Intelligent power grid simulation method

Abstract

The invention provides an intelligent power grid simulation method, and belongs to the technical field of power grid simulation. A simulation requirement convertibility quantification method based on semantic integrity and uniqueness is adopted, an original natural language requirement is regarded as a description body composed of multi-dimensional semantic elements, and a requirement quality evaluation matrix is constructed through requirement element recognition, scene key item coverage analysis and ambiguity item confusion degree calculation; taking an integrity standard-reaching rate and a semantic confusion degree as core indexes, determining whether the demand can be directly converted into a structured demand description through a threshold judgment mechanism, and forming a verification closed loop of demand input in combination with user confirmation; for the user revised content, the quantization process will be re-performed to maintain the stability of the scene construction. According to the method, the original demand understanding process depending on artificial experience is converted into a computable, verifiable and traceable semantic evaluation process, and the normalization, stability and controllability of a simulation demand input stage are remarkably improved.

Description

Technical Field

[0001] This invention provides an intelligent power grid simulation method, belonging to the field of power grid simulation technology. Background Technology

[0002] As the application of large-scale model technology in power grid simulation deepens, the method of automatically generating simulation code based on natural language input is gradually emerging. However, the inherent ambiguity of natural language and the differences in the professional background and expression habits of simulation personnel can easily lead to incomplete input requirements, semantic ambiguity, and missing key scenario elements, resulting in significant deviations between the generated code and actual simulation requirements. Especially in complex simulation scenarios such as power grid transients, if core parameters such as fault simulation types and safety and stability verification indicators are not accurately expressed, it will directly affect the completeness of the simulation scenario construction and the credibility of the results. Existing methods generally lack mechanisms for quantitatively evaluating the semantic completeness, uniqueness, and convertibility of natural language simulation requirements, and do not fully consider personalized calibration and completion of requirements based on user profiles and historical interactions, making it difficult to promptly identify and correct defects at the requirement level. Therefore, it is necessary to add a requirement translation layer to the outer layer of the intelligent system to perform semantic quality assessment, structured transformation, and dynamic calibration based on user profiles to improve the accuracy and controllability of intelligent power grid simulation.

[0003] In existing intelligent simulation technology systems, semantic-driven simulation script generation and knowledge graph-supported data consistency verification have gradually become important means for multi-process simulation and digital twin applications. Related technologies improve the automation level of simulation processes and data quality control capabilities to some extent by constructing simulation script knowledge graphs, using semantic statistical models or CNN semantic analysis models to parse input information, and relying on graph traversal or entity relationship inference to complete script generation, data classification, and rationality verification. However, these methods mainly focus on the logical organization of simulation scripts or the structured verification of urban model data, relying on the accuracy of the input semantics itself. They lack a quantitative evaluation mechanism for the quality of natural language requirements and do not establish personalized requirement understanding models for different users' professional backgrounds, expression habits, and interaction histories. Therefore, it is difficult to identify problems such as missing expressions, semantic ambiguity, or omission of key elements in the requirement stage. With the increasing complexity of simulation scenarios and the widespread application of natural language input in simulation modeling, existing technologies still have significant shortcomings in terms of requirement parsing accuracy, semantic integrity assurance, and user intent inference, limiting the further application of intelligent simulation in high-reliability and high-precision scenarios. Summary of the Invention

[0004] This invention provides an intelligent power grid simulation method, which solves the following two technical problems:

[0005] 1. Existing simulation requirements based on large models lack a systematic semantic evaluation and convertibility judgment mechanism, making it difficult to promptly identify missing expressions, ambiguities, and omissions of scene elements. In power grid simulation, if key information such as fault type and safety and stability indicators are not clearly defined, it will lead to incomplete scenarios and biased generated results. Because the quality of requirements cannot be quantified, the uncertainty of natural language input is difficult to constrain, affecting the controllability and accuracy of simulation generation.

[0006] This invention proposes a method for quantifying the convertibility of simulation requirements based on semantic completeness and uniqueness. First, by identifying requirement elements in the original natural language and combining this with the coverage of requirement dimensions, the degree of expression of key scenario items, and the semantic confusion of ambiguous items, a requirement quality assessment matrix is ​​constructed to quantitatively measure the completeness and uniqueness of natural language requirements. For example, for a power grid transient simulation task, the system will detect whether the user input explicitly contains scenario elements such as fault simulation type or stability indicators, and assess whether its expression is executable. When the completeness compliance rate is higher than a preset threshold and the semantic confusion is lower than a threshold, the system automatically converts the requirement into a structured description and submits it for user confirmation to support direct execution of the simulation. If the user supplements or corrects the requirement, the system returns to recalculate the assessment matrix, achieving dynamic calibration of the requirement input.

[0007] 2. When natural language requirements fail to reach the conversion threshold due to insufficient expression, semantic bias, or missing elements, relying solely on semantic quantization is insufficient for effective calibration. Existing methods do not consider differences in users' professional backgrounds, expression styles, and historical usage habits, making it difficult to accurately infer their true intentions. Furthermore, they cannot proactively supplement key information when the quality of the requirements is insufficient, resulting in inadequate calibration capabilities and affecting the stability and reliability of simulation task inputs.

[0008] This invention proposes a simulation requirement calibration and completion method based on user profiles and thought transmission chains. By constructing a dynamic simulation user profile library, it extracts expression habit features from users' professional fields, technical titles, and past interaction data. Combined with temporal analysis and semantic source tracing techniques, it reconstructs the user's thought transmission chain from initial needs to final structured needs, identifying stable preferences and logical patterns in user expression. When the calibration process is triggered, it matches the corresponding expression habit rule library based on user profile tags and combines it with a simulation knowledge graph incorporating user features to automatically calibrate the original input into a structured requirement description that better reflects the user's true intentions, providing feedback to the user for confirmation. If the user approves, simulation generation begins; if the user continues to adjust, the semantic integrity assessment process is re-executed, achieving layer-by-layer requirement optimization. This method, through the synergistic fusion of semantic quantification and user profile inference, constructs an iterative and complete natural language requirement translation mechanism, significantly improving the credibility and accuracy of simulation task input.

[0009] The specific technical solution provided by this invention:

[0010] A method for simulating an intelligent power grid includes the following steps:

[0011] S1: Quantification of the transformability of simulation requirements based on semantic integrity and uniqueness

[0012] The original natural language requirements are viewed as a description composed of multi-dimensional semantic elements. A requirement quality assessment matrix is ​​constructed through requirement element identification, scenario key item coverage analysis, and ambiguity calculation. Completeness compliance rate and semantic ambiguity are used as core indicators. A threshold judgment mechanism determines whether a requirement can be directly transformed into a structured requirement description, and user confirmation forms a closed-loop verification of the requirement input. For user corrections, the quantification process is re-executed to maintain the stability of the scenario construction. Specifically, the following sub-steps are included:

[0013] S1.1 Constructing the semantic element space of simulation requirements and quantifying its completeness

[0014] Let the set of raw natural language input by the simulated user be denoted as ,in Indicates the first One natural language input sample, For natural language indexing, For the total number of natural languages ​​to be processed, for natural languages Using the existing semantic recognition model BERT to process natural language The input yields a set of requirements. ,in Indicates the first One requirement, For demand indexing, Given the total demand quantity, construct an integrity compliance rate assessment vector. ,in Indicates the first One dimension of demand, For demand indexes, there are a total of The semantic requirement dimensions include the business being simulated, the goal of the simulation, the constraints of the simulation, the topology, the fault type, the element type, the number of elements, and the position of the elements.

[0015] S1.2 Structured Requirement Transformation and Feedback Confirmation Based on Requirement Completeness and Confusion

[0016] In the process of constructing the semantic confusion matrix of ambiguous terms, the recognition model identifies the requirements. The process will output U possibilities, and the set of potential demands consisting of all possibilities is represented as follows: , Let the u-th potential demand identified as demand n be represented by the probability set denoted as . , This invention uses entropy to describe semantic confusion. , is represented as:

[0017] (1)

[0018] The final user demand semantic confusion metric The calculation is as follows:

[0019] (2)

[0020] Define alignment matrix , of which elements Indicates the first The requirement for the first The degree of explanation for each demand dimension, with a value range of [0,1].

[0021] For abstract semantic variables, calculate the cosine similarity of semantic vectors; for concrete parameter variables, detect the matching degree of named entities. The formula is as follows:

[0022] (3)

[0023] In the formula, It is a semantic embedding vector; For the first The semantic embedding vector of each requirement; For the first Semantic embedding vectors for each demand dimension; CosSim is the cosine similarity function; As an indicator function, when from the demand The demand dimension was successfully extracted. The value is 1 if it corresponds to a numerical value or entity, otherwise it is 0. For the requirement dimension type indicator variable, when the requirement dimension The value is 1 when it is an abstract semantic class variable, and 1 when it is a requirement dimension. The value is 0 when it is a concrete parameter class variable; Indicates from demand Extract the demand dimension from it.

[0024] For each dimension of demand It is necessary to determine whether it is included in the demand set. Covered by one or more requirements in the document. Define requirement dimensions. Effective coverage , is represented as:

[0025] (4)

[0026] Based on the business logic of the simulation scenario, a weight vector is introduced. Characterizes the importance of different variables, among which This represents the preset weight for the p-th requirement dimension. The final user requirement completeness index. The calculation is as follows:

[0027] (5)

[0028] In the formula, For the first Preset weights for each demand dimension. Ultimately... The larger the value, the more complete the user's description of their needs.

[0029] The integrity compliance rate and confusion threshold conditions are expressed as follows:

[0030] (6)

[0031] (7)

[0032] In the formula, The threshold for completeness compliance rate, This is the confusion threshold.

[0033] When both the integrity compliance rate and the confusion level meet the threshold conditions, the intelligent agent is directly simulated using a large model to convert natural language into structured requirements. Provide a description and feedback to the user for confirmation. Large-scale simulation of the intelligent agent's parameter model. Represented as:

[0034] (8)

[0035] The formula consists of two parts. It represents the inherent parameter set of a large model agent and is responsible for basic semantic understanding and structure generation; This represents an auxiliary parameter set derived from a domain knowledge graph, which improves the accuracy of the generated knowledge graph through external knowledge injection. A dynamic masking mechanism is introduced; when the knowledge graph is not invoked, masking operations are used to... It relies solely on inherent parameters for inference.

[0036] Structured requirements are described using a directed graph. , is represented as:

[0037] (9)

[0038] In the formula, It is a set of directed edges, representing the execution order and logical dependencies between modules. Indicates a node Pointing to node The directed edge, Indicates the first One node; It is a set of nodes, where I is the number of nodes, and each node A corresponding "structured module" includes module type and module parameters, represented as follows:

[0039] (10)

[0040] In the formula, Indicates the module type. Indicates module parameters.

[0041] If the user confirms that the calibration is correct, simulation generation will proceed; if the user makes corrections or additions, S1.1 will be repeated. If either of these steps fails to meet the threshold, the S2 calibration and completion process will be performed.

[0042] S2: Simulation requirement calibration and completion based on user profiles and thought processes

[0043] A dynamic user profile library is constructed at the requirement translation layer. Stable expression features are extracted from users' professional fields, technical titles, and historical interaction records. Through time-series analysis, the thought transmission chain from initial input to final confirmation is reconstructed to characterize individualized expression patterns. Based on this, a simulated domain knowledge graph integrating profile features is constructed, forming a rule base for expression habits corresponding to different profile tags. When the semantic quantification result does not meet the standard, the original requirement is semantically corrected and element-completed according to the user profile matching rule base. Combined with the knowledge graph, the automatic mapping from natural language to structured requirements is completed, generating candidate structured descriptions and providing feedback to the user for confirmation. After user approval, the simulation generation begins. If further modifications are required, the process returns to semantic evaluation, achieving a cyclical optimization of "evaluation-calibration-confirmation".

[0044] Specifically, it includes the following sub-steps:

[0045] S2.1 Constructing a Dynamic Simulation User Profile Database

[0046] In the requirements translation layer, a set of simulated users is introduced. ,in Indicates the first Users, Index for users, The number of users participating in the simulation configuration. (For the first...) users In the In this interaction, a user profile is constructed, comprising three core modules: professional domain profile. Technical title profile and historical interactive portraits Among them, historical interactive portraits Further includes a collection of the user's past natural language input records. and the final confirmed set of structured requirements. ,in Indicates the first In the first interaction, the... The user's number A sample of natural language input. The above information is then organized into a user profile vector. :

[0047] (11)

[0048] This serves as the foundation for subsequent extraction of individual user characteristics and demand calibration. A user profile database is formed by dynamically updating all user profiles. , is represented as:

[0049] (12)

[0050] S2.2 Extract user expression habits and construct a thought transmission chain model

[0051] Based on user profile database For users Historical interaction records are used for execution timing analysis and semantic tracing. Assume the user... The This interaction ranges from natural language input to the final structured output. The whole process , is represented as:

[0052] (13)

[0053] Analyzing sequences using long short-term memory networks or Transformer models that incorporate attention mechanisms. Evolutionary paths and modification locations of semantic fragments in Chinese, mining user expression preference feature sets. , is represented as:

[0054] (14)

[0055] In the formula, This refers to an LSTM or Transformer model that incorporates an attention mechanism. This represents the default preference vector, indicating the types of parameters that users often omit. This represents the focus bias vector, indicating the emphasis of the user's description; It represents the logical narrative sequence characteristics, characterizing the user's thought process in constructing the scenario.

[0056] S2.3 Constructing a simulation domain knowledge graph that integrates user profile features

[0057] In obtaining user profiles Characteristics of expression habits Subsequently, a simulation domain knowledge graph integrating user characteristics is constructed. , is represented as:

[0058] (15)

[0059] In the formula, This represents a personalized inference rule used to describe a user. A unique thought transmission chain: after each interaction between the user and the simulation model, the thought chain of that interaction is stored in the user's knowledge graph.

[0060] S2.4 Requirements calibration, completion, and confirmation process based on profile tags

[0061] This invention calculates the similarity of dynamic profiles and introduces a knowledge graph of similar users when the calibration process is triggered, thereby achieving personalized semantic correction and element completion for users' natural language needs.

[0062] First, calculate the user With other users in the user database In the Dynamic profile similarity of secondary interactions

[0063] Represented as:

[0064] (16)

[0065] Furthermore, for users In the Natural language input requirements in this interaction By combining its own historical habits with the rules of users in similar fields, a weighted and fused structured requirement description is generated. , is represented as:

[0066] (17)

[0067] In the formula, This represents a structured requirement generation function, which represents the generation of natural language. Input parameters are Structured requirements and natural language obtained from large-scale model simulation of intelligent agents The mapping relationship between them. Large-scale model simulation of intelligent agent parameters. Parameters in This represents the set of auxiliary parameters after integrating similar user profile knowledge graphs, expressed as:

[0068] (18)

[0069] In the formula, Indicates user Knowledge graph auxiliary parameter set, Indicates the relationship with the current user The index of users with the highest profile similarity. Symbols representing feature fusion.

[0070] The above process transforms the user's natural language into a structured requirement description, which is then fed back to the user for confirmation. If the user makes corrections or additions, step S1.1 is repeated.

[0071] The technical effects of this invention are as follows:

[0072] 1. This invention proposes a quantitative method for the convertibility of simulation requirements based on semantic integrity and uniqueness. The method first parses the original natural language requirements into a multi-dimensional semantic element space. Through requirement element identification and scenario key item coverage analysis, it systematically models core dimensions such as business objectives, constraints, topology, faults, and parameters. Then, it introduces an entropy-based semantic confusion metric to quantitatively characterize the ambiguity in the identification results, and combines requirement dimension coverage and weighting mechanisms to form a unified requirement integrity index. Based on this, a threshold judgment mechanism automatically determines whether the natural language requirements meet the conditions for direct conversion into structured simulation requirements, and generates structured module descriptions in the form of directed graphs, providing user confirmation and thus forming a closed-loop process of "quantitative evaluation—structured generation—user verification." This method transforms the requirement understanding process, which originally relied on human experience, into a computable, verifiable, and traceable semantic evaluation process, significantly improving the standardization, stability, and controllability of the simulation requirement input stage.

[0073] 2. This invention proposes a simulation requirement calibration and completion method based on user profiles and thought transmission chains. This method constructs a dynamic user profile library at the requirement translation layer, extracting stable expressive features from professional fields, technical titles, and historical interaction behaviors. Through temporal semantic analysis, it reconstructs the user's thought transmission chain from initial input to final confirmation, characterizing their long-term expressive preferences and logical habits. Based on this, user profile features are integrated into the simulation domain knowledge graph to form a personalized reasoning rule base. When the requirement quantification result is unsatisfactory, based on the user's own profile and the knowledge graph of similar users, semantic correction, element completion, and structured mapping are performed on the original natural language to generate candidate requirements and provide feedback to the user for confirmation until the conversion conditions are met. This achieves a cyclical optimization mechanism of "evaluation—calibration—re-evaluation—confirmation." Its advantage lies in its ability to adaptively correct for the expressive differences of different users, effectively reducing the cost of repeated manual communication, improving the conversion rate of low-quality natural language requirements, and enhancing the overall simulation configuration efficiency. Attached Figure Description

[0074] Figure 1 This is a flowchart of the present invention. Detailed Implementation

[0075] The specific technical solution of the present invention will be described in conjunction with the accompanying drawings.

[0076] This invention proposes an intelligent power grid simulation method, the flowchart of which is shown below. Figure 1 As shown, it includes:

[0077] S1. A method for quantifying the convertibility of simulation requirements based on semantic integrity and uniqueness.

[0078] S2 consists of two parts: simulation requirement calibration and completion method based on user profile and thought transmission chain.

[0079] Specifically:

[0080] S1: A method for quantifying the convertibility of simulation requirements based on semantic integrity and uniqueness.

[0081] To address the issues of incomplete demand expression, difficulty in quantifying and identifying semantic ambiguities, and uncontrollable demand convertibility in natural language-driven power grid simulation, this invention proposes a method for quantifying the convertibility of simulation demands based on semantic completeness and uniqueness. This method treats the original natural language demand as a description composed of multi-dimensional semantic elements. It constructs a demand quality assessment matrix through demand element identification, scenario key item coverage analysis, and ambiguity item confusion calculation, achieving a joint measurement of demand completeness, uniqueness, and convertibility. This invention uses completeness compliance rate and semantic confusion as core indicators, employing a threshold judgment mechanism to determine whether a demand can be directly converted into a structured demand description, and combining this with user confirmation to form a closed-loop verification of demand input. For user corrections, the quantification process is re-executed to maintain the stability of scenario construction. Based on semantic statistics and scenario constraint analysis, this method transforms traditional experience-based demand parsing into a quantifiable and verifiable semantic evaluation process, effectively improving the standardization and controllability of the simulation demand input stage. The following steps describe the convertibility quantification process.

[0082] S1.1 Constructing the semantic element space of simulation requirements and quantifying its completeness

[0083] Let the set of raw natural language input by the simulated user be denoted as ,in Indicates the first One natural language input sample, For natural language indexing, For the total number of natural languages ​​to be processed, for natural languages Using the existing semantic recognition model BERT to process natural language The input yields a set of requirements. ,in Indicates the first One requirement, For demand indexing, Given the total demand quantity, construct an integrity compliance rate assessment vector. ,in Indicates the first One dimension of demand, For demand indexes, there are a total of The semantic requirement dimensions include the business being simulated, the objectives being simulated, the constraints being simulated, the topology, the fault type, the element type, the number of elements, and the location of the elements.

[0084] S1.2 Structured Requirement Transformation and Feedback Confirmation Based on Requirement Completeness and Confusion

[0085] In the process of constructing the semantic confusion matrix of ambiguous terms, the recognition model identifies the requirements. The process will output U possibilities, and the set of potential demands consisting of all possibilities is represented as follows: , Let the u-th potential demand identified as demand n be represented by the probability set denoted as . , This invention uses entropy to describe semantic confusion. , is represented as:

[0086] (1)

[0087] The final user demand semantic confusion metric The calculation is as follows:

[0088] (2)

[0089] Define alignment matrix , of which elements Indicates the first The requirement for the first The degree of explanation for each demand dimension (value range [0,1]).

[0090] For abstract semantic variables (such as business or goal), calculate the cosine similarity of semantic vectors; for concrete parameter variables (such as position or number), detect the matching degree of named entities. The formula is as follows:

[0091] (3)

[0092] In the formula, It is a semantic embedding vector; For the first The semantic embedding vector of each requirement; For the first Semantic embedding vectors for each demand dimension; CosSim is the cosine similarity function; As an indicator function, when from the demand The demand dimension was successfully extracted. The value is 1 if it corresponds to a numerical value or entity, otherwise it is 0. For the requirement dimension type indicator variable, when the requirement dimension The value is 1 when it is an abstract semantic class variable, and 1 when it is a requirement dimension. The value is 0 when it is a concrete parameter class variable; Indicates from demand Extract the demand dimension from it.

[0093] For each dimension of demand It is necessary to determine whether it is included in the demand set. Covered by one or more requirements in the document. Define requirement dimensions. Effective coverage , is represented as:

[0094] (4)

[0095] Based on the business logic of the simulation scenario, a weight vector is introduced. Characterizes the importance of different variables, among which This represents the preset weight for the p-th requirement dimension. The final user requirement completeness index. The calculation is as follows:

[0096] (5)

[0097] In the formula, For the first Preset weights for each demand dimension. Ultimately... The larger the value, the more complete the user's description of their needs.

[0098] The integrity compliance rate and confusion threshold conditions are expressed as follows:

[0099] (6)

[0100] (7)

[0101] In the formula, The threshold for completeness compliance rate, This is the confusion threshold.

[0102] When both the integrity compliance rate and the confusion level meet the threshold conditions, the intelligent agent is directly simulated using a large model to convert natural language into structured requirements. Provide a description and feedback to the user for confirmation. Large-scale simulation of the intelligent agent's parameter model. Represented as:

[0103] (8)

[0104] The formula consists of two parts. It represents the inherent parameter set of a large model agent and is responsible for basic semantic understanding and structure generation; This method represents an auxiliary parameter set derived from a domain knowledge graph, which improves the accuracy of the generated knowledge graph through external knowledge injection. The method introduces a dynamic masking mechanism; when the knowledge graph is not invoked, masking operations are used to... It relies solely on inherent parameters for inference, reducing computational burden.

[0105] This invention uses a directed graph to describe structured requirements. , is represented as:

[0106] (9)

[0107] In the formula, It is a set of directed edges, representing the execution order and logical dependencies between modules. Indicates a node Pointing to node The directed edge, Indicates the first One node; It is a set of nodes, where I is the number of nodes, and each node A corresponding "structured module" includes module type and module parameters, represented as follows:

[0108] (10)

[0109] In the formula, This indicates the module type, such as topology generation, power flow calculation, short-circuit calculation, optimization solution, constraint verification, data reading, etc. This represents module parameters, such as node size, time scale, fault type / location, objective function, and constraint thresholds.

[0110] If the user confirms that the calibration is correct, simulation generation will proceed; if the user makes corrections or additions, S1.1 will be repeated. If either of these steps fails to meet the threshold, the S2 calibration and completion process will be performed.

[0111] S2: Simulation Requirement Calibration and Completion Method Based on User Profiles and Thought Transmission Chains

[0112] To address the problem that original natural language simulation requirements often fail to reach a convertible threshold due to insufficient expression, significant semantic deviations, or missing key elements, this invention proposes a simulation requirement calibration and completion method based on user profiles and thought processes. This method constructs a dynamic user profile library at the requirement translation layer, extracting stable expression features from users' professional fields, technical titles, and historical interaction records. It then reconstructs the thought process chain from initial input to final confirmation through temporal analysis to characterize individualized expression patterns. Based on this, a simulation domain knowledge graph integrating profile features is constructed, forming a rule base for expression habits corresponding to different profile tags. When semantic quantification results fail to meet the standards, this invention performs semantic correction and element completion on the original requirement based on the user profile matching rule base. It also automatically maps natural language to structured requirements using the knowledge graph, generating candidate structured descriptions and providing user confirmation. Once user approval is granted, the simulation begins; if further modifications are required, the process returns to semantic evaluation, achieving a cyclical optimization of "evaluation-calibration-confirmation." This method synergistically improves the convertibility of low-quality requirements through profile inference and semantic analysis. The simulation requirement calibration and completion process is described step-by-step below.

[0113] S2.1 Constructing a Dynamic Simulation User Profile Database

[0114] In the requirements translation layer, a set of simulated users is introduced. ,in Indicates the first Users, Index for users, The number of users participating in the simulation configuration. (For the first...) users In the In this interaction, a user profile is constructed, comprising three core modules: professional domain profile. Technical title profile and historical interactive portraits Among them, historical interactive portraits Further includes a collection of the user's past natural language input records. and the final confirmed set of structured requirements. ,in Indicates the first In the first interaction, the... The user's number A sample of natural language input. The above information is then organized into a user profile vector. :

[0115] (11)

[0116] This serves as the foundation for subsequent extraction of individual user characteristics and demand calibration. A user profile database is formed by dynamically updating all user profiles. , is represented as:

[0117] (12)

[0118] This profile library can be continuously expanded as new users are added and historical interactions accumulate, and is used to depict the differences in the expressive styles of different simulated personnel.

[0119] S2.2 Extract user expression habits and construct a thought transmission chain model

[0120] Based on user profile database For users Historical interaction records are used for execution timing analysis and semantic tracing. Assume the user... The This interaction ranges from natural language input to the final structured output. The whole process , is represented as:

[0121] (13)

[0122] Analyzing sequences using Long Short-Term Memory (LSTM) networks or Transformer models incorporating attention mechanisms. Evolutionary paths and modification locations of semantic fragments in Chinese, mining user expression preference feature sets. , is represented as:

[0123] (14)

[0124] In the formula, This refers to an LSTM or Transformer model that incorporates an attention mechanism. This represents the default preference vector, indicating the types of parameters that users often omit. This represents the focus bias vector, indicating the emphasis of the user's description; It represents the logical narrative sequence characteristics, characterizing the user's thought process in constructing the scenario.

[0125] S2.3 Constructing a simulation domain knowledge graph that integrates user profile features

[0126] In obtaining user profiles Characteristics of expression habits Subsequently, a simulation domain knowledge graph integrating user characteristics is constructed. , is represented as:

[0127] (15)

[0128] In the formula, This represents a personalized inference rule used to describe a user. With its unique thought transmission chain, after each interaction between the user and the simulation model, the thought chain of this interaction is stored in the user's knowledge graph.

[0129] S2.4 Requirements calibration, completion, and confirmation process based on profile tags

[0130] This invention calculates the similarity of dynamic profiles and introduces a knowledge graph of similar users when the calibration process is triggered, thereby achieving personalized semantic correction and element completion for users' natural language needs.

[0131] First, calculate the user With other users in the user database In the Dynamic profile similarity of secondary interactions

[0132] Represented as:

[0133] (16)

[0134] Furthermore, for users In the Natural language input requirements in this interaction By combining its own historical habits with the rules of users in similar fields, a weighted and fused structured requirement description is generated. , is represented as:

[0135] (17)

[0136] In the formula, This represents a structured requirement generation function, which represents the generation of natural language. Input parameters are Structured requirements and natural language obtained from large-scale model simulation of intelligent agents The mapping relationship between them. Large-scale model simulation of intelligent agent parameters. Parameters in This represents the set of auxiliary parameters after integrating similar user profile knowledge graphs, expressed as:

[0137] (18)

[0138] In the formula, Indicates user Knowledge graph auxiliary parameter set, Indicates the relationship with the current user The index of users with the highest profile similarity. Symbols representing feature fusion.

[0139] The above process transforms the user's natural language into a structured requirement description, which is then fed back to the user for confirmation. If the user makes corrections or additions, step S1.1 is repeated.

Claims

1. A method for simulating intelligent power grids, characterized in that, Includes the following processes: A simulation requirement transformation quantification method based on semantic integrity and uniqueness is adopted. The original natural language requirement is regarded as a description composed of multi-dimensional semantic elements. A requirement quality assessment matrix is ​​constructed through requirement element identification, scenario key item coverage analysis and ambiguity item confusion calculation. The integrity compliance rate and semantic confusion are used as core indicators. A threshold judgment mechanism is used to determine whether the requirement can be directly transformed into a structured requirement description. Combined with user confirmation, a requirement input verification closed loop is formed. For user corrections, the quantification process is re-executed to maintain the stability of scenario construction.

2. The intelligent power grid simulation method according to claim 1, characterized in that, The method for quantifying the convertibility of simulation requirements based on semantic integrity and uniqueness specifically includes the following sub-steps: S1.1 Constructing the semantic element space of simulation requirements and quantifying its completeness Let the set of raw natural language input by the simulated user be denoted as ,in Indicates the first One natural language input sample, For natural language indexing, For the total number of natural languages ​​to be processed, for natural languages Using the existing semantic recognition model BERT to process natural language The input yields a set of requirements. ,in Indicates the first One requirement, For demand indexing, Given the total demand quantity, construct an integrity compliance rate assessment vector. ,in Indicates the first One dimension of demand, For demand indexes, there are a total of The semantic requirement dimensions include the business being simulated, the objective being simulated, the constraints being simulated, the topology, the fault type, the element type, the number of elements, and the position of the elements. S1.2 Structured Requirement Transformation and Feedback Confirmation Based on Requirement Completeness and Confusion In the process of constructing the semantic confusion matrix of ambiguous terms, the recognition model identifies the requirements. The process will output U possibilities, and the set of potential demands consisting of all possibilities is represented as follows: , Let the u-th potential demand identified as demand n be represented by the probability set denoted as . , This invention uses entropy to describe semantic confusion. , is represented as: （1） The final user demand semantic confusion metric The calculation is as follows: （2） Define alignment matrix , of which elements Indicates the first The requirement for the first The degree of explanation for each demand dimension, with a value range of [0,1]; For abstract semantic variables, calculate the cosine similarity of semantic vectors; for concrete parameter variables, detect the matching degree of named entities; the formula is as follows: （3） In the formula, It is a semantic embedding vector; For the first The semantic embedding vector of each requirement; For the first Semantic embedding vectors for each demand dimension; CosSim is the cosine similarity function; As an indicator function, when from the demand The demand dimension was successfully extracted. The value is 1 if it corresponds to a numerical value or entity, otherwise it is 0. For the requirement dimension type indicator variable, when the requirement dimension The value is 1 when it is an abstract semantic class variable, and 1 when it is a requirement dimension. The value is 0 when it is a concrete parameter class variable; Indicates from demand Extract the demand dimension from; For each dimension of demand It is necessary to determine whether it is included in the demand set. Covered by any one or more requirements; defining requirement dimensions. Effective coverage , is represented as: （4） Based on the business logic of the simulation scenario, a weight vector is introduced. Characterizes the importance of different variables, among which The preset weight for the p-th requirement dimension; the final user requirement completeness index. The calculation is as follows: （5） In the formula, For the first Preset weights for each demand dimension; ultimately The larger the value, the more complete the user's description of their needs; The integrity compliance rate and confusion threshold conditions are expressed as follows: （6） （7） In the formula, The threshold for completeness compliance rate, The confusion threshold; When both the integrity compliance rate and the confusion level meet the threshold conditions, the intelligent agent is directly simulated using a large model to convert natural language into structured requirements. Provide a description and feedback to the user for confirmation; large-scale simulation of the intelligent agent's parameter model. Represented as: （8） The formula consists of two parts. It represents the inherent parameter set of a large model agent and is responsible for basic semantic understanding and structure generation; This represents an auxiliary parameter set derived from a domain knowledge graph, which improves the accuracy of the generated knowledge graph through external knowledge injection. A dynamic masking mechanism is introduced; when the knowledge graph is not invoked, masking operations are used to... Inference relies solely on inherent parameters; Structured requirements are described using a directed graph. , is represented as: （9） In the formula, It is a set of directed edges, representing the execution order and logical dependencies between modules. Indicates a node Pointing to node The directed edge, Indicates the first One node; It is a set of nodes, where I is the number of nodes, and each node A corresponding "structured module" includes module type and module parameters, represented as follows: （10） In the formula, Indicates the module type. Indicates module parameters; If the user confirms that the calibration is correct, simulation generation will proceed; if the user makes corrections or additions, S1.1 will be repeated; if either of these conditions fails to meet the threshold, the calibration and completion process will be performed.

3. The intelligent power grid simulation method according to claim 2, characterized in that, The calibration and completion process employs a simulation-based requirement calibration and completion method based on user profiles and thought processes. A dynamic user profile library is constructed at the requirement translation layer, extracting stable expression features from the user's professional field, technical title, and historical interaction records. Through temporal analysis, the thought process chain from initial input to final confirmation is reconstructed to characterize individualized expression patterns. Based on this, a simulation domain knowledge graph integrating profile features is constructed, forming a rule base for expression habits corresponding to different profile tags. When the semantic quantification result is unsatisfactory, the original requirement is semantically corrected and element-completed according to the user profile matching rule base. Combined with the knowledge graph, automatic mapping from natural language to structured requirements is completed, generating candidate structured descriptions and providing feedback to the user for confirmation. After user approval, the process enters simulation generation; if further modifications are needed, it returns to semantic evaluation, achieving a cyclical optimization of "evaluation—calibration—confirmation." 4. The intelligent power grid simulation method according to claim 3, characterized in that, The simulation requirement calibration and completion method based on user profiles and thought transmission chains specifically includes the following sub-steps: S2.1 Constructing a Dynamic Simulation User Profile Database In the requirements translation layer, a set of simulated users is introduced. ,in Indicates the first Users, Index for users, The number of users participating in the simulation configuration; for the first users In the In this interaction, a user profile is constructed, comprising three core modules: professional domain profile. Technical title profile and historical interactive portraits ; Among them, historical interactive portraits Further includes a collection of the user's past natural language input records. and the final confirmed set of structured requirements. ,in Indicates the first In the first interaction, the... The user's number A sample of natural language input; organize the above information into a user profile vector. : （11） This serves as the foundation for subsequent extraction of individual user characteristics and demand calibration; by dynamically updating all user profiles, a user profile database is formed. , is represented as: （12） S2.2 Extract user expression habits and construct a thought transmission chain model Based on user profile database For users Historical interaction records are used for execution timing analysis and semantic tracing; assuming the user... The This interaction ranges from natural language input to the final structured output. The whole process , is represented as: （13） Analyzing sequences using long short-term memory networks or Transformer models that incorporate attention mechanisms. Evolutionary paths and modification locations of semantic fragments in Chinese, mining user expression preference feature sets. , is represented as: （14） In the formula, This refers to an LSTM or Transformer model that incorporates an attention mechanism. This represents the default preference vector, indicating the types of parameters that users often omit. This represents the focus bias vector, indicating the emphasis of the user's description; It represents the temporal characteristics of logical narration and characterizes the thought process of the user in constructing the scenario; S2.3 Constructing a simulation domain knowledge graph that integrates user profile features In obtaining user profiles Characteristics of expression habits Subsequently, a simulation domain knowledge graph integrating user characteristics is constructed. , is represented as: （15） In the formula, This represents a personalized inference rule used to describe a user. A unique thought transmission chain: after each interaction between the user and the simulation model, the thought chain of this interaction is stored in the user's knowledge graph; S2.4 Requirements calibration, completion, and confirmation process based on profile tags When the calibration process is triggered, the similarity of the dynamic profile is calculated and the knowledge graph of similar users is introduced to realize personalized semantic correction and element completion for the user's natural language needs. First, calculate the user With other users in the user database In the Dynamic profile similarity of secondary interactions Represented as: （16） Furthermore, for users In the Natural language input requirements in this interaction By combining its own historical habits with the rules of users in similar fields, a weighted and fused structured requirement description is generated. , is represented as: （17） In the formula, This represents a structured requirement generation function, which represents the generation of natural language. Input parameters are Structured requirements and natural language obtained from large-scale model simulation of intelligent agents Mapping relationships between them; large-scale model simulation agent parameters Parameters in This represents the set of auxiliary parameters after integrating similar user profile knowledge graphs, expressed as: （18） In the formula, Indicates user Knowledge graph auxiliary parameter set, Indicates the relationship with the current user The index of users with the highest profile similarity. Indicates feature fusion symbols; Transform the user's natural language into a structured requirement description and provide feedback to the user for confirmation; if the user makes corrections or additions, then repeat step S1.1.

Images