Intelligent test design method and system based on distributed network architecture

Abstract

The present application relates to a kind of intelligent test design method and system based on distributed network architecture.The method includes: loading historical dialogue record, input file of test design algorithm and user request information.In the backend middleware layer, the legality of input file is verified, if the legality of input file is abnormal, historical dialogue record and user request information are used as training sample data set, trigger the big model of business layer to make decision training according to the input file on training sample data set, output test design strategy, and spit out dialogue to front-end interactive layer.Otherwise, reconstruct input parameter in input file, call the calculation core of test design algorithm to start independent thread, so that big model executes calculation task, and outputs calculation result.After loading the dialogue of big model to calculation result, spit out analysis report to front-end interactive layer.The method can improve the development efficiency and versatility of virtual computing test.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of virtual computing experiment, in particular to an intelligent experiment design method and system based on a distributed network architecture. BACKGROUND

[0002] Traditional experiment design methods are widely used in industrial optimization and scientific research, but their parameter configuration relies on manual experience and is difficult to meet the optimization needs of complex scenarios. In the prior art, experiment design systems mostly use localized software or fixed function modules, which have the problems of poor flexibility and insufficient intelligence. While large model technology has shown great capabilities in data analysis and pattern prediction, it has not been deeply integrated with experiment design methods.

[0003] Currently, traditional experiment design methods require manual setting of parameter ranges and sample sizes, which is low in efficiency and susceptible to subjective experience, making it difficult to ensure the optimality of parameter combinations. Existing systems cannot dynamically adjust the experiment scheme according to historical data or real-time feedback, leading to repeated experiments or insufficient coverage. Most experiment design systems use fixed function modules, and adding new algorithms or modifying existing logic requires code reconstruction, which is high in development cost and long in cycle. New methods cannot be quickly integrated through plug-in methods, making it difficult to meet the needs of multiple scenarios. SUMMARY

[0004] Therefore, it is necessary to provide an intelligent experiment design method and system based on a distributed network architecture to improve the development efficiency and universality of virtual computing experiments.

[0005] An intelligent experiment design method based on a distributed network architecture is applied to a distributed network architecture, which includes a front-end interaction layer, a back-end middleware layer, and a business layer. The method comprises:

[0006] Loading historical conversation records, input files of experiment design algorithms, and user request information.

[0007] In the back-end middleware layer, the legality of the input file is verified. If the legality of the input file is abnormal, the historical conversation records and the user request information are used as a training sample data set, and the business layer triggers a large model to perform decision training on the training sample data set according to the input file, outputs an experiment design strategy, and outputs a conversation to the front-end interaction layer. Otherwise, the input parameters in the input file are reconstructed, and the calculation core of the experiment design algorithm is called to start an independent thread to enable the large model to perform a calculation task and output a calculation result.

[0008] After loading the calculation result into the large model conversation, an analysis report is output to the front-end interaction layer.

[0009] An intelligent experiment design system based on a distributed network architecture, the system comprises:

[0010] An information loading module is configured to load historical dialogue records, input files of the experimental design algorithm, and user request information.

[0011] An experiment execution module is configured to verify the legality of the input files in the backend middleware layer. If the legality of the input files is abnormal, the historical dialogue records and the user request information are used as a training sample dataset, and a large model of a business layer is triggered to perform decision training on the training sample dataset according to the input files, to output an experimental design strategy, and to output a dialogue to a front-end interactive layer. Otherwise, input parameters in the input files are reconstructed, and a calculation core of the experimental design algorithm is called to start an independent thread, so that the large model performs a calculation task and outputs a calculation result.

[0012] A report display module is configured to output an analysis report to the front-end interactive layer after the calculation result is loaded into the dialogue of the large model.

[0013] The intelligent experimental design method and system based on the distributed network architecture have the following advantages. The modular architecture is adopted in the design of the function loading middleware, the functional components such as the experimental design algorithm and the large model decision are grouped into dynamically loadable modules, new methods can be quickly integrated in the form of plug-ins without modifying the core code, the development cost and cycle are greatly reduced, and flexible adaptation to multiple scene requirements is achieved. The whole-process automation mechanism runs through the input file legality verification, the training sample generation, the large model decision training, the calculation task execution, and the analysis report output, to form a closed loop of “design-verification-improvement”, and avoid the efficiency loss caused by manual intervention. The intelligent interaction and decision support driven by the large model can automatically generate a training sample based on the historical dialogue records and the user request information when the legality of the input files is abnormal, drive the large model to perform decision training and output a strategy, and when the input is legal, an independent thread is started by reconstructing parameters to call the algorithm calculation core, to realize the efficient execution of the calculation task by the large model, and effectively improve the test efficiency and reduce the calculation cost in combination with the multi-task flexible scheduling mechanism. This design not only enables the system to quickly adapt to different industry requirements, but also provides standardized interfaces for the expansion of new methods and new scenes through the function loading characteristics of the middleware, realizes the flexibility of the technical architecture and the adaptability of the business requirements, and improves the development efficiency and universality of the virtual calculation test. BRIEF DESCRIPTION OF DRAWINGS

[0014] Figure 1 FIG. 1 is an application scenario diagram of the intelligent experimental design method based on the distributed network architecture in one embodiment;

[0015] Figure 2 FIG. 2 is a flowchart of the intelligent experimental design method based on the distributed network architecture in one embodiment;

[0016] Figure 3A system function flowchart in one embodiment;

[0017] Figure 4 A structural block diagram of an intelligent experiment design system based on a distributed network architecture in one embodiment. DETAILED DESCRIPTION

[0018] In order to make the purposes, technical solutions and advantages of the present application clearer, the present application is further described in detail below with reference to the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and do not limit the present application.

[0019] The intelligent experiment design method based on the distributed network architecture provided by the present application can be applied to the framework structure as shown in Figure 1 . It contains front and back ends, a back end and a business layer. The front and back ends provide function buttons (for example, Latin hypercube design) of the experiment design method, support user parameter input and result visualization. The back end contains function loading middleware (for example, get_crazy_function) and business total control middleware (for example, latinHypercube), which can load algorithm modules of various experiment design methods, analyze and call the specific implementation logic of the business layer, and coordinate the interaction between the large model and the experiment design method. The business layer is mainly the specific implementation of the experiment design method, including the large model and the Latin hypercube calculation implementation lhs.

[0020] In one embodiment, as shown in Figure 2 , an intelligent experiment design method based on a distributed network architecture is provided. Taking the framework structure in Figure 1 as an example, the method includes the following steps:

[0021] Step 202, load historical dialogue records, input files of experiment design algorithms and user request information.

[0022] Specifically, the system first loads the input information, accepts the user's request, loads the historical dialogue records, and can load the input files of the experiment design method and some parameters (such as dimension, sample size, etc.) customized by the user.

[0023] Step 204, in the back end middleware layer, verify the legality of the input file. If the legality of the input file is abnormal, the historical dialogue records and the user request information are used as a training sample data set, the large model of the business layer is triggered to perform decision training on the training sample data set according to the input file, an experiment design strategy is output, and a dialogue is output to the front end interaction layer. Otherwise, the input parameters in the input file are reconstructed, the calculation core of the experiment design algorithm is called to start an independent thread, so that the large model performs a calculation task and outputs a calculation result.

[0024] Specifically, check whether the loaded file is complete, including checking the parameter file, parameter requirements, whether the values ​​are reasonable, reference standards and other related information.

[0025] Furthermore, if the parameter file fails the validity check, for example, if the parameter file handle does not exist, the system triggers a large model prompt to the front-end. Alternatively, if the parameter file handle exists but the parameters are unreasonable or illegal, the system also prompts the front-end with a natural language prompt stating: "Input parameters not detected, unreasonable or illegal." The user is asked to provide relevant information and explain the input parameters in conjunction with a project case. The conversation is then saved and returned.

[0026] Furthermore, if the parameter file is legal, the system will reconstruct the input parameters read from the file, including conversion types, splicing and packaging, etc. After reading is completed, the independent thread is started by calling the calculation core of the experimental design method to perform the calculation task. The present invention supports multiple experimental design methods, including but not limited to Latin hypercube sampling, factorial design, central composite design, etc. The following is a detailed introduction to the embodiment:

[0027] Example 1: Experimental Design Method for Latin Hypercube Sampling

[0028] Latin Hypercube Sampling (LHS) is a statistical method for generating approximately random samples of parameter values ​​from a multidimensional distribution. This method is often used to construct computer experiments or Monte Carlo integration. The following is its mathematical formula:

[0029] Assume that n samples need to be extracted from d-dimensional space, and the value range of each dimension is normalized to [0,1). For each sample and each dimension , and its coordinate calculation formula is:

[0030] ;

[0031] in, Indicates the dimensional random permutation function, Randomly shuffle the order, represents independent uniform random numbers, .

[0032] Example 2: Execution of a central composite experimental design method

[0033] Central Composite Design (CCD) is a commonly used experimental design method, mainly used in Response Surface Methodology (RSM) to optimize processes. CCD mainly includes the following parts: factor points (standard operating points in full factor or partial factor design, usually at high and low levels of each factor), center points (repeated runs with all factors set at intermediate values, helpful for estimating pure error and detecting curvature), star points (extending outward from the center point along each factor axis, distance determined by determined, which can be selected for orthogonality, rotatability or other characteristics).

[0034] For CCD, some key formulas are as follows:

[0035] Total number of experiments (N):The total number of experiments can be calculated using the following formula:

[0036] ;

[0037] where, is the number of center points, is the number of factors.

[0038] Value: used to determine the position of the star point, different strategies can be chosen to determine value, such as rotatability, then value can be calculated by the following formula:

[0039] ;

[0040] where is the number of factor points.

[0041] Second-order regression model formula:

[0042] The relationship between response variable y and k factors:

[0043] ;

[0044] where, represents the constant term, represents the linear effect coefficient, represents the quadratic effect coefficient, represents the interaction effect coefficient, represents the random error.

[0045] Example three, perform factorial experimental design method

[0046] Factorial Design is a statistical experimental design method used to study how a response variable is affected by one or more factors, each factor having at least two levels in the experiment. For a completely randomized factorial design, if there are factors, each factor has levels, then the total number of experimental runs can be calculated by the following formula:

[0047] ;

[0048] Each experimental combination corresponds to a response value , and the following model can be established:

[0049] ;

[0050] where represents the overall average response, represents the main effect of the th factor, represents the interaction effect of the th and th factors, represents the coding value of the th factor (usually -1 or 1), represents the coding value of the th factor, represents the coding value of the th factor, represents the interaction effect of the th and th and th factors, represents the experimental error.

[0051] During the calculation process, the main thread is always interacting with the front end, and the front end displays a progress bar and pushes the task status in real time (e.g., "calculating, 30% completed").

[0052] Further, during the waiting process for the completion of the calculation, the system starts a new thread to support the active push of interactive messages, such as the application of a certain experimental design method in a certain evaluation, and the main functions of the system can also introduce to the user, and the interactive messages also support the user's active questioning. After the calculation is completed, the user can stop the task through the "stop" button, and the system destroys the corresponding thread and releases the resources.

[0053] Step 206, load the calculation results into the large model dialogue, and output the analysis report to the front-end interaction layer.

[0054] Specifically, after the calculation is completed, the system will accept the original result data and convert it into a standard format for storage, ensuring compatibility and analyzability. At the same time, the data is sent to the large model, triggering the result explanation module on the large model side, based on historical data, reasonable analysis of the calculation results combined with the engineering background, and reasonable application and suggestions for the result data, generating an analysis report and sending it to the user side. At the same time, the user side can be visualized, by drawing scientific calculation charts, scatter plots, and layered scatter plots, to help users intuitively evaluate the uniformity of sample distribution.

[0055] Further, after all operations and calculations are completed, the original data, visual charts, and large model analysis reports can be packaged into standardized files (such as ZIP compression packages) and stored in the local system. The front-end interface provides a download link, and users can obtain the file package containing complete results with one click, supporting multiple formats, while pushing notification messages (such as "task completed, click here to download").

[0056] Further, if the calculation core is interrupted due to insufficient resources or logical errors, the system automatically records error logs, triggers the large model to generate repair suggestions (such as "sample number exceeds memory limit, suggest batch calculation"), and notifies the user to resubmit the task. Regardless of the success or failure of the task, the system automatically releases the occupied computing resources, ensuring efficient use of resources.

[0057] In the above intelligent test design method based on a distributed network architecture, the design of the function loading middleware adopts a modular architecture, which groups test design algorithms, large model decision-making, and other functional components into dynamically loadable modules. Without modifying the core code, new methods can be quickly integrated through plug-in form, significantly reducing development cost and cycle, and achieving flexible adaptation to multiple scene requirements. The whole-process automation mechanism runs through input file legality verification, training sample generation, large model decision-making training, calculation task execution, and analysis report output, forming a closed loop of "design-verification-improvement", avoiding efficiency loss caused by manual intervention. The intelligent interaction and decision support driven by the large model can automatically generate training samples based on historical dialogue records and user request information when the input file is legal, drive the large model to perform decision-making training and output strategies; when the input is legal, it starts an independent thread to call the algorithm calculation core, realizes efficient execution of the large model on the calculation task, and combines with the multi-task flexible scheduling mechanism to effectively improve the test efficiency and reduce the calculation cost. This design not only enables the system to quickly adapt to different industry requirements, but also provides standardized interfaces for the expansion of new methods and new scenarios through the function loading characteristics of the middleware, realizes the unity of flexibility of the technical architecture and adaptability of business requirements, and improves the development efficiency and universality of virtual calculation tests.

[0058] In one of the embodiments, in the front-end interactive layer, according to the input information of the user in the system interface, the historical dialogue record, the input file of the experimental design algorithm and the user request information are loaded.

[0059] In one of the embodiments, in the back-end middleware layer, the input file is subjected to legality verification according to the preset verification condition, and the specific steps are as follows: if the parameter file handle does not exist in the input file, the legality of the input file is abnormal. If the parameter file handle exists and the parameter is abnormal, the legality of the input file is abnormal. Except for the above two abnormal cases, the legality of the input file is normal.

[0060] In one of the embodiments, if the legality of the input file is abnormal, the historical dialogue record and the user request information are taken as the training sample data set, the large model of the business layer is triggered to perform decision training on the training sample data set according to the input file, and the experimental design strategy is output. After saving the current dialogue, the current dialogue is output to the front-end interactive layer to generate natural language to prompt the user to adjust the input information of the experimental design.

[0061] In one of the embodiments, if the legality of the input file is normal, the input parameters in the input file are reconstructed, the calculation core of the experimental design algorithm is called to start an independent thread, and in the independent thread, the large model of the business layer is triggered to take the reconstructed input parameters as the training sample to execute the calculation task corresponding to the experimental design algorithm, and the calculation result is output.

[0062] It is worth noting that from the parameter file loading, the legality verification, to the calculation task submission, the result analysis and the visualization display, the whole process is automated without human intervention. The system realizes one-key operation through middleware scheduling and algorithm encapsulation, improves the experimental efficiency, reduces the calculation cost, and avoids the efficiency bottleneck caused by the traditional dependence on manual experience.

[0063] In one of the embodiments, during the waiting process of the calculation completion, an independent new thread is started, in the independent new thread, the application of the experimental design algorithm in the engineering test and the platform function are generated into visual natural language through the front-end interactive layer, and are displayed to the user, so that the user can control the dialogue output of the independent new thread through the function button. If the user selects the stop function button, the dialogue output is terminated and the independent new thread is destroyed.

[0064] It is worth noting that the intelligent interaction and decision support driven by the large model will continuously iterate based on the historical data and user feedback to generate parameter optimization strategies and result explanation reports, and truly realize the automation of the closed loop of "design-verification-improvement". At the same time, the large model is deeply embedded in the experimental design process to actively provide natural language prompts at key nodes such as parameter input, reduce the learning threshold, and facilitate user use.

[0065] In one embodiment, the calculation results are converted into a standard format and stored in the local system.

[0066] In one embodiment, if the calculation core is interrupted, a large model is generated to generate a repair suggestion, and a dialogue is output to the front-end interaction layer to prompt the user to resubmit the calculation task.

[0067] In one embodiment, as shown in the system function flowchart, the specific steps are as follows: Figure 3

[0068] 1. Start: Load historical dialogue, Latin hypercube parameter input file, user request, etc.

[0069] 2. Parameter file legality verification:

[0070] 2.1 Verify illegal, parameter file handle does not exist, prompt to the front end: no input parameter detected; or the parameter file handle exists, prompt to the front end: parameter is unreasonable or illegal; the prompt is fed back to the large model to generate natural language, and the dialogue is output to the front end, including the following information: what kind of parameter file is needed, parameter number requirement, reasonable value range of each parameter, reference standard, and engineering case explanation input parameter. Save the dialogue and return the program.

[0071] 2.2 Verify legal, reconstruct the parameters read from the file, including type conversion, splicing and packaging, etc.

[0072] 2.3 Call Latin hypercube calculation core lhs: start a new thread, trigger large model: while waiting for calculation, the front end introduces the application of Latin hypercube in engineering design of experiment and the function of the platform to the user, and the thread is destroyed when the dialogue is completed or the user uses the stop button to stop the dialogue.

[0073] 3. Calculate the results and accept the result data.

[0074] 4. Draw scientific calculation charts, including scatter plots, layered scatter plots, etc.: load the calculation results into the large model dialogue and output the dialogue to the user, including: typical raw data table of calculation results, calculation result rationality analysis combined with engineering background, how to apply result data, drawing display suggestion, etc. And show the scatter plot examples generated in the background. Then, save all the raw data and charts of the calculation results to the file, and provide a front-end download link to prompt the user and return the program.

[0075] It should be understood that, although Figures 2-3 ​The steps in the flowchart are shown in sequence as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified in this document, there is no strict order restriction for the execution of these steps, and these steps can be executed in other orders. In addition, Figures 2-3 At least part of the steps may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily executed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed in turn or alternately with other steps or at least part of the sub-steps or stages of other steps.

[0076] In one embodiment, Figure 4 As shown, an intelligent experiment design system based on a distributed network architecture is provided, including: an information loading module 402, an experiment execution module 404 and a report display module 406, wherein:

[0077] The information loading module 402 is used to load historical conversation records, input files of the experimental design algorithm, and user request information.

[0078] Experiment execution module 404 verifies the validity of input files at the backend middleware layer. If the input files are not valid, it uses historical conversation records and user request information as a training sample dataset, triggering the business layer's large model to perform decision training on the training sample dataset based on the input files, output the experiment design strategy, and transmit the conversation to the frontend interaction layer. Otherwise, it reconstructs the input parameters in the input file, calls the computational core of the experiment design algorithm, and starts an independent thread, allowing the large model to execute the computation task and output the results.

[0079] The report display module 406 is used to load the calculation results into the large model dialogue and then output the analysis report to the front-end interaction layer.

[0080] The specific definition of the intelligent experiment design system based on a distributed network architecture can be found in the definition of the intelligent experiment design method based on a distributed network architecture above, and will not be repeated here. The various modules in the above-mentioned intelligent experiment design system based on a distributed network architecture can be implemented in whole or in part through software, hardware, and a combination thereof. The above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.

[0081] Those skilled in the art will understand that Figure 4The structure shown in the figure is only a block diagram of part of the structure related to the scheme of the present application, and does not constitute a limitation on the computer device to which the scheme of the present application is applied. The specific computer device can include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0082] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer readable storage medium, and when executed, can include the processes of the above-mentioned embodiment methods. Any reference to memory, storage, database or other medium used in each embodiment of the present application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration but not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0083] Each technical feature of the above embodiments can be combined arbitrarily. In order to make the description concise, all possible combinations of each technical feature in the above embodiments are not described, but as long as the combination of the technical features does not exist contradictory, it should be considered as the scope of the present application.

[0084] The above embodiments only express several implementation manners of the present application, and the description is more specific and detailed, but it should not be understood as a limitation on the scope of the application. It should be pointed out that for those skilled in the art, without departing from the concept of the present application, a number of modifications and improvements can be made, which are within the scope of the present application. Therefore, the protection scope of the present application should be subject to the appended claims.

Claims

1. An intelligent experiment design method for a distributed network architecture, characterized in that: Applied to a distributed network architecture, the distributed network architecture includes a front-end interaction layer, a back-end middleware layer, and a business layer, the method includes: Load historical conversation records, experimental design algorithm input files, and user request information; At the back-end middleware layer, the validity of the input file is checked. If the validity of the input file is abnormal, the historical conversation records and the user request information are used as a training sample data set, triggering the large model of the business layer to perform decision training on the training sample data set based on the input file, output the experimental design strategy, and spit out the conversation to the front-end interaction layer; otherwise, the input parameters in the input file are reconstructed, and the computing core of the experimental design algorithm is called to start an independent thread, so that the large model executes the computing task and outputs the computing results; In the back-end middleware layer, the input file is subjected to a validity check according to preset verification conditions, specifically: if the parameter file handle in the input file does not exist, the validity of the input file is abnormal; if the parameter file handle exists and the parameters are abnormal, the validity of the input file is abnormal; except for the above two abnormal situations, the validity of the input file is normal; After the calculation results are loaded into the large model dialogue, an analysis report is output to the front-end interaction layer.

2. The method according to claim 1, characterized in that Loads historical session records, input files for the DOE algorithm, and user request information, including: In the front-end interaction layer, historical conversation records, input files of the experimental design algorithm, and user request information are loaded according to the user input information in the system interface.

3. The method according to claim 1, characterized in that If the legality of the input file is abnormal, the historical conversation record and the user request information are used as a training sample data set, triggering the large model of the business layer to perform decision training on the training sample data set based on the input file, output the experimental design strategy, and spit out the conversation to the front-end interaction layer, including: If the legality of the input file is abnormal, the historical conversation records and the user request information are used as a training sample data set, triggering the large model of the business layer to perform decision training on the training sample data set based on the input file and output an experimental design strategy; After saving this dialogue, the dialogue is sent to the front-end interaction layer to generate natural language to prompt the user to adjust the input information of the experimental design.

4. The method according to claim 3, characterized in that If the legality of the input file is normal, reconstruct the input parameters in the input file, call the computation core of the experimental design algorithm to start an independent thread, so that the large model performs the computation task and outputs the computation result, including: If the legality of the input file is normal, reconstruct the input parameters in the input file, call the computing core of the experimental design algorithm to start an independent thread, and in the independent thread, trigger the large model of the business layer to use the reconstructed input parameters as training samples, execute the computing tasks corresponding to the experimental design algorithm, and output the calculation results.

5. The method according to claim 4, characterized in that Before the step of outputting the calculation results, it also includes: While waiting for the calculation to be completed, an independent new thread is started. In the independent new thread, the application of the experimental design algorithm in the engineering operation experiment and the platform function are generated into a visual natural language through the front-end interaction layer and displayed to the user, so that the user can control the dialogue output of the independent new thread through the function button. If the user selects the stop function button, the dialogue output is terminated and the independent new thread is destroyed.

6. The method according to claim 5, characterized in that After the calculation results are loaded into the large model dialogue, and before the step of outputting the analysis report to the front-end interaction layer, the method further includes: The calculation results are converted into a standard format and stored in a local system.

7. The method according to claim 6, characterized in that After the calculation results are loaded into the large model dialogue and the analysis report is output to the front-end interaction layer, the method further includes: If the computing core is interrupted, the large model is triggered to generate repair suggestions and spit out a dialogue to the front-end interaction layer, prompting the user to resubmit the computing task.

8. An intelligent experiment design system based on a distributed network architecture, characterized in that: The system comprises: Information loading module, used to load historical conversation records, input files of experimental design algorithms and user request information; The test execution module is used to verify the legality of the input file at the back-end middleware layer. If the legality of the input file is abnormal, the historical conversation record and the user request information are used as a training sample data set to trigger the large model of the business layer to perform decision training on the training sample data set according to the input file, output the test design strategy, and spit out the conversation to the front-end interaction layer; otherwise, the input parameters in the input file are reconstructed, and the computing core of the test design algorithm is called to start an independent thread so that the large model executes the computing task and outputs the computing result; at the back-end middleware layer, the input file is subjected to a legality check according to preset inspection conditions, and the specific steps are as follows: if the parameter file handle in the input file does not exist, the legality of the input file is abnormal; if the parameter file handle exists and the parameters are abnormal, the legality of the input file is abnormal; except for the above two abnormal situations, the legality of the input file is normal; The report display module is used to load the calculation results into the large model dialogue and then output the analysis report to the front-end interaction layer.

Images