorithms, we simulated adversarial scenarios for a preset number of rounds and calculated the joint utility value of training coverage and adversarial intensity under each strategy combination. Based on the joint utility value, the Pareto front search is performed at the leaf nodes of the game tree using a multi-objective particle swarm optimization algorithm to lock the optimal initial values ​​of radar and target parameters that enable training coverage and adversarial intensity to reach Nash equilibrium.

8. A radar adversarial simulation training system based on a multimodal large model, characterized in that, This system is applied to a radar countermeasure simulation training platform based on a multimodal large model. The platform includes interconnected intelligent interaction units for the large model, intelligent scene generation units, intelligent radar signal generation units, real-time simulation engine units, intelligent countermeasure strategy units, intelligent evaluation and feedback units, and a model library. The system comprises: The distribution module is used to parse the training requirements when the large model intelligent interaction unit receives training requirements input by the trainer via natural language, such as voice or text, extract the core elements, and complete the initialization settings of radar parameters and target parameters. The generation module is used to generate complete electromagnetic environment scenarios based on training requirements, and generate multi-mode radar signal parameters that conform to tactical situations based on intelligent scene generation units and intelligent radar signal generation units. It synthesizes signal waveforms and simulates complex environmental effects, and outputs standardized scenarios and signal files. The simulation module is used to simulate radar startup using the real-time simulation engine unit, and to simulate radar search and scanning beams to generate target echo signals. The training module is used by the intelligent countermeasure strategy unit to perceive the countermeasure situation in real time, complete radar signal interception, parameter measurement, threat identification, generate adaptive countermeasure strategies through large model inference, and simulate jamming transmission and countermeasure effects. The evaluation module is used by the intelligent evaluation feedback unit to collect training operation logs, decision time series, and adversarial effect data throughout the process. It uses a large model to perform multi-dimensional quantitative scoring, tactical compliance checks, and causal reasoning on the rationality of strategies, and generates training evaluation results in real time.

9. A readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the steps of the method as described in any one of claims 1 to 7.

10. An electronic device, characterized in that, It includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the program, implements the steps of the method as described in any one of claims 1 to 7.

Images