Intelligent analysis method, system and equipment for roadbed settlement of high-speed railway and storage medium

Abstract

The invention provides an intelligent analysis method, system and equipment for high-speed railway subgrade settlement and a storage medium, and the method comprises the steps: collecting high-speed railway subgrade settlement multi-source observation data, and building a subgrade settlement observation data set; creating an external function for data reading and analysis processing, and constructing an automatic response multi-round dialogue model based on a large language model; and based on the automatic response multi-round dialogue model, automatically reading and analyzing the roadbed settlement observation data set by calling a target external function according to a question of a user, and generating an automatic question and answer result. According to the method, the intelligent analysis model combining the multi-source observation data and knowledge dual-drive is established, intelligent analysis can be performed on roadbed settlement according to the collected high-speed railway roadbed settlement multi-source observation data, and the analysis result is fed back in real time through multi-round dialogue interaction with the user, so that a scientific basis is provided for decision making; and finally, intelligent and precise roadbed settlement monitoring and early warning are realized.

Description

Technical Field

[0001] The present invention relates to the technical field of subgrade settlement monitoring, and particularly to an intelligent analysis method, system, device and storage medium for high-speed railway subgrade settlement. Background Art

[0002] With the continuous progress of high-speed railway construction, subgrade settlement problems have become one of the important factors affecting railway safety, comfort and operation efficiency. Subgrade settlement not only directly affects the operation stability of the railway, but also may lead to damage to railway infrastructure and occurrence of safety accidents. Therefore, it is particularly important to conduct real-time monitoring and intelligent analysis of high-speed railway subgrade settlement. Traditional subgrade settlement monitoring technologies have problems such as difficult data acquisition, insufficient analysis accuracy, slow response speed, etc., and lack systematic intelligent analysis means, often relying on manual analysis and manual experience, and cannot meet the requirements of high efficiency and accuracy for subgrade settlement monitoring in high-speed railway construction and operation.

[0003] In recent years, with the rapid development of artificial intelligence technology, especially the maturity of natural language processing (NLP) and multi-source data fusion technology, using intelligent analysis methods to monitor and predict high-speed railway subgrade settlement has gradually become a hot topic in current research and practice. The present invention can effectively improve the automation, accuracy and real-time performance of subgrade settlement monitoring by establishing an intelligent analysis model that combines multi-source observation data and knowledge dual drive. Summary of the Invention

[0004] This application provides an intelligent analysis method, system, device and storage medium for high-speed railway subgrade settlement to solve the problems of low automation, insufficient analysis accuracy and poor real-time response of existing subgrade settlement monitoring technologies, which cannot meet the monitoring requirements.

[0005] According to a first aspect, in one embodiment, an intelligent analysis method for high-speed railway subgrade settlement is provided, and the method includes:

[0006] Collect multi-source observation data of high-speed railway subgrade settlement and establish a subgrade settlement observation data set;

[0007] Create an external function for data reading and analysis processing, and construct an automatic answering multi-round dialogue model based on a large language model;

[0008] Based on the automatic answering multi-round dialogue model, automatically read and analyze the subgrade settlement observation data set according to the user's question and by calling the target external function, and generate an automatic answering result.

[0009] Further, collecting multi-source observation data of high-speed railway subgrade settlement and establishing a subgrade settlement observation data set specifically includes:

[0010] Perform multi-source data observation using a pre-established multi-source data observation device for subgrade settlement;

[0011] The multi-source observation data of subgrade settlement includes soil stress values, subgrade settlement values, moisture content inside the soil mass, average daily temperature, and maximum daily temperature difference at a preset distance below the subgrade surface, in the middle of the subgrade, and at the bottom of the subgrade.

[0012] Furthermore, create external functions for data reading and analysis processing, specifically including:

[0013] Create a data reading function, which is used to identify and parse the format of multi-source observation data, cache and incrementally update data, and perform data encryption processing;

[0014] Create a data analysis and processing function, which is used to convert and parse the natural language input by the user to obtain the user's intention, perform complex condition queries and integrations on multi-source observation data, and visualize and display the data and results.

[0015] Furthermore, construct an automatic answering multi-round dialogue model based on a large language model, specifically including:

[0016] Create a dialogue answering function check_code_run for making external function calls and generating model answers according to each user question;

[0017] Based on the created dialogue answering function check_code_run, create a multi-round dialogue answering function chat_with_inter for interactive multi-round dialogue answering.

[0018] Furthermore, the dialogue answering function check_code_run is specifically used for:

[0019] a. Determine whether there is an external function library:

[0020] If not, perform ordinary dialogue tasks and directly call the large language model to obtain the response result. If so, it is necessary to flexibly select external functions and answer;

[0021] b. Call the large language model for the first time:

[0022] Call the large language model, pass in parameters, and obtain the first response result of the model;

[0023] c. Determine whether it is necessary to call an external function:

[0024] Judge whether it is necessary to call an external function to answer the question by checking whether the first response result of the model contains a preset function call field;

[0025] If an external function needs to be called, obtain the function name, function object, and function parameters to be called;

[0026] Determine whether the called external function is automatically executed. If it is not automatically executed, prompt the user to confirm whether to execute the code, and execute the code or exit the operation according to the user's selection. If it is automatically executed, directly execute the external function without user confirmation;

[0027] Call the external function with the parsed function parameters, obtain the calculation result of the external function, and add the result returned by the external function to the message; after adding the response of the external function, send the complete message to the model; obtain the final model answer and use it as the final result;

[0028] If an external function does not need to be called, directly use the content of the model's first response as the final result;

[0029] d. Delete the message list;

[0030] e. Return the final response content of the model.

[0031] Furthermore, the multi-round dialogue Q&A function chat_with_inter is specifically used for:

[0032] a. Project file creation: Obtain the name of the current analysis project input by the user; create or obtain the project folder path;

[0033] b. Input of the analysis phase name: Obtain the name of the current analysis phase input by the user;

[0034] c. Set the multi-round dialogue threshold: Set the token threshold for multi-round dialogue according to the large language model type;

[0035] d. Initialize the message list and user input: Initialize the message list and add the user input to the message list; calculate the initial number of tokens;

[0036] e. Enter the multi-round dialogue loop: Call the check_code_run function, obtain the model answer and print the model answer;

[0037] f. Record the answer or enter the question again:

[0038] Ask the user whether to record the current answer or enter the question again;

[0039] If the user chooses to record the answer, save the question and answer to the specified folder and add the question and answer of this round to the message list;

[0040] If the user selects to input the question again, prompt the user to input a new question, update the user input in the message list, and call the check_code_run function to obtain the model answer again and print the new model answer;

[0041] g. Ask the user if there are any other questions:

[0042] If not, delete the message list and break out of the outer loop; otherwise, record the new round of questions in the message list and start a new round of conversation;

[0043] h. Calculate the current total number of tokens and delete the conversation content that exceeds the token threshold.

[0044] Furthermore, construct an automatic answering multi-round dialogue model based on a large language model, specifically including:

[0045] Divide the data reading and analysis processing tasks into different analysis stages, including data exploration, data preprocessing, and machine learning prediction stages. Set multiple test questions for each analysis stage respectively, input the test questions of each stage into the automatic answering multi-round dialogue model, and obtain the multi-round dialogue Q&A results to complete the model test.

[0046] According to the second aspect, an intelligent analysis system for high-speed railway subgrade settlement is provided in an embodiment. The system includes:

[0047] A data collection module, which is used to collect multi-source observation data of high-speed railway subgrade settlement and establish a subgrade settlement observation data set;

[0048] A function and model creation module, which is used to create external functions for data reading and analysis processing and construct an automatic answering multi-round dialogue model based on a large language model;

[0049] An intelligent analysis module, which is used to automatically read and analyze the subgrade settlement observation data set based on the automatic answering multi-round dialogue model according to the user's question and by calling the target external function, and generate an automatic Q&A result.

[0050] According to the third aspect, an electronic device is provided in an embodiment. The device includes: a processor and a memory;

[0051] The memory is used to store one or more program instructions;

[0052] The processor is used to run one or more program instructions to execute the steps of an intelligent analysis method for high-speed railway subgrade settlement as described in any one of the above.

[0053] According to a third aspect, in one embodiment, a computer-readable storage medium is provided. A computer program is stored on the computer-readable storage medium, and when the computer program is executed by a processor, the steps of an intelligent analysis method for high-speed railway subgrade settlement as described in any one of the above are implemented.

[0054] This application provides an intelligent analysis method, system, device, and storage medium for high-speed railway subgrade settlement. It collects multi-source observation data of high-speed railway subgrade settlement and establishes a subgrade settlement observation data set; creates external functions for data reading and analysis processing, and constructs an automatic answering multi-round dialogue model based on a large language model; based on the automatic answering multi-round dialogue model, according to user questions and by calling target external functions, it automatically reads and analyzes the subgrade settlement observation data set to generate automatic answering results. By establishing an intelligent analysis model combining multi-source observation data and knowledge dual-driving, this application can intelligently analyze subgrade settlement based on the collected multi-source observation data of high-speed railway subgrade settlement, and through multi-round dialogue interaction with users, it can provide real-time feedback on the analysis results, providing a scientific basis for decision-making, and ultimately realizing intelligent and precise subgrade settlement monitoring and early warning. Brief Description of the Drawings

[0055] Figure 1 It is a flowchart of an intelligent analysis method for high-speed railway subgrade settlement provided by an embodiment of the present invention;

[0056] Figure 2 It is a schematic diagram of a multi-source data observation device in an intelligent analysis method for high-speed railway subgrade settlement provided by an embodiment of the present invention;

[0057] Figure 3 It is the functional test result of the external function extract_data in an intelligent analysis method for high-speed railway subgrade settlement provided by an embodiment of the present invention;

[0058] Figure 4 It is the functional test result of the external function python_inter in an intelligent analysis method for high-speed railway subgrade settlement provided by an embodiment of the present invention;

[0059] Figure 5 It is the implementation process of the Chat dialogue answering function check_code_run in an intelligent analysis method for high-speed railway subgrade settlement provided by an embodiment of the present invention;

[0060] Figure 6 It is the implementation process of the multi-round dialogue answering function chat_with_inter in an intelligent analysis method for high-speed railway subgrade settlement provided by an embodiment of the present invention;

[0061] Figure 7Multi-round Q&A results in the data exploration stage of an intelligent analysis method for high-speed railway subgrade settlement provided by an embodiment of the present invention;

[0062] Figure 8 Multi-round Q&A results in the data preprocessing stage of an intelligent analysis method for high-speed railway subgrade settlement provided by an embodiment of the present invention;

[0063] Figure 9 Multi-round Q&A results in the prediction stage of the random forest algorithm in an intelligent analysis method for high-speed railway subgrade settlement provided by an embodiment of the present invention;

[0064] Figure 10 Schematic diagram of the logical structure of an intelligent analysis system for high-speed railway subgrade settlement provided by an embodiment of the present invention. Detailed implementation manners

[0065] The present invention will be further described in detail below in conjunction with the specific implementation manners and the accompanying drawings. Similar elements in different implementation manners are labeled with related similar element numbers. In the following implementation manners, many details are described to enable a better understanding of the present application. However, those skilled in the art can easily recognize that some of the features can be omitted in different situations, or can be replaced by other elements, materials, or methods. In some cases, some operations related to the present application are not shown or described in the specification to avoid the core part of the present application being overwhelmed by excessive description. For those skilled in the art, it is not necessary to describe these related operations in detail, and they can fully understand the related operations based on the description in the specification and the general technical knowledge in the art.

[0066] In addition, the features, operations, or characteristics described in the specification can be combined in any appropriate manner to form various implementation manners. At the same time, the steps or actions in the method description can also be reordered or adjusted in an obvious manner by those skilled in the art. Therefore, the various sequences in the specification and the drawings are only for clearly describing a certain embodiment and do not mean a necessary sequence unless it is stated that a certain sequence must be followed.

[0067] The first embodiment of the present invention provides an intelligent analysis method for high-speed railway subgrade settlement. The following is a detailed description in conjunction with Figure 1 for detailed description.

[0068] As Figure 1 shown, in step S100, multi-source observation data of high-speed railway subgrade settlement is collected to establish a subgrade settlement observation data set.

[0069] The above steps specifically include: performing multi-source data observation using a pre-established multi-source data observation device for subgrade settlement; the multi-source observation data for subgrade settlement includes soil stress values at 0.5 m below the subgrade top surface, in the middle of the subgrade, and at the bottom of the subgrade, subgrade settlement values, moisture content inside the soil body, average daily temperature, and maximum daily temperature difference.

[0070] In this embodiment, the multi-source data observation device for subgrade settlement is as Figure 2 shown, and includes stress sensors, static level gauges, moisture meters, thermometers, a data transmission system, and a power supply system. The stress sensors are used to measure the soil stress values at 0.5 m below the subgrade top surface, in the middle of the subgrade, and at the bottom of the subgrade. The static level gauges are used to obtain the settlement data of the foundation and the subgrade pavement. The moisture meters are used to obtain the moisture content of the soil inside the foundation. The thermometers are used to obtain the ambient temperature near the subgrade.

[0071] As Figure 1 shown, in step S200, an external function for data reading and analysis processing is created, and an automatic answering multi-round dialogue model based on a large language model is constructed.

[0072] The above steps specifically include:

[0073] Step S210, creating an external function for data reading and analysis processing, specifically including:

[0074] S211, creating a data reading function, which is used to perform format recognition and parsing, data caching and incremental updating, and data encryption processing on the multi-source observation data;

[0075] S212, creating a data analysis and processing function, which is used to convert and parse the natural language input by the user to obtain the user's intention, perform complex condition queries and integration on the multi-source observation data, and perform data visualization and result display.

[0076] In this embodiment, two external functions, namely local data reading and data processing, are created to implement the reading and processing of relevant data in the subgrade settlement observation section. These two external functions will be automatically called as basic functions in the subsequent dialogue system, specifically as follows:

[0077] 1. Construct an external function extract_data for data extraction, which is used to read the local subgrade settlement data and save it to the local Python environment. It is the core module connecting the multi-source data observation system and the intelligent analysis agent model. The specific functions of this external function are as follows:

[0078] (1) Multi-source data dynamic adaptation mechanism: Adapt to the diverse data formats of different sensors (such as stress sensors, static level gauges, etc.). By introducing a dynamic adaptation mechanism, the function can automatically identify and parse the data structures of different devices and uniformly convert them into standardized data tables. Specifically, first, match the data file format (such as CSV, JSON, XML) through regular expressions, then call the corresponding parser (such as pandas.read_csv, json.load) to load the data according to the recognition result, and finally convert the parsed data into a unified format and add metadata tags (such as sensor type, acquisition time).

[0079] (2) Data caching and incremental update: In the settlement monitoring of high-speed railway subgrades, the data volume is huge and the updates are frequent. By introducing a caching mechanism, the function can store historical data locally to avoid repeated reading; at the same time, combined with timestamp marking, incremental data loading is achieved to improve the efficiency of large-scale data processing.

[0080] (3) Data encryption and secure transmission: During data transmission and storage, sensitive data (such as subgrade stress values) may face the risk of leakage. By integrating SSL / TLS protocols and hash verification technologies, the security and integrity of the data are ensured.

[0081] By constructing an interpretation list of the functions parameter of the external function extract_data, specifically passing the information of this external function to the model, so that the large language model can identify, interpret, and call this function.

[0082] 2. Construct an external function python_inter for data processing, whose main function is to achieve various queries and processing of subgrade settlement data through the user's natural language questions. The specific functions of this external function are as follows:

[0083] (1) Support multiple natural language expressions: Support questions in multiple natural language expressions. Different users may be accustomed to different ways of asking questions, and the system can identify and process these differences. In this embodiment, a statement interpretation module is designed to convert the natural language questions input by the user into standardized commands or operation requests. This module automatically judges the intention of the question and matches it to the appropriate function module based on a combination of a deep learning model and a rule engine.

[0084] (2) Support for complex query conditions and operation functions: By introducing natural language parsing technology, the function can understand and execute complex query conditions; by supporting multi-table association queries, the function can integrate multi-source data and generate comprehensive analysis results.

[0085] (3) Data visualization and result display: This function provides rich visualization capabilities to help users understand the analysis results. Based on different types of data, python_inter can generate dynamic charts and also generate interactive reports. Users can select different analysis views in the report to view data from different dimensions.

[0086] By constructing an interpretation list of the functions parameter of the external function python_inter, information about this external function is specifically passed to the model so that the large language model can recognize, interpret, and call this function.

[0087] Function calling function tests are as Figure 3 、 Figure 4 shown, indicating that both of the above external functions can be normally called by the large language model.

[0088] Step S220, constructing an automatic response multi-round dialogue model based on the large language model, specifically including:

[0089] S221, creating a dialogue Q&A function check_code_run that can perform external function calls and generate model answers according to each user question.

[0090] In this embodiment, a Chat dialogue Q&A function check_code_run that can automatically execute external function calls is created, which is specifically used in the construction process of the code interpreter. Its core function is to achieve seamless integration and interaction between the intelligent dialogue system and external functions.

[0091] The dialogue Q&A function check_code_run created in this embodiment has parameters including: messages (a list of messages for interacting with the model), functions_list (a list object containing all external functions), functions (a list in Schema format containing explanations of all external function parameters), model (the Chat large language model used), auto_run (whether to automatically run in the case of calling an external function). The specific function description of this function is as follows, and the specific implementation process is as Figure 5 shown.

[0092] a. Determine whether there is an external function library:

[0093] If functions_list is None, perform ordinary dialogue tasks and directly call the model to obtain a response.

[0094] If functions_list is not None, it is necessary to flexibly select external functions and give answers.

[0095] b. First call the chat model:

[0096] Use the client.chat.completions.create method to call the chat model, passing in the model, messages, and functions parameters to obtain the first response of the model.

[0097] c. Determine whether to call external functions:

[0098] Check whether the first response of the model contains the function_call field, that is, determine whether to call external functions to answer questions.

[0099] ① If external functions need to be called: Obtain the function name, function object, and function parameters, and convert the parameters from JSON format to a Python dictionary.

[0100] If auto_run is False, prompt the user to confirm whether to execute the code. The user can choose: execute the code (continue to execute the external function, obtain the result and return) or exit the run (terminate the current code execution and return None). If auto_run is True, directly execute the external function without user confirmation.

[0101] Call the external function using the parsed function parameters, obtain the function calculation result, and add the result returned by the function to the message as the function response. After adding the response of the external function, send the complete message (including user input, model response, external function output, etc.) to the model for the second call. Obtain the final model answer and use it as the final result.

[0102] ② If external functions do not need to be called: Directly use the content of the first response as the final result.

[0103] d. Delete the message list: To save memory, delete the messages list.

[0104] e. Return the final result: Return the final response content of the model.

[0105] S222. Based on the created dialogue Q&A function check_code_run, create a multi-round dialogue Q&A function chat_with_inter for interactive multi-round dialogue Q&A.

[0106] In this embodiment, based on the dialogue Q&A function check_code_run, a multi-turn dialogue Q&A model chat_with_inter is constructed to achieve interactive multi-turn dialogue. It can not only execute code after user confirmation, but also save the Q&A results of each key question (after user review) to a local document, facilitating the preparation of the final data analysis report. At the same time, a historical dialogue message function is added to the multi-turn dialogue model to achieve long-term memory of historical messages.

[0107] The multi-turn dialogue Q&A function chat_with_inter created in this embodiment has the following parameters: functions_list: (a list object containing all external functions), prompt (the initial prompt or question entered by the user), model (the specified Chat model to use), system_message: (system message used to set the behavior and role of the model), auto_run (whether to automatically make a second response in the case of calling an external function). The specific function description is as follows, and the specific implementation process is as Figure 6 shown.

[0108] a. Project file creation:

[0109] Obtain the current analysis project name project_name entered by the user through the input function.

[0110] Call the create_or_get_folder function (already defined) to create or obtain the project folder path.

[0111] b. Input of analysis phase name:

[0112] Obtain the current analysis phase name doc_name entered by the user through the input function, such as "Data Exploration Phase", "Data Preprocessing Phase", etc.

[0113] c. Set the multi-turn dialogue threshold:

[0114] Set the token threshold tokens_thr for multi-turn dialogue according to the model type: If the model contains gpt-4, then tokens_thr is 6000; if the model contains 16k, then tokens_thr is 14000; otherwise, tokens_thr is 3000.

[0115] d. Initialize the message list and user input:

[0116] Initialize the message list message and add the user input to the message list. Calculate the initial token count tokens_count.

[0117] e. Enter the multi-round conversation loop:

[0118] Call the check_code_run function to obtain the model's answer and print the model's answer.

[0119] f. Record the answer or enter the question again:

[0120] Ask the user through the input function whether to record the current answer (enter 1) or enter the question again and generate an answer (enter 2).

[0121] If the user chooses to record the answer: Construct the question string Q_temp and the answer string A_temp. Call the append_or_create_doc_in_folder function (already defined) to save the question and answer to the specified folder, and add the question and answer of this round to the message list.

[0122] If the user chooses to enter the question again: Prompt the user to enter a new question and update the user input user_input. Update the user input in the message list and call the check_code_run function to obtain the model's answer again. Print the new model's answer.

[0123] g. Ask the user if there are any other questions:

[0124] Ask the user through the input function if there are any other questions, and enter "exit" to end the conversation. If the user enters "exit", delete the message list and break out of the outer loop. Otherwise, record the new round of questions in the message list and start a new round of conversation.

[0125] h. Calculate the current total number of tokens and delete the conversation content that exceeds the token threshold:

[0126] Calculate the current total number of tokens tokens_count, including the newly generated model's answer and the user input. If the current total number of tokens exceeds the threshold tokens_thr, delete the conversation content in the message list that exceeds the threshold until the number of tokens is lower than the threshold.

[0127] As Figure 1 shown, in step S300, based on the automatic answering multi-round conversation model, according to the user's question and by calling the target external function, the subgrade settlement observation data set is automatically read and analyzed to generate an automatic Q&A result.

[0128] This embodiment tests the construction of an automatic response multi-round dialogue model based on a large language model, specifically including: dividing the data reading and analysis processing tasks into different analysis stages, including data exploration, data preprocessing, and machine learning prediction stages. Multiple test questions are respectively set for each analysis stage, and the test questions of each stage are respectively input into the automatic response multi-round dialogue model to obtain the multi-round dialogue Q&A results to complete the model test.

[0129] Specifically, in this embodiment, around the local subgrade settlement dataset, the multi-round dialogue model chat_with_inter is used to conduct multi-round Q&A tests on three stages: data exploration, data preprocessing, and random forest algorithm prediction, and the Q&A that passes the review is saved locally.

[0130] 1. In the data exploration stage, 5 test questions are proposed to the large model, and the Q&A information is saved locally. The 5 questions are respectively:

[0131] 1) Please read the local'subgrade_settlement_' table and name it Data.

[0132] 2) How many rows and columns does this data have in total, and what are the names of each column?

[0133] 3) Does this data have missing values? If so, how many missing values are there in total?

[0134] 4) Are there any outliers in this group of data?

[0135] 5) Please conduct descriptive statistics on each column, and calculate the mean, maximum value, minimum value, and variance of each column respectively.

[0136] The large model's answers to the above 5 questions are as Figure 7 shown. It can be seen that the large model has successfully answered the above 5 questions and saved the questions to the local document "Q&A in the Data Exploration Stage.docx". The multi-round Q&A results are as Figure 7 shown.

[0137] 2. In the data preprocessing stage, 5 test questions are proposed to the large model, and the Q&A information is saved locally. The 5 questions are respectively:

[0138] 1) Please read the local'subgrade_settlement' table and name it Data.

[0139] 2) Fill in the missing values in the data table. The first column is filled with the mode of this column, and the rest of each column is filled with the mean of this column. Save the filled table to the Data_1 variable.

[0140] 3) Save the variable Data_1 to the local table named "subgrade_settlement1".

[0141] 4) Please help me visualize the proportion of different values in the first column of Data_1.

[0142] 5) Save the picture to the local. The processing results of the large model for the above 5 questions are as Figure 8 shown.

[0143] It can be seen that the large model has successfully processed the five requirements for the table above and saved the Q&A results to the local document "Q&A in the data preprocessing stage.docx". The multi-round Q&A results are as Figure 8 shown.

[0144] 3. In the prediction stage of the random forest algorithm, 4 questions are asked to the large model and the Q&A information is saved locally. The 5 questions are as follows:

[0145] 1) Please read the local'subgrade_settlement_' table and name it Data.

[0146] 2) Extract the feature and label matrices of the data. The first 8 columns are features and the last column is the label, which are saved to variables X and Y respectively.

[0147] 3) Use the random forest algorithm to train with the training set data and verify the model effect with the test set data to obtain the evaluation indicators of the model: goodness of fit, MAE.

[0148] 4) If the evaluation result is not satisfactory, use grid search for hyperparameter optimization. The hyperparameters selected are n_estimators and max_depth. After training, use the test set data for prediction and obtain the evaluation indicators of the model: goodness of fit, MAE.

[0149] The answers of the large model to the above 4 questions are as Figure 9 shown. It can be seen that the large model has successfully processed the four requirements for the table above and saved the Q&A results to the local document "Q&A in the random forest algorithm prediction stage.docx". The multi-round Q&A results are as Figure 9 shown.

[0150] Corresponding to the above-disclosed intelligent analysis method for high-speed railway subgrade settlement, an embodiment of the present invention also discloses an intelligent analysis system for high-speed railway subgrade settlement, as Figure 10 shown, which specifically includes:

[0151] A data collection module for collecting multi-source observation data of high-speed railway subgrade settlement and establishing a subgrade settlement observation data set;

[0152] A function and model creation module for creating external functions for data reading, analysis, and processing, and constructing an automatic answering multi-round dialogue model based on a large language model;

[0153] An intelligent analysis module for automatically reading and analyzing the subgrade settlement observation data set based on the automatic answering multi-round dialogue model according to the user's question and by calling the target external function, and generating an automatic Q&A result.

[0154] It should be noted that for the detailed description of a high-speed railway subgrade settlement intelligent analysis system provided by an embodiment of the present invention, reference can be made to the relevant description of a high-speed railway subgrade settlement intelligent analysis method provided by an embodiment of the present application, which will not be elaborated here.

[0155] In addition, an embodiment of the present invention further provides an electronic device, which includes: a processor and a memory; the memory is used to store one or more program instructions; the processor is used to run one or more program instructions to execute the steps of a high-speed railway subgrade settlement intelligent analysis method as described in any one of the above.

[0156] It should be noted that for the detailed description of an electronic device provided by an embodiment of the present invention, reference can be made to the relevant description of a high-speed railway subgrade settlement intelligent analysis method provided by an embodiment of the present application, which will not be elaborated here.

[0157] In addition, an embodiment of the present invention further provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, it implements the steps of a high-speed railway subgrade settlement intelligent analysis method as described in any one of the above.

[0158] It should be noted that for the detailed description of a computer-readable storage medium provided by an embodiment of the present invention, reference can be made to the relevant description of a high-speed railway subgrade settlement intelligent analysis method provided by an embodiment of the present application, which will not be elaborated here.

[0159] Those skilled in the art can understand that all or part of the functions of the various methods in the above embodiments can be implemented in a hardware manner or in a computer program manner. When all or part of the functions in the above embodiments are implemented in a computer program manner, the program can be stored in a computer-readable storage medium, which can include: read-only memory, random access memory, magnetic disks, optical disks, hard disks, etc. The above functions can be realized by a computer executing the program. For example, the program is stored in the memory of the device, and when the processor executes the program in the memory, the above all or part of the functions can be realized. In addition, when all or part of the functions in the above embodiments are implemented in a computer program manner, the program can also be stored in a storage medium such as a server, another computer, magnetic disk, optical disk, flash drive or mobile hard disk, and saved to the memory of the local device by downloading or copying, or the system of the local device is updated. When the processor executes the program in the memory, all or part of the functions in the above embodiments can be realized.

[0160] The above uses specific examples to elaborate on the present invention, which is only used to help understand the present invention and is not intended to limit the present invention. For those skilled in the art of the present invention, according to the idea of the present invention, several simple deductions, deformations or substitutions can also be made.

Claims

1. An intelligent analysis method for high-speed railway subgrade settlement, characterized in that: The method comprises: Collect multi-source observation data on high-speed railway subgrade settlement and establish a subgrade settlement observation dataset; Create external functions for data reading and analysis, and build an automatic response multi-round dialogue model based on a large language model; Based on the automatic answering multi-round dialogue model, the roadbed settlement observation data set is automatically read and analyzed according to the user's questions and by calling the target external function to generate an automatic question and answer result.

2. The intelligent analysis method for high-speed railway subgrade settlement according to claim 1, characterized in that: Collect multi-source observation data of high-speed railway subgrade settlement and establish a subgrade settlement observation data set, including: Use the pre-established roadbed settlement multi-source data observation device to conduct multi-source data observation; The multi-source observation data of roadbed settlement include soil stress values ​​at a preset distance below the roadbed top surface, in the middle of the roadbed and at the bottom of the roadbed, roadbed settlement values, soil internal moisture content, average temperature of the day and maximum temperature difference of the day.

3. The intelligent analysis method for high-speed railway subgrade settlement according to claim 1, characterized in that: Create external functions for data reading and analysis, including: Create a data reading function, which is used to perform format recognition and parsing, data caching and incremental update, and data encryption processing on multi-source observation data; Create a data analysis processing function, which is used to convert and parse the natural language input by the user to obtain the user's intention, perform complex conditional query and integration on multi-source observation data, and visualize and display the data results.

4. The intelligent analysis method for high-speed railway subgrade settlement according to claim 1, characterized in that: Build an automatic response multi-round dialogue model based on a large language model, including: Create a dialog question-answering function check_code_run that can call external functions and generate model answers based on each user question; Based on the created dialogue question and answer function check_code_run, create a multi-round dialogue question and answer function chat_with_inter for interactive multi-round dialogue question and answer.

5. The intelligent analysis method for high-speed railway subgrade settlement according to claim 4, characterized in that: The dialogue question and answer function check_code_run is specifically used to: a. Determine whether there is an external function library: If not, perform the normal dialogue task and directly call the large language model to obtain the response result. If yes, you need to flexibly select the external function and give the answer; b. First call to the large language model: Call the large language model, pass in parameters, and get the first response result of the model; c. Determine whether an external function needs to be called: By checking whether the model's first response contains the preset function call field, it is determined whether an external function needs to be called to answer the question; If you need to call an external function, get the function name, function object, and function parameters that need to be called; Determine whether the called external function is automatically executed. If not, prompt the user to confirm whether to execute the code. Execute the code or exit the run according to the user's choice. If it is automatically executed, directly execute the external function without user confirmation. Use the parsed function parameters to call the external function, obtain the calculation result of the external function, and add the result returned by the external function to the message; Send the complete message to the model, after adding the response from the external function; Get the final model answer and use it as the final result; If there is no need to call an external function, the content of the model's first response is directly used as the final result; d. Delete the message list; e. Return the final response content of the model.

6. The intelligent analysis method for high-speed railway subgrade settlement according to claim 4, characterized in that: The multi-round dialogue question-answering function chat_with_inter is specifically used for: a. Project file creation: Get the current analysis project name entered by the user; create or get the project folder path; b. Input of analysis phase name: Get the name of the current analysis phase input by the user; c. Set the multi-round dialogue threshold: Set the token threshold for multi-round dialogues according to the large language model type; d. Initialize message list and user input: Initialize the message list and add user input to the message list; Calculate the initial token number; e. Enter a multi-round dialogue loop: call the check_code_run function, obtain the model's answer, and print the model's answer; f. Record your answer or re-enter the question: Ask the user whether to record the answer or re-enter the question; If the user chooses to record the answer, the question and answer will be saved in the specified folder, and the answer to this round of questions will be added to the message list; If the user chooses to enter the question again, the user is prompted to enter a new question, the user input in the message list is updated, and the check_code_run function is called to re-obtain the model answer and print the new model answer; g. Ask the user if they have any other questions: If not, delete the message list and exit the outer loop; otherwise, record a new round of questions in the message list and start a new round of conversation; h. Calculate the current total number of tokens and delete conversations that exceed the token threshold.

7. The intelligent analysis method for high-speed railway subgrade settlement according to claim 1, characterized in that: Build an automatic response multi-round dialogue model based on a large language model, including: The data reading and analysis tasks are divided into different analysis stages, including data exploration, data preprocessing, and machine learning prediction stages. Multiple test questions are set for each analysis stage. The test questions of each stage are input into the automatic answering multi-round dialogue model respectively, and the multi-round dialogue question and answer results are obtained to complete the model testing.

8. An intelligent analysis system for high-speed railway subsidence, characterized in that: The system comprises: Data collection module, used to collect multi-source observation data of high-speed railway roadbed settlement and establish roadbed settlement observation data set; Function and model creation module, used to create external functions for data reading and analysis, and to build an automatic response multi-round dialogue model based on a large language model; The intelligent analysis module is used to automatically read and analyze the roadbed settlement observation data set based on the automatic answering multi-round dialogue model according to the user's questions and by calling the target external function to generate an automatic question and answer result.

9. An electronic device, characterized in that: The device comprises: a processor and a memory; The memory is used to store one or more program instructions; The processor is used to run one or more program instructions to execute the steps of a high-speed railway roadbed settlement intelligent analysis method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the high-speed railway subgrade settlement intelligent analysis method as described in any one of claims 1 to 7 are implemented.

Images