Target threat assessment method based on large language model

Abstract

The invention discloses a target threat assessment method based on a large language model, and the method comprises the following steps: carrying out the modeling of a confrontation environment, integrating multi-dimensional information, and determining a plurality of threat assessment targets; generating an evaluation index through the large language model; evaluating the generated evaluation indexes in combination with expert knowledge, if the evaluation indexes do not meet the requirements, further generating the evaluation indexes by the large language model, and if the evaluation indexes meet the requirements, generating an evaluation matrix by the large language model; performing evaluation by combining an evaluation matrix symmetrically formed by expert knowledge, if the evaluation matrix does not meet requirements, further generating the evaluation matrix by the large language model, and if the evaluation matrix meets the requirements, calculating the distance between each target and the positive and negative ideal solutions; and performing threat level sorting on the target according to the distance. By optimizing the decision process, more efficient threat assessment in a complex confrontation environment is realized, and decision support with higher reference value is provided for decision makers.

Description

Technical Field

[0001] The present invention belongs to the technical field of target threat assessment, and in particular relates to a target threat assessment method based on a large language model. Background Art

[0002] Threat assessment is the process of systematically analyzing and quantifying potential threats' capabilities, intentions, and potential harm. Its core is to provide a scientific basis for operational decision-making through the integration of multi-dimensional data. In modern simulated combat scenarios, accurate threat assessment is crucial for combat decision-making, directly impacting the scientific nature of command decisions and the effectiveness of combat.

[0003] Threat assessment is a multi-attribute decision-making problem that requires comprehensive consideration of the target's physical characteristics (such as speed and range) and tactical intent (such as heading angle and mission type). It includes both quantitative indicators (such as target range and speed) and qualitative indicators (such as stealth performance and mission type), which must be standardized through the Analytic Hierarchy Process (AHP) or membership functions. Threat assessments must be updated in real time to adapt to changes in the adversarial environment (such as target maneuvers and unexpected missions).

[0004] As adversarial environments become increasingly complex, traditional multi-attribute decision-making methods face challenges in coping with rapidly changing and multi-dimensional data, making them difficult to meet real-time adversarial demands. For example, data fusion complexity hinders the real-time integration and conflict resolution of heterogeneous data from multiple sources (radar, infrared, and electronic reconnaissance); incomplete adversarial environment information (such as target camouflage) leads to assessment bias; and threat assessment of large-scale target groups (such as drone swarms) requires low-latency computing, which relies on edge computing and distributed architectures.

[0005] Therefore, it is required to continuously optimize the threat assessment model to shift confrontation decisions from "experience-driven" to "data-driven" to improve response speed and accuracy. Summary of the Invention

[0006] Aiming at the limitations of existing threat assessment methods in data processing and real-time performance, the present invention proposes a threat assessment method combined with intelligent technology, aiming to improve the accuracy of assessment and real-time response capability.

[0007] To achieve the above objectives, the present application discloses a target threat assessment method based on a large language model, comprising the following steps:

[0008] Model the adversarial environment, integrate multi-dimensional information, and determine multiple threat assessment targets;

[0009] Generate evaluation metrics through large language models;

[0010] The generated evaluation indicators are evaluated in combination with expert knowledge. If they do not meet the requirements, the large language model will generate further evaluation indicators. If they meet the requirements, the large language model will generate an evaluation matrix.

[0011] The evaluation matrix formed by combining expert knowledge is evaluated. If it does not meet the requirements, the large language model will further generate an evaluation matrix. If it meets the requirements, the distance between each target and the positive and negative ideal solutions is calculated;

[0012] Rank targets by threat level based on distance.

[0013] Preferably, the modeling of the adversarial environment comprises:

[0014] Target information modeling: abstract various adversarial targets into objects with attributes such as firepower threat, mobility, and survivability. Each attribute is expressed in the IFS language to represent its credibility, unreliability, and uncertainty.

[0015] Situational information modeling: including topography, electromagnetic interference, and population density in the confrontation area as evaluation factors integrated into the model to construct environmental constraints and combat adaptability indicators;

[0016] Mission information modeling: extract instruction documents, tactical objectives, and confrontation intentions, and establish a "mission-objective" mapping relationship.

[0017] Preferably, the threat assessment target t j Characterized by multiple attributes, the attribute values ​​are expressed as triples <μ ij ,v ij ,π ij >, where π ij represents the target t j The degree of uncertainty about whether there is a threat under indicator j, μ ij is the membership degree, v ij Degree of non-affiliation.

[0018] Preferably, the multi-dimensional information includes:

[0019] Local static data: equipment performance parameter library, tactical regulations, historical threat assessment models, and simulation confrontation results;

[0020] Real-time dynamic data: information returned by the adversarial environment perception system, text summaries of drone reconnaissance images, and heterogeneous sensor data;

[0021] Open interconnected data: open source intelligence, online public opinion, equipment network information, knowledge base;

[0022] The large language model performs semantic analysis, feature extraction and context association on multi-dimensional information, and automatically generates a threat assessment indicator system that is adapted to the current adversarial environment.

[0023] Preferably, generating a construction decision matrix includes:

[0024] Data collection and standardization: Target-related data is collected from local databases, environmental perception systems, and internet intelligence sources. Structured data is normalized, and unstructured data is processed using a large language model for semantic extraction, standard mapping, and quantitative scoring.

[0025] Multi-dimensional indicator filling mechanism: Based on the evaluation indicator system, the large language model combines contextual information and target features to fill in the target's evaluation value under each indicator item by item;

[0026] Intelligent completion of missing values ​​and enhanced robustness: In cases of incomplete information or sensor failure, the model performs intelligent completion estimation based on historical databases, domain logic, and similar target data;

[0027] Adaptive adjustment of task situation: According to the task type, the weights of relevant indicators in the matrix are adjusted or partially reconstructed.

[0028] Preferably, calculating the distance from each target to the positive and negative ideal solutions includes:

[0029] It receives the evaluation matrix generated by the large model as input, determines the positive and negative ideal solutions, calculates the distance of each target to the ideal solution, and ranks the final threat level.

[0030] Preferably, determining the positive and negative ideal solutions includes:

[0031] Positive ideal solution A + and negative ideal solution A ― Represent the optimal and worst values ​​of all evaluation indicators respectively; under each indicator, the ideal solution A + and negative ideal solution A ― The values ​​of membership and non-membership are:

[0032]

[0033] in

[0034]

[0035] Where m is the number of attributes, n is the number of targets, is the membership degree of the positive ideal solution, is the negative ideal solution non-membership degree, is the membership degree of the negative ideal solution, is the non-membership degree of the negative ideal solution. The membership degree in the positive ideal solution is obtained from the maximum score among the membership values ​​of the indicator, and the non-membership degree is obtained from the maximum score among the non-membership values ​​of the indicator.

[0036] Calculate the Euclidean distance of each target to the positive and negative ideal solutions, including:

[0037] The distance from the target to the positive ideal solution

[0038]

[0039] The distance from the target to the negative ideal solution

[0040]

[0041] in

[0042]

[0043] is the weighted hesitation of the target, is the negative ideal solution hesitation, It is the ideal solution hesitation.

[0044] Preferably, the threat level ranking includes:

[0045] Based on the calculated positive and negative ideal solution distances, the relative proximity of each target is calculated. The relative proximity reflects the closeness of the target to the positive ideal solution. The calculation formula is as follows:

[0046]

[0047] The larger the relative proximity value is, the farther away it is from the negative ideal solution, that is, the closer it is to the positive ideal solution, and the higher the threat level of the target.

[0048] Finally, all targets are sorted according to their relative proximity, and the threat level of each target is obtained by sorting them from large to small according to their relative proximity.

[0049] By optimizing the decision-making process, the present invention achieves more efficient threat assessment in complex confrontation environments and provides decision makers with more valuable decision support. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 A threat assessment method framework enabled by large models;

[0051] Figure 2 Large model construction indicator system;

[0052] Figure 3 Large model generation evaluation indicators. DETAILED DESCRIPTION

[0053] The present invention will be further described below with reference to the accompanying drawings, but the present invention is not limited in any way. Any changes or substitutions made based on the teachings of the present invention fall within the scope of protection of the present invention.

[0054] refer to Figure 1 The present invention provides a target threat assessment method based on the empowerment of a large language model, which is suitable for modern adversarial environments with complex information, multi-source heterogeneity and high real-time requirements. In order to effectively realize the rapid judgment and ranking of the threat levels of various targets, this embodiment first models the adversarial environment, integrates multi-dimensional information, including target performance, tactical background, environmental factors and command intelligence, and constructs a semantic evaluation space that can be processed by intelligent models. On this basis, combined with the intuitive fuzzy set TOPSIS method, quantitative threat assessment of multiple targets is carried out, and the automatic generation and dynamic optimization of evaluation indicators are realized through the large language model to improve the objectivity and efficiency of the evaluation process.

[0055] The core problem addressed by this application is how to efficiently construct a rational evaluation index system and quickly and accurately rank the threat levels of multiple potential targets in the context of complex data sources, diverse threat types, and rapidly changing adversarial scenarios. This invention uses a large language model to automatically extract key attributes from adversarial scenario semantic data and construct evaluation indicators. This method also utilizes an expert interaction mechanism for fine-tuning, forming a closed-loop optimization process to ensure the practicality and interpretability of the evaluation system.

[0056] refer to Figure 1 ,This invention first models the information environment involved in target threat assessment, and transforms the actual adversarial environment data into a structured semantic representation that can be understood and processed by the model.,This mainly includes three aspects:

[0057] 1. Target Information Modeling: Various adversarial targets (such as drones and electronic jamming equipment) are abstracted into objects with attributes such as firepower threat, mobility, and survivability. Each attribute is expressed in the IFS language to represent its credibility, unreliability, and uncertainty.

[0058] 2. Situational information modeling: including topography, electromagnetic interference, population density in the confrontation area, etc., are integrated into the model as evaluation influencing factors to construct environmental constraints and combat adaptability indicators.

[0059] 3. Mission Information Modeling: Extract semantic information such as command documents, tactical objectives, and confrontation intentions, establish a "mission-objective" mapping relationship, and enable the model to understand the tactical importance and mission relevance of the target.

[0060] Through the above-mentioned environmental modeling, the present invention transforms the traditional process of threat judgment that relies on expert experience into a semantic reasoning and quantitative evaluation process driven by a large language model, realizing the key transformation of threat assessment tasks from "experience-driven" to "intelligent empowerment."

[0061] Explanation of symbols

[0062]

[0063]

[0064] Problem Definition:

[0065] In this application, under the framework of the intuitionistic fuzzy set TOPSIS method, the part that needs to rely on expert experience is replaced by a large language model. There is a solution set T consisting of n solutions. In a certain adversarial environment, it is necessary to sort the threat values ​​of the n targets and determine which targets are high-threat targets and need to be processed first.

[0066] Set each target t in the target set T j Characterized by multiple attributes, the attribute values ​​are expressed as triples <μ ij ,v ij ,π ij >, where π ij represents the target t j It is uncertain whether there is a threat level under indicator j.

[0067] The target threat level ranking process proposed in this application is divided into the following steps: 1. Indicator generation (indicator weight estimation); 2. Construction of a decision matrix; 3. Expert review and optimization; 4. Calculation of the distance from each target to the positive and negative ideal solutions; 5. Threat ranking output.

[0068] The following, combined with the embodiments of the present invention, further details the indicator weight estimation and decision matrix construction process in the target threat assessment method enabled by a large language model. Unless otherwise specified, the following steps can be automatically executed by a computer system and can be integrated into the threat assessment device proposed in this invention.

[0069] S1 indicator generation

[0070] This paper proposes a threat assessment indicator generation method that integrates the intelligent perception capabilities of a large language model (LLM) with human-computer interaction mechanisms. This method offers the advantages of high automation, real-time updates, scalability, and auditability. This method aims to address the challenges of traditional threat assessment, such as the heavy reliance on expert experience, delayed updates, and difficulty covering heterogeneous data from multiple sources.

[0071] The present invention supports multi-source data input, and the data sources include but are not limited to:

[0072] Local static data: such as weapon and equipment performance parameter database, tactical regulations, historical threat assessment models, simulation confrontation results, etc.

[0073] Real-time dynamic data: such as information returned by the adversarial environment perception system, text summaries of drone reconnaissance images, and heterogeneous sensor data;

[0074] Open interconnected data: such as open source intelligence, online public opinion, online equipment information, Wikipedia knowledge base, etc.

[0075] By connecting to local databases and internet resources, the large language model can perform semantic analysis, feature extraction, and contextual association on large amounts of structured and unstructured information in seconds, automatically generating a threat assessment indicator system adapted to the current adversarial environment. This process integrates natural language understanding, contextual logical reasoning, and military domain knowledge transfer capabilities to construct a multi-level, multi-dimensional indicator structure covering core areas such as combat capability, tactical intent, environmental adaptability, and equipment coordination.

[0076] After the initial generation of indicators, this invention introduces an expert review and feedback mechanism to dynamically revise the indicator system through a "human-machine collaboration + closed-loop optimization" approach. Experts can provide modification suggestions for the effectiveness, quantifiability, and practical applicability of indicators based on current mission requirements. The model then rapidly adjusts the logical chain based on this feedback, forming a high-quality, highly consistent final indicator system.

[0077] Furthermore, the indicator generation mechanism constructed by this invention possesses continuous learning capabilities. The model can accumulate expert knowledge and self-optimize over multiple rounds of interaction, further reducing reliance on human intervention in subsequent tasks. Compared to traditional methods, this invention not only significantly improves the automation and intelligence level of indicator system construction, but also achieves rapid response, precise adaptation, and knowledge accumulation in indicator design through a "model generation - expert correction - model iteration" mechanism.

[0078] This indicator generation module breaks the shackles of "manual setting of indicators + application of fixed templates" in traditional threat assessment, and truly realizes a data-driven, intelligently evolved human-machine collaborative assessment system, providing a solid foundation for threat perception and intelligent decision-making in dynamic confrontation environments.

[0079] At the same time, the present invention proposes an indicator weight estimation method that combines the knowledge reasoning ability of a large language model and an expert feedback mechanism. The specific form and process are similar to the next step.

[0080] S2 builds a decision matrix

[0081] After completing the construction of a threat assessment indicator system and estimating indicator weights, this paper further proposes an efficient and scalable method for constructing a decision matrix to support subsequent threat ranking, situation analysis, and decision-making. This method combines the semantic parsing capabilities of a large language model for multi-source information with an expert feedback mechanism to construct an accurate, adaptable, and interpretable threat assessment decision matrix in dynamic and uncertain environments.

[0082] The decision matrix constructed by the present invention takes the target entity as the evaluation object and the generated multidimensional indicator system as the evaluation dimension. The matrix unit represents the quantitative score or qualitative judgment result of the target under specific indicators. Specifically, it includes the following key steps:

[0083] (1) Data collection and standardization: Target-related data is collected from local databases, environmental perception systems, Internet intelligence sources, and other channels. Normalization is performed on structured data, and semantic extraction, standard mapping, and quantitative scoring are performed on unstructured data (such as text descriptions and image summaries) using a large language model to ensure that data from different sources and formats can be compared on a unified scale.

[0084] (2) Multi-dimensional indicator filling mechanism: Based on the indicator system generated above, the large language model combines contextual information and target characteristics to fill in the target's evaluation value under each indicator item by item. For example, under the "firepower intensity" indicator, the model can integrate factors such as the enemy's weapon range, ammunition type, and strike accuracy to comprehensively calculate the score.

[0085] (3) Intelligent completion of missing values ​​and enhanced robustness: In cases of incomplete information or sensor failure, the model can perform intelligent completion estimation based on historical databases, domain logic, and similar target data to improve the integrity and anti-interference ability of the matrix.

[0086] (4) Expert verification and collaborative revision mechanism: After the preliminary matrix is ​​constructed, experts are allowed to review and revise the key indicator values ​​through the human-computer collaborative interface. The model automatically records the modified content and performs reverse knowledge update to optimize the subsequent filling strategy.

[0087] (5) Task situation adaptability adjustment: According to the task type, the weights of relevant indicators in the matrix are adjusted or partially reconstructed to enhance the tactical adaptability and accuracy of the decision-making results.

[0088] The final matrix is ​​weighted by the indicator weights and can be directly used as the input basis for the multi-attribute decision-making algorithm. It also supports customized needs. By modifying the prompt words, the large model can be given the ability to take into account the decision maker's preferences. Compared with the traditional matrix construction method that relies on static templates and manual form filling, this invention realizes the integration of data-driven, semantic understanding and human-computer collaborative construction, further consolidating the core decision-making foundation of the intelligent threat assessment system.

[0089] S3 expert review optimization mechanism

[0090] To ensure the practical relevance and professional credibility of threat assessment results, this paper proposes an expert review optimization mechanism that integrates expert knowledge feedback with the semantic reasoning capabilities of a large language model. By incorporating a "human-machine collaboration + closed-loop optimization" technical approach, this mechanism effectively improves the accuracy, adaptability, and interpretability of the assessment system.

[0091] refer to Figure 2 and Figure 3 After the evaluation process is initially completed, the indicator system, weight configuration and decision matrix automatically generated by the model will be presented to domain experts through a visual interface. Experts can review key evaluation factors item by item based on tactical mission requirements and confrontation experience. For indicator ambiguity, weight deviation or sorting anomalies in the model output, experts can directly propose revisions in natural language or in the form of structured suggestions. The present invention uses a semantic understanding module to identify the intent and perform structured analysis of the expert input content, and makes local adjustments to the model reasoning path to achieve rapid self-correction of the evaluation system.

[0092] The expert review process of this invention is not just a one-time modification; it also enables continuous optimization and knowledge accumulation. Each round of expert revisions is automatically archived, forming an internal "expert experience mapping archive" within the model. This archive drives the initial generation and judgment logic of the evaluation model in subsequent tasks, gradually building an intelligent evaluation system with continuous learning capabilities. This mechanism effectively reduces the reliance on continuous expert intervention, enabling the model to adapt to diverse environments and tasks while maintaining a high level of consistency and practicality.

[0093] Compared with the traditional threat assessment method led by experts and executed in a templated manner, the expert review optimization mechanism proposed in this invention takes the large model as the core and builds a set of efficient linkage mechanisms of "automatic generation - expert correction - model iteration", which greatly enhances the collaborative intelligence, dynamic adaptability and result interpretability of the threat assessment system.

[0094] S4 calculates the distance from each target to the positive and negative ideal solutions

[0095] This paper uses a multi-objective threat assessment model based on the IFS-TOPSIS method, allowing users to calculate the threat level of a target using multiple evaluation metrics. The method receives an evaluation matrix generated by a large model as input and includes steps such as determining positive and negative ideal solutions, calculating the distance of each target from the ideal solution, and finally ranking the threat levels. The specific steps are as follows:

[0096] (1) Definition of positive and negative ideal solutions

[0097] According to the method definition, the positive ideal solution A + and negative ideal solution A ―Represent the optimal and worst values ​​of all evaluation indicators respectively. Under each indicator, the membership and non-membership values ​​of the positive and negative ideal solutions are:

[0098]

[0099] in

[0100]

[0101] The membership degree of the positive ideal solution is obtained from the maximum score among the membership values ​​of the indicator. In one embodiment, the membership degree is the maximum value among the membership values ​​of the indicator. The non-membership degree is obtained from the maximum score among the non-membership values ​​of the indicator. The non-membership degree is the maximum value among the non-membership values ​​of the indicator. The purpose of this is to find the most "perfect" solution. The opposite is true for the negative ideal solution.

[0102] (2) Calculate the Euclidean distance from each target to the positive and negative ideal solutions

[0103] Distance to positive ideal solution

[0104]

[0105] Distance to negative ideal solution

[0106]

[0107] in

[0108]

[0109] S5 Threat Ranking Output

[0110] Based on the calculated distances between the positive and negative ideal solutions, the relative proximity of each target is calculated. The relative proximity reflects the degree of closeness between the target and the positive ideal solution. The formula is as follows

[0111]

[0112] The larger the relative proximity value is, the farther it is from the negative ideal solution, that is, the closer it is to the positive ideal solution, and the higher the threat level of the target.

[0113] Finally, all targets are ranked according to their relative proximity, from highest to lowest, to obtain the threat level of each target. Targets with higher threat levels have greater threat values ​​and should be prioritized. Through this ranking mechanism, the present invention can accurately assess the threat level of targets and provide a scientific basis for decision support.

[0114] By optimizing the decision-making process, the present invention achieves more efficient threat assessment in complex confrontation environments and provides decision makers with more valuable decision support.

[0115] As used herein, the word "preferred" is intended to serve as an example, instance, or illustration. Any aspect or design described herein as "preferred" is not necessarily to be construed as advantageous over other aspects or designs. Rather, the use of the word "preferred" is intended to present concepts in a concrete manner. As used in this application, the term "or" is intended to mean an inclusive "or" rather than an exclusive "or." That is, unless otherwise specified or clear from the context, "X employs A or B" is intended to mean any of the naturally inclusive permutations. That is, if X employs A; X employs B; or X employs both A and B, then "X employs A or B" is satisfied in any of the foregoing examples.

[0116] Moreover, although the present disclosure has been shown and described with respect to one or implementation, those skilled in the art will think of equivalent variations and modifications based on reading and understanding of this specification and the accompanying drawings. The present disclosure includes all such modifications and variations and is limited only by the scope of the appended claims. In particular, with respect to the various functions performed by the above-mentioned components (such as elements, etc.), the terms used to describe such components are intended to correspond to any component (unless otherwise indicated) that performs the specified function of the component (such as it is functionally equivalent), even if structurally different from the disclosed structure that performs the function in the exemplary implementation of the present disclosure shown herein. In addition, although the specific features of the present disclosure have been disclosed with respect to only one of several implementations, such features can be combined with one or other features of other implementations that can be desired and advantageous for a given or specific application. Moreover, insofar as the terms "including", "having", "containing" or their variations are used in specific embodiments or claims, such terms are intended to be included in a manner similar to the term "comprising".

[0117] The functional units in the embodiments of the present invention may be integrated into a single processing module, or each unit may exist physically separately, or multiple or more units may be integrated into a single module. The aforementioned integrated module may be implemented in the form of hardware or in the form of a software functional module. If the integrated module is implemented in the form of a software functional module and sold or used as an independent product, it may also be stored in a computer-readable storage medium. The aforementioned storage medium may be a read-only memory, a magnetic disk, or an optical disk, etc. The aforementioned devices or systems may execute the storage method in the corresponding method embodiment.

[0118] In summary, the above embodiment is one implementation method of the present invention, but the implementation method of the present invention is not limited to the described embodiment. Any other changes, modifications, substitutions, combinations, and simplifications that deviate from the spirit and principles of the present invention should be equivalent replacement methods and are included in the scope of protection of the present invention.

Claims

1. A target threat assessment method based on a large language model, characterized in that: The following steps are involved: Model the adversarial environment, integrate multi-dimensional information, and determine multiple threat assessment targets; Generate evaluation metrics through large language models; The generated evaluation indicators are evaluated in combination with expert knowledge. If they do not meet the requirements, the large language model will generate further evaluation indicators. If they meet the requirements, the large language model will generate an evaluation matrix. The evaluation matrix formed by combining expert knowledge is evaluated. If it does not meet the requirements, the large language model will further generate an evaluation matrix. If it meets the requirements, the distance between each target and the positive and negative ideal solutions is calculated; Rank targets by threat level based on distance.

2. The target threat assessment method based on a large language model according to claim 1, characterized in that: The modeling of the adversarial environment includes: Target information modeling: abstract various adversarial targets into objects with attributes such as firepower threat, mobility, and survivability. Each attribute is expressed in the IFS language to represent its credibility, unreliability, and uncertainty. Situational information modeling: including topography, electromagnetic interference, and population density in the confrontation area as evaluation factors integrated into the model to construct environmental constraints and combat adaptability indicators; Mission information modeling: Extract instruction documents, tactical objectives, and confrontation intentions to establish a "mission-objective" mapping relationship.

3. The target threat assessment method based on a large language model according to claim 1, characterized in that: The threat assessment target t j The attribute value is expressed as a triplet <μ ij ,υ ij ,π ij >, where π ij represents the target t j The degree of uncertainty about whether there is a threat under indicator j, μ ij is the membership degree, v ij Degree of non-affiliation.

4. The target threat assessment method based on a large language model according to claim 1, characterized in that: The multi-dimensional information includes: Local static data: equipment performance parameter library, tactical regulations, historical threat assessment models, and simulation confrontation results; Real-time dynamic data: information returned by the adversarial environment perception system, text summaries of drone reconnaissance images, and heterogeneous sensor data; Open interconnected data: open source intelligence, online public opinion, equipment network information, knowledge base; The large language model performs semantic analysis, feature extraction and context association on multi-dimensional information, and automatically generates a threat assessment indicator system that is adapted to the current adversarial environment.

5. The target threat assessment method based on a large language model according to claim 1, characterized in that: Generating a build decision matrix involves: Data collection and standardization: Target-related data is collected from local databases, environmental perception systems, and internet intelligence sources. Structured data is normalized, and unstructured data is processed using a large language model for semantic extraction, standard mapping, and quantitative scoring. Multi-dimensional indicator filling mechanism: Based on the evaluation indicator system, the large language model combines contextual information and target features to fill in the target's evaluation value under each indicator item by item; Intelligent missing value completion and robustness enhancement: In cases of incomplete information or sensor failure, the model performs missing value estimation based on historical databases, domain logic, and similar target data; Adaptive adjustment of task situation: According to the task type, the weights of relevant indicators in the matrix are adjusted or partially reconstructed.

6. The target threat assessment method based on a large language model according to claim 1, characterized in that: Calculate the distance from each target to the positive and negative ideal solutions, including: It receives the evaluation matrix generated by the large model as input, determines the positive and negative ideal solutions, calculates the distance of each target to the ideal solution, and ranks the final threat level.

7. The target threat assessment method based on a large language model according to claim 6, characterized in that: Determining positive and negative ideal solutions involves: Positive ideal solution A + and negative ideal solution A ― Represent the optimal and worst values ​​of all evaluation indicators respectively; under each indicator, the ideal solution A + and negative ideal solution A ― The values ​​of membership and non-membership are: in Where m is the number of attributes, n is the number of targets, is the membership degree of the positive ideal solution, is the negative ideal solution non-membership degree, is the membership degree of the negative ideal solution, is the non-membership degree of the negative ideal solution. The membership degree in the positive ideal solution is determined by the maximum score among the membership values ​​of the indicator, and the non-membership degree is determined by the maximum score among the non-membership values ​​of the indicator. Calculate the Euclidean distance of each target to the positive and negative ideal solutions, including: The distance from the target to the positive ideal solution The distance from the target to the negative ideal solution in is the weighted hesitation of the target, is the negative ideal solution hesitation, It is the ideal solution hesitation.

8. The target threat assessment method based on a large language model according to claim 7, characterized in that: The threat level rankings include: Based on the calculated positive and negative ideal solution distances, the relative proximity of each target is calculated. The relative proximity reflects the closeness of the target to the positive ideal solution. The calculation formula is as follows: The larger the relative proximity value is, the farther away it is from the negative ideal solution, that is, the closer it is to the positive ideal solution, and the higher the threat level of the target. Finally, all targets are sorted according to their relative proximity, and the threat level of each target is obtained by sorting them from large to small according to their relative proximity.

Images