Natural language-oriented fuel performance analysis method

Abstract

The invention discloses a natural language-oriented fuel performance analysis method, which comprises the following steps of: automatically converting natural language description input by a user into a standardized input card required by a fuel performance analysis program by integrating a natural language processing technology, fuel performance analysis professional knowledge and a fuel performance analysis program framework; the user use threshold is obviously reduced; and the analysis efficiency is improved. Comprising the following steps: 1, constructing a multi-level semantic library in the field of fuel performance analysis; 2, using a pre-training language model professional corpus and a multi-level semantic library; 3, analyzing the natural language input of the user by using the learned pre-training language model, and extracting key parameters and calculation requirements; 4, generating a structured input card in combination with context semantics, fuel performance domain knowledge and a source code of a fuel performance analysis program; 5, verifying the logic integrity of the input card, feeding back and correcting; 6, automatically submitting the final input card to a fuel performance analysis program to execute calculation; and 7, analyzing a calculation result, and generating a visual result and a natural language analysis report.

Description

Technical Field

[0001] The present invention belongs to the technical field of nuclear fuel performance analysis, and particularly relates to a fuel performance analysis method based on natural language. Background Art

[0002] When using existing fuel performance analysis programs (such as BISON, FROBA, FRAPCON, FRAPTRAN, BEEs, etc.), users need to manually write complex input cards and extract data from the result files to form technical reports, requiring users to be familiar with the usage methods, source codes, parameter definitions, and physical models of the programs (and underlying dependent platforms if any). The following problems exist in this process:

[0003] 1. Long learning cycle: Personnel with fuel performance analysis requirements need to learn the usage methods of specific fuel performance analysis programs (and underlying platforms if any) on the basis of being familiar with the fuel performance analysis steps. The cycle ranges from several months to a year, featuring a long learning cycle and high difficulty. When converting the fuel performance analysis tool from Program A to Program B, relevant personnel need to learn the usage method of Program B again, which takes a long time.

[0004] 2. High professional threshold: Non-professional personnel are difficult to accurately describe the calculation requirements, and it is easy to cause calculation failures due to omission of input card variable values or incorrect variable input.

[0005] 3. Low efficiency: Preparing input cards based on text editors, EXCEL, or visual interfaces is rather abstract for users, increasing the preparation time before calculation, especially when there is a large amount of repetitive work in multi-condition comparative analysis.

[0006] 4. Poor program compatibility: Different versions or types of programs have significant differences in the requirements for input card formats, and users need to repeatedly adjust, which takes a long time.

[0007] 5. Long post-processing time: Fuel performance analysis programs generally output a large amount of calculation data (such as calculation history data, result data). It takes a huge amount of time for users to extract key parameters from the massive results, and generally, unit conversion, graphing, etc. need to be carried out, further increasing the calculation time.

[0008] Currently, some studies attempt to simplify the input through graphical interfaces or standardized input cards, but still require users to be familiar with the usage methods of specific programs and fill in parameters item by item, without solving the mapping problem between natural language and structured data, nor realizing adaptive generation in combination with the code logic of the target program. Therefore, there is an urgent need for an intelligent system that can directly understand the user's intention and generate input cards with strong compatibility.

[0009] In view of the above background, this patent proposes a method for fuel performance analysis oriented to natural language, which uses natural language processing technology to convert the descriptive requirements input by users into standardized input cards required by the fuel performance analysis system, achieving a seamless connection from natural language to fuel performance analysis. Summary of the Invention

[0010] To overcome the above defects, the object of the present invention is to provide a method for fuel performance analysis oriented to natural language, the core of which is: constructing a knowledge graph in the field of fuel performance analysis to achieve precise association between natural language and professional parameters; parsing the source code of the target program, extracting input rules and integrating them with the semantic library to ensure that the generated input cards conform to the program logic; introducing a multi-level verification mechanism to intercept potential conflicts and support dynamic correction.

[0011] To achieve the above object, the present invention adopts the following technical solutions:

[0012] A method for fuel performance analysis oriented to natural language uses natural language processing technology to convert the descriptive requirements input by users into standardized input cards required by the fuel performance analysis system program, achieving a seamless connection from natural language to fuel performance analysis;

[0013] The method includes the following steps:

[0014] Step 1: Construct a multi-level semantic library in the field of fuel performance analysis, including a basic parameter library (such as geometric dimensions, material properties), a working condition rule library (such as boundary conditions, power distribution), and a logical relationship library (such as parameter dependency relationships); at the same time, parse the source code and configuration files of the target fuel performance analysis program, and extract the syntax rules, logical constraints, and physical model dependency relationships of the input parameters through static code analysis and dynamic parameter tracking to construct a fuel performance analysis program logic library;

[0015] Step 2: Input data to enable the pre-trained language model to learn the professionally organized corpus in the field of fuel performance analysis and the multi-level semantic library constructed in Step 1;

[0016] Step 3: Receive the natural language text or speech input by the user, perform semantic parsing through a pre-trained language model (such as BERT, ChatGPT, DeepSeek, etc.), and identify key entities (such as fuel type, geometric structure), operation instructions (such as steady-state analysis, transient simulation), and constraint conditions (such as temperature range, system pressure);

[0017] If there are problems with polysemy or ambiguous expressions in the user input, resulting in unsuccessful semantic parsing, the interaction requests the user to further describe and perform semantic parsing again until the parsing is successful and proceed to Step 4;

[0018] Step 4: Based on the semantic parsing results, integrate the fuel performance analysis program logic library and the multi-level semantic library, match the terms in the fuel performance analysis program logic library, and generate a preliminary structured input card;

[0019] Step 5: Conduct multi-level logical verification on the generated input card, including parameter integrity check, unit consistency check, conflict condition detection, and fuel performance analysis program compatibility check. If there are errors or omissions in the logical verification results, request the user to supplement or correct through interactive prompts, and return to Step 4 until the final input card is generated;

[0020] Step 6: Automatically submit the final input card to the fuel performance analysis program (such as BISON, FROBA, FRAPCON, FRAPTRAN, BEEs, etc.) for calculation, and monitor the calculation status in real time;

[0021] Step 7: Analyze the calculation results, generate a visual report (such as temperature field distribution, stress nephogram) and a summary of key indicators, and feedback to the user in the form of a report file or natural language.

[0022] The present invention converts the descriptive requirements input by the user into the standardized input card required by the fuel performance analysis system program through natural language processing technology, realizing the seamless connection from natural language to fuel performance analysis. Compared with the prior art, the present invention has the following advantages:

[0023] 1. Natural language interaction reduces the threshold for users to use the fuel performance analysis system program, enabling users to perform fuel performance analysis calculations without fully understanding the code framework and input card format;

[0024] 2. The efficiency of input card generation is improved. The traditional method requires manual writing of input card parameters item by item, which is time-consuming and error-prone. The present invention realizes the automatic generation of input cards through a pre-trained language model and a rule engine, greatly shortening the calculation preparation time, especially suitable for multi-condition comparison analysis scenarios;

[0025] 3. Multi-modal input and adaptive compatibility support text, voice, and image input. Geometric parameters can be extracted through image recognition and supplemented to the preliminary structured input card. At the same time, the system analyzes the input rules of the target program through static code analysis and dynamic tracking technology, and automatically adapts to different versions or types of fuel performance analysis programs;

[0026] 4. The result feedback is visualized. The output of traditional programs is mostly raw data files, which require secondary processing by professionals. After analyzing the calculation results of the present invention, a visual report (such as temperature field cloud map, stress distribution map) and a natural language summary are automatically generated, improving the readability of the results and the decision-making support ability. Description of the Drawings

[0027] Figure 1 This is the logic flow chart of the method of the present invention. Detailed implementation manners

[0028] The present invention will be further described in detail below in conjunction with the accompanying drawings and specific implementation manners:

[0029] The present invention discloses a fuel performance analysis method for natural language. Existing fuel performance analysis programs (such as BISON, FROBA, FRAPCON, FRAPTRAN, BEEs, etc.) require users to manually write complex input cards, and require users to be familiar with the program source code, parameter definitions, and physical models. The threshold for use is high, and the time-consuming for learning and calculation preparation is long. Through natural language processing technology, it is possible to convert the descriptive requirements input by users into the standardized input cards required by the fuel performance analysis system program, so that the same set of methods can be used to quickly implement fuel performance analysis calculations with reduced requirements for program mastery; the method logic flow is as Figure 1 shown;

[0030] The method includes the following steps:

[0031] Step 1: Construct a multi-level semantic library in the field of fuel performance analysis. The multi-level semantic library includes a basic parameter library, a working condition rule library, and a logical relationship library; at the same time, parse the source code and configuration files of the fuel performance analysis program, extract the syntax rules, logical constraints, and physical model dependencies of the input parameters through static code analysis and dynamic parameter tracking, integrate the official documents of the fuel performance analysis program, and construct a fuel performance analysis program logic library;

[0032] Construction of the multi-level semantic library:

[0033] 1) Basic parameter library: Covers fuel types (such as UO2, MOX), geometric models (such as rod-shaped, plate-shaped), material properties (such as thermal conductivity, expansion coefficient), etc., and extracts structured data from technical documents and historical input cards through entity extraction tools;

[0034] 2) Working condition rule library: Defines boundary conditions (convective heat transfer coefficient, wall temperature), power distribution, failure criteria, etc., and integrates typical parameter ranges in industry standards;

[0035] 3) Logical relationship library: Describes the dependency relationship between parameters (such as "time step needs to be defined for transient analysis"), and realizes dynamic association through knowledge graph technology;

[0036] Construction of the program logic library:

[0037] 1) Static code analysis: Use a parser to extract the input parameter definitions and the dependency relationships between models and parameters of the target program (such as BISON, FROBA, FRAPCON, FRAPTRAN, BEEs, etc.).

[0038] 2) Dynamic parameter tracking: Monitor the parameter call chain during program operation, record error feedback (such as "fuel_density parameter not found"), and improve the logical constraint rules.

[0039] 3) Document integration: Parse the user manual and official documents of the fuel performance analysis program, and supplement the implicit rules.

[0040] Step 2: Input data to enable the pre-trained language model to learn the professional corpus in the field of fuel performance analysis organized manually and the multi-level semantic library in the field of fuel performance analysis constructed in Step 1;

[0041] Step 3: Receive the natural language text or speech input by the user, and perform semantic parsing through the pre-trained language model that has learned the professional corpus in the field of fuel performance analysis to identify key calculation objects, operation instructions, and constraint conditions;

[0042] If there are problems with polysemy or ambiguous expressions in the user input, resulting in unsuccessful semantic parsing, the interaction requests the user to describe further and perform semantic parsing again until the parsing is successful and enter Step 4;

[0043] Step 4: Based on the semantic parsing results, integrate the fuel performance analysis program logic library and the multi-level semantic library, match the terms in the fuel performance analysis program logic library, and generate a preliminary structured input card;

[0044] Step 5: Perform multi-level logical verification on the generated input card, including parameter integrity check, unit consistency check, conflict condition detection, and fuel performance analysis program compatibility check. If there are errors or omissions in the logical verification results, request the user to supplement or correct through interactive prompts and return to Step 4 until the final input card is generated;

[0045] Step 6: Automatically submit the final input card to the fuel performance analysis program for calculation and monitor the calculation status in real time;

[0046] Step 7: Parse the calculation results, extract key variables, generate a visual report and a summary of key indicators, and feedback to the user in the form of a report file or natural language.

Claims

1. A fuel performance analysis method for natural language, characterized in that: The descriptive requirements input by the user are converted into the standardized input cards required by the fuel performance analysis program through natural language processing technology, realizing the seamless connection from natural language to fuel performance analysis; This method includes the following steps: Step 1: Construct a multi-level semantic library in the field of fuel performance analysis. The multi-level semantic library includes a basic parameter library, a working condition rule library, and a logical relationship library; at the same time, parse the source code and configuration files of the fuel performance analysis program, and extract the syntax rules, logical constraints, and physical model dependencies of the input parameters through static code analysis and dynamic parameter tracking to construct a fuel performance analysis program logic library; Step 2: Input materials to enable the pre-trained language model to learn the professional corpus in the field of fuel performance analysis organized manually and the multi-level semantic library in the field of fuel performance analysis constructed in Step 1; Step 3: Receive the natural language text or speech input by the user, and perform semantic parsing through the pre-trained language model that has learned the professional corpus in the field of fuel performance analysis to identify key calculation objects, operation instructions, and constraint conditions; If there are problems with polysemy or ambiguous expressions in the user input, resulting in unsuccessful semantic parsing, the interaction requests the user to describe further and perform semantic parsing again until the parsing is successful and proceeds to Step 4; Step 4: Based on the semantic parsing results, integrate the fuel performance analysis program logic library and the multi-level semantic library, match the terms in the fuel performance analysis program logic library, and generate a preliminary structured input card; Step 5: Perform multi-level logical verification on the generated input card, including parameter integrity check, unit consistency check, conflict condition detection, and fuel performance analysis program compatibility check. If there are errors or omissions in the logical verification results, request the user to supplement or correct through interactive prompts and return to Step 4 until the final input card is generated; Step 6: Automatically submit the final input card to the fuel performance analysis program for calculation and monitor the calculation status in real time; Step 7: Analyze the calculation results, generate a visual report and a summary of key indicators, and feedback to the user in the form of a report file or natural language.

2. The fuel performance analysis method for natural language according to claim 1, characterized in that: The basic parameter library includes fuel geometric dimensions and material properties; the working condition rule library includes boundary conditions and fuel-related parameter settings; the logical relationship library includes the dependency relationships between parameters.

3. The fuel performance analysis method for natural language according to claim 1, wherein: The pre-trained language model uses BERT, ChatGPT, or DeepSeek, and the professional corpus includes the technical documents of the fuel performance analysis program and standard reference input cards.

4. A method for analyzing fuel performance oriented to natural language according to claim 1, characterized in that: When accepting user input, multi-modal input is supported, including text, voice, and images. Geometric parameters are extracted through image recognition technology and supplemented to the preliminary structured input card.

5. The fuel performance analysis method for natural language according to claim 1, characterized in that: The fuel performance analysis program logic library parses the source code of the fuel performance analysis program through a static code analysis tool to extract the definition methods and dependency relationships of the input parameters; extracts the actual usage scenarios and error feedback mechanisms of the parameters by simulating the execution process of the input parameters through dynamic parameter tracking technology; integrates the official documents and user manuals of the target program to supplement implicit rules.

6. The fuel performance analysis method for natural language according to claim 1, characterized in that: The fuel performance analysis program uses a self-programmed fuel performance analysis program, the light water reactor fuel element performance analysis program FuelRodAnalysisProgram-Consortium, the fuel transient performance analysis program FRAPTRAN, the fuel performance analysis program FROBA developed by the Nuclear Reactor Thermal-Hydraulics Laboratory of Xi'an Jiaotong University, or the fuel performance analysis programs BISON and BEEs developed by secondary development based on the open-source platform MOOSE.

7. A method for analyzing fuel performance oriented to natural language according to claim 1, characterized in that: The conflict condition detection in the multi-level logic verification includes identifying thermodynamic model incompatibility problems, and the model incompatibility problems include conflicts between the fuel materials used and the input thermodynamic models.

8. The fuel performance analysis method for natural language according to claim 1, characterized in that: The interactive prompt includes repair suggestions, and the repair suggestions include adopting the default adaptive grid strategy or adjusting the parameter units.

9. A fuel performance analysis method for natural language according to claim 1, characterized in that: The visualization report includes the temperature field distribution and the stress nephogram; the key index summary includes the maximum stress and the local peak temperature.

10. A fuel performance analysis method for natural language according to claim 1, characterized in that: The natural language-oriented fuel performance analysis method performs data interaction with the fuel performance analysis program through the RESTful API interface; the natural language-oriented fuel performance analysis method deploys computing nodes through containerization technology and supports cloud-based distributed task scheduling.

Images