Low-code data docking detection method based on artificial intelligence

Abstract

The invention discloses a low-code data docking detection method based on artificial intelligence. The method comprises the steps of component arrangement, template testing, establishment of a data docking detection model, task docking scheduling and updating feedback. The invention belongs to the technical field of data engineering, and particularly relates to a low-code data docking detection method based on artificial intelligence, and the method comprises the steps: inputting analog data, completing data circulation through docking analog logic, prompting an error level through a background color, and guaranteeing the reliability of data docking; the sample neighborhood and the neighborhood granularity are defined, the local range and the fine degree of the data are determined, the local characteristics of the data can be better grasped, the characteristic similarity is measured, and the accuracy of a data docking detection model is jointly improved; a task allocation scheme for finding the minimum overall delay is adopted, reasonable allocation is performed by considering the performance of an accelerator and task characteristics, and tasks are divided according to layers and execution positions are reasonably planned to ensure normal task scheduling.

Description

Technical Field

[0001] The present invention belongs to the field of data engineering technology, and specifically is a low-code data docking detection method based on artificial intelligence. Background Art

[0002] Low-code data docking detection is a method that uses artificial intelligence technology to monitor and analyze the data docking process in a low-code environment, aiming to speed up software development, reduce traditional programming work, and minimize the workload of manual coding. However, the existing low-code data docking detection has technical problems such as lengthy and complicated template testing, extremely low reuse, and inflexible configuration; inaccurate data docking detection, resulting in unnecessary repeated operations and a waste of time and resources; inefficient data docking task allocation, inefficient change response, and inability to achieve effective task management and monitoring. Summary of the invention

[0003] In view of the above situation, in order to overcome the defects of the prior art, the present invention provides a low-code data docking detection method based on artificial intelligence. In view of the technical problems that the template testing time is lengthy and complicated, the reuse degree is extremely low, and the configuration is not flexible enough, the simulated data is input and the data flow is completed through the docking simulation logic, and the error level is prompted by the background color to ensure the reliability of data docking; in view of the technical problems that the data docking detection is not accurate enough, resulting in unnecessary repeated operations and waste of time and resources, the sample neighborhood and neighborhood granularity are defined, the local range and precision of the data are clarified, which helps to better grasp the local characteristics of the data, measure the feature similarity, and deeply reveal the relationship between the data features. The attention mechanism layer is introduced to mark the isomorphic and heterogeneous features, and the accuracy of the data docking detection model is jointly improved; in view of the technical problems that the data docking task allocation is not efficient enough, the change response is not agile enough, and effective task management and monitoring cannot be achieved, the task allocation scheme is adopted by finding a task allocation scheme that minimizes the overall delay, and the performance of the accelerator and the characteristics of the task are reasonably allocated. The tasks are divided into layers and the execution positions are reasonably planned. The task queuing information is visually displayed, and the disabling and calling of tasks are supported to ensure the normal progress of task scheduling.

[0004] The technical solution adopted by the present invention is as follows: The present invention provides a low-code data docking detection method based on artificial intelligence, which comprises the following steps:

[0005] Step S1: component orchestration, specifically generating a docking template and completing node configuration to perform component orchestration;

[0006] Step S2: template test, specifically obtaining simulation data for template test, and indicating the error level through background color;

[0007] Step S3: Establish a data docking detection model, specifically define sample neighborhoods and neighborhood granularity, measure feature similarity, use local gram matrix modeling to obtain the relationship between data features, and finally introduce an attention mechanism layer to transfer feature similarity and relationship, mark isomorphic and heterogeneous features, and realize the establishment of a data docking detection model;

[0008] Step S4: Task docking scheduling, specifically to clarify the task scheduling goal is to find a task allocation plan that minimizes the overall delay, calculate the delay when the accelerator allocates tasks, allocate tasks in sequence, divide the CNN task into subtasks, plan the execution position of each layer, and then configure the task scheduling rules and record logs;

[0009] Step S5: Update feedback.

[0010] Furthermore, in step S1, the component arrangement includes the following steps:

[0011] Step S11: Generate a docking template, drag the component into the workbench through visual interaction, and use it as the basic unit of scheduling in the docking template;

[0012] Step S12: Complete node configuration, click on the component, and complete the node configuration according to the interface configuration items. The node includes a data reading node, a data cleaning node, and a data writing node;

[0013] The data reading node is mainly responsible for selecting the data source and setting parameters, reading the data entry of the data source, and verifying whether the data reading is normal through preview testing to ensure that the data can be accessed smoothly;

[0014] The data cleaning node mainly pre-processes, screens and filters the data source data according to the data storage requirements, selects appropriate scripts, verification, statistics, and conversion components, completes the configuration work, and achieves the purpose of cleaning and optimization;

[0015] The data writing node selects the appropriate input component to load the processed data into the target business system to achieve the final writing and application of the data;

[0016] Step S13: Perform orchestration. The low-code platform sets the execution order of components through visual operations, and determines the flow direction of subsequent steps based on the execution results, thereby achieving the orchestration of the relationships and processes between components. It also supports control over whether components are executed concurrently or linearly, and flexibly constructs business flows and logical architectures that meet various needs.

[0017] Further, in step S2, the template test includes the following steps:

[0018] Step S21: Get simulation data, get some slice data obtained by the node test preview, configure corresponding generation rules according to own needs, create some simulation data, and let users customize data according to actual conditions and specific requirements, so as to better carry out subsequent analysis and testing, which increases the flexibility and customizability of data processing;

[0019] Step S22: Performing a test, including the following steps:

[0020] Step S221: Create a test instance and select simulation data, and enable the template test instance;

[0021] Step S222: According to the operation result, the visualization component on the interface will accurately indicate the step where the data flow error occurs through the background color, and the prompt information is divided into three levels: green normal, yellow warning, and red error;

[0022] The yellow warning indicates that an error may occur when data with certain characteristics is transferred here and the execution cannot continue;

[0023] The red error indicates that the current step configuration is unreliable and has caused serious problems in the simulation data, resulting in process interruption;

[0024] Step S23: Modify the template. Based on the modification suggestions provided by this test, the user can directly adjust the template in the test instance until the template logic is normal.

[0025] Furthermore, in step S3, the establishment of the data docking detection model includes the following steps:

[0026] Step S31: define the sample neighborhood, the neighborhood granularity of the sample on the feature set is defined, and the formula used is as follows:

[0027] ;

[0028] In the formula, Representation sample neighborhood, U1 is the set containing all features, x represents the sample point in the feature set, θ is the threshold of the neighborhood granularity, represents the distance metric, Represents the sample point x and the sample The distance measure between them;

[0029] Step S32: measure feature similarity, model the relationship according to the neighborhood granularity of the sample, and measure feature similarity by calculating the absolute value of the elements in the local gram matrix. The formula used is as follows:

[0030] ;

[0031] In the formula, represents the element in the i-th row and j-th column of the local gram matrix, represents the i-th eigenvector in the sample neighborhood granularity, represents the transpose of the jth eigenvector in the sample neighborhood granularity, The larger this absolute value is, the higher the feature similarity is;

[0032] Step S33: Use the local gram matrix to model and obtain the relationship between data features. The formula used is as follows:

[0033] ;

[0034] ;

[0035] In the formula, n represents the number of neighborhood granularities, g represents the index of the specific neighborhood granularity, is the jth row of the coefficient matrix, represents the i-th row of the coefficient matrix, N represents the number of rows of the coefficient matrix, that is, the number dimension of elements involved in the calculation, represents the Laplacian matrix of the neighborhood granularity g, It is a diagonal matrix with a neighborhood granularity of g, which is used to encode the similarity between pairs of features, and tr() represents the trace of the matrix;

[0036] Step S34: Introduce the attention mechanism layer, pass the relationship between feature similarity and data features to the attention mechanism layer, assign different attention weights to each feature, mark isomorphic and heterogeneous features according to the similarity threshold, evaluate the importance and weight the heterogeneous features, and then realize the establishment of the data docking detection model.

[0037] Furthermore, in step S4, the task docking scheduling includes the following steps:

[0038] Step S41: clarify the task scheduling goal and find the task allocation scheme that minimizes the overall delay, that is, find the allocation method that minimizes the maximum value of the sum of delays in all computing nodes. The formula used is as follows:

[0039] ;

[0040] In the formula, U represents the overall goal of minimization, that is, minimizing the overall delay, k represents different computing nodes, represents the set of tasks assigned to the kth computing node, Indicates a specific task. Indicates different tasks, Indicates execution of tasks Assign to task Delay time on

[0041] Step S42: Calculate the delay when the accelerator assigns tasks, sort the CNN tasks from large to small, traverse the sorted task list, and for each task, calculate the delay when assigning the task to each accelerator in turn, while considering the existing task queues on each accelerator, select the accelerator with the smallest delay and the smallest impact on the overall performance for assignment, and repeat the steps until all tasks are assigned;

[0042] Step S43: Divide the CNN task into subtasks by layer, determine the characteristics and requirements of each subtask, evaluate the performance of each accelerator and the task type suitable for processing, and select the most suitable accelerator for each subtask. When processing two consecutive layers on the same accelerator, the output of the previous layer is directly placed in the buffer for use by the next layer; if two adjacent layers are assigned to different accelerators, the output of the previous layer is placed in the backup buffer, and another accelerator is allowed to load data from the backup buffer. The access costs of the two rounds of the backup buffer are calculated and recorded.

[0043] Step S44: Plan the execution location of each layer, and only assign the convolutional layer to different accelerators. When a specific convolutional layer is assigned to an accelerator, the subsequent pooling layers will also be executed on the same accelerator, and the fully connected layer will be arranged to be executed on the same accelerator as the last convolutional layer;

[0044] Step S45: Configure the task scheduling rules. The user can specify that the task is executed at fixed minute intervals or hour intervals, and accurately specify the specific time point of each day to start the task, so that the user can make reasonable arrangements according to actual needs and business scenarios;

[0045] The scheduler will deliver tasks to the queue according to the task scheduling rules, and the consumer thread will wake up and execute tasks in the order of queue delivery. It supports disabling tasks during the scheduling process. When the task is called up again, it will no longer be scheduled due to the disabled state, completing the scheduling unloading. The task queue information will also be provided to the user in a visual display. If there is an abnormal execution of a task, it can be manually kicked out of the queue according to the actual situation to ensure that subsequent tasks can be scheduled normally. This part of the operation can also be automatically monitored and executed by the scheduler by configuring the task execution timeout.

[0046] Step S46: Record logs. Through task and node listeners, record task execution logs and node data, and store them persistently for use in subsequent troubleshooting.

[0047] Furthermore, in step S5, the update feedback is specifically real-time monitoring of template usage, task docking scheduling and the running effect of the detection model, and the relevant data and information collected based on the monitoring results are fed back to the staff so as to make timely adjustments and optimizations.

[0048] The beneficial results achieved by the present invention using the above scheme are as follows:

[0049] (1) In order to solve the technical problems of lengthy and complicated template testing, extremely low reuse, and inflexible configuration, the simulation data is input and connected to the simulation logic to complete the data flow, and the error level is indicated by the background color to ensure the reliability of data connection;

[0050] (2) To address the technical problem of inaccurate data docking detection, which leads to unnecessary repeated operations and waste of time and resources, we define sample neighborhoods and neighborhood granularity to clarify the local scope and degree of refinement of the data, which helps to better grasp the local characteristics of the data, measure feature similarity, and deeply reveal the relationship between data features. We also introduce an attention mechanism layer to mark homogeneous and heterogeneous features, which can jointly improve the accuracy of the data docking detection model.

[0051] (3) To address the technical issues of inefficient data docking task allocation, slow response to changes, and inability to achieve effective task management and monitoring, we adopt a task allocation solution that seeks to minimize overall delay, considers the performance of the accelerator and the characteristics of the task, divides the tasks into layers and reasonably plans the execution locations, and ensures the normal progress of task scheduling by visually displaying task queuing information and supporting the disabling and calling up of tasks. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 A flowchart of a low-code data docking detection method based on artificial intelligence provided by the present invention;

[0053] Figure 2 is a schematic flow chart of step S2;

[0054] Figure 3 is a schematic flow chart of step S3;

[0055] Figure 4 It is a schematic diagram of the process of step S4.

[0056] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention. DETAILED DESCRIPTION

[0057] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0058] In the description of the present invention, it should be understood that terms such as “upper”, “lower”, “front”, “back”, “left”, “right”, “top”, “bottom”, “inside” and “outside” indicating directions or positional relationships are based on the directions or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific direction, be constructed and operated in a specific direction, and therefore should not be understood as limiting the present invention.

[0059] Example 1, see Figure 1 The present invention provides a low-code data docking detection method based on artificial intelligence, which includes the following steps:

[0060] Step S1: component orchestration, specifically generating a docking template and completing node configuration to perform component orchestration;

[0061] Step S2: template test, specifically obtaining simulation data for template test, and indicating the error level through background color;

[0062] Step S3: Establish a data docking detection model, specifically define sample neighborhoods and neighborhood granularity, measure feature similarity, use local gram matrix modeling to obtain the relationship between data features, and finally introduce an attention mechanism layer to transfer feature similarity and relationship, mark isomorphic and heterogeneous features, and realize the establishment of a data docking detection model;

[0063] Step S4: Task docking scheduling, specifically to clarify the task scheduling goal is to find a task allocation plan that minimizes the overall delay, calculate the delay when the accelerator allocates tasks, allocate tasks in sequence, divide the CNN task into subtasks, plan the execution position of each layer, and then configure the task scheduling rules and record logs;

[0064] Step S5: Update feedback.

[0065] Example 2, see Figure 1 This embodiment is based on the above embodiment. In step S1, the component arrangement includes the following steps:

[0066] Step S11: Generate a docking template, drag the component into the workbench through visual interaction, and use it as the basic unit of scheduling in the docking template;

[0067] Step S12: Complete node configuration, click on the component, and complete the node configuration according to the interface configuration items. The node includes a data reading node, a data cleaning node, and a data writing node;

[0068] The data reading node is mainly responsible for selecting the data source and setting parameters, reading the data entry of the data source, and verifying whether the data reading is normal through preview testing to ensure that the data can be accessed smoothly;

[0069] The data cleaning node mainly pre-processes, screens and filters the data source data according to the data storage requirements, selects appropriate scripts, verification, statistics, and conversion components, completes the configuration work, and achieves the purpose of cleaning and optimization;

[0070] The data writing node selects the appropriate input component to load the processed data into the target business system to achieve the final writing and application of the data;

[0071] Step S13: Perform orchestration. The low-code platform sets the execution order of components through visual operations, and determines the flow direction of subsequent steps based on the execution results, thereby achieving the orchestration of the relationships and processes between components. It also supports control over whether components are executed concurrently or linearly, and flexibly constructs business flows and logical architectures that meet various needs.

[0072] Example 3, see Figure 1 and Figure 2 This embodiment is based on the above embodiment. In step S2, the template test includes the following steps:

[0073] Step S21: Get simulation data, get some slice data obtained by the node test preview, configure corresponding generation rules according to own needs, create some simulation data, and let users customize data according to actual conditions and specific requirements, so as to better carry out subsequent analysis and testing, which increases the flexibility and customizability of data processing;

[0074] Step S22: Performing a test, including the following steps:

[0075] Step S221: Create a test instance and select simulation data, and enable the template test instance;

[0076] Step S222: According to the operation result, the visualization component on the interface will accurately indicate the step where the data flow error occurs through the background color, and the prompt information is divided into three levels: green normal, yellow warning, and red error;

[0077] The yellow warning indicates that an error may occur when data with certain characteristics is transferred here and the execution cannot continue;

[0078] The red error indicates that the current step configuration is unreliable and has caused serious problems in the simulation data, resulting in process interruption;

[0079] Step S23: Modify the template. Based on the modification suggestions provided by this test, the user can directly adjust the template in the test instance until the template logic is normal.

[0080] By executing the above operations, the simulation data is input and the data flow is completed through the docking simulation logic. The error level is prompted by the background color to ensure the reliability of data docking. This solves the technical problems of lengthy and complicated template testing, extremely low degree of reuse, and inflexible configuration.

[0081] Example 4, see Figure 1 and Figure 3 This embodiment is based on the above embodiment. In step S3, the establishment of the data docking detection model includes the following steps:

[0082] Step S31: define the sample neighborhood, the neighborhood granularity of the sample on the feature set is defined, and the formula used is as follows:

[0083] ;

[0084] In the formula, Representation sample neighborhood, U1 is the set containing all features, x represents the sample point in the feature set, θ is the threshold of the neighborhood granularity, represents the distance metric, Represents the sample point x and the sample The distance measure between them;

[0085] Step S32: measure feature similarity, model the relationship according to the neighborhood granularity of the sample, and measure feature similarity by calculating the absolute value of the elements in the local gram matrix. The formula used is as follows:

[0086] ;

[0087] In the formula, represents the element in the i-th row and j-th column of the local gram matrix, represents the i-th eigenvector in the sample neighborhood granularity, represents the transpose of the jth eigenvector in the sample neighborhood granularity, The larger this absolute value is, the higher the feature similarity is;

[0088] Step S33: Use the local gram matrix to model and obtain the relationship between data features. The formula used is as follows:

[0089] ;

[0090] ;

[0091] In the formula, n represents the number of neighborhood granularities, g represents the index of the specific neighborhood granularity, is the jth row of the coefficient matrix, represents the i-th row of the coefficient matrix, N represents the number of rows of the coefficient matrix, that is, the number dimension of elements involved in the calculation, represents the Laplacian matrix of the neighborhood granularity g, It is a diagonal matrix with a neighborhood granularity of g, which is used to encode the similarity between pairs of features, and tr() represents the trace of the matrix;

[0092] Step S34: Introduce the attention mechanism layer, pass the relationship between feature similarity and data features to the attention mechanism layer, assign different attention weights to each feature, mark isomorphic and heterogeneous features according to the similarity threshold, evaluate the importance and weight the heterogeneous features, and then realize the establishment of the data docking detection model.

[0093] By performing the above operations, defining sample neighborhoods and neighborhood granularity, clarifying the local scope and degree of refinement of the data, it helps to better grasp the local characteristics of the data, measure feature similarities, and deeply reveal the relationship between data features. The attention mechanism layer is introduced, and isomorphic and heterogeneous features are marked, which jointly improves the accuracy of the data docking detection model and solves the technical problem that data docking detection is not accurate enough, resulting in unnecessary repeated operations and waste of time and resources.

[0094] Example 5, see Figure 1 and Figure 4 This embodiment is based on the above embodiment. In step S4, the task docking scheduling includes the following steps:

[0095] Step S41: clarify the task scheduling goal and find the task allocation scheme that minimizes the overall delay, that is, find the allocation method that minimizes the maximum value of the sum of delays in all computing nodes. The formula used is as follows:

[0096] ;

[0097] In the formula, U represents the overall goal of minimization, that is, minimizing the overall delay, k represents different computing nodes, represents the set of tasks assigned to the kth computing node, Indicates a specific task. Indicates different tasks, Indicates execution of tasks Assign to task Delay time on

[0098] Step S42: Calculate the delay when the accelerator assigns tasks, sort the CNN tasks from large to small, traverse the sorted task list, and for each task, calculate the delay when assigning the task to each accelerator in turn, while considering the existing task queues on each accelerator, select the accelerator with the smallest delay and the smallest impact on the overall performance for assignment, and repeat the steps until all tasks are assigned;

[0099] Step S43: Divide the CNN task into subtasks by layer, determine the characteristics and requirements of each subtask, evaluate the performance of each accelerator and the task type suitable for processing, and select the most suitable accelerator for each subtask. When processing two consecutive layers on the same accelerator, the output of the previous layer is directly placed in the buffer for use by the next layer; if two adjacent layers are assigned to different accelerators, the output of the previous layer is placed in the backup buffer, and another accelerator is allowed to load data from the backup buffer. The access costs of the two rounds of the backup buffer are calculated and recorded.

[0100] Step S44: Plan the execution location of each layer, and only assign the convolutional layer to different accelerators. When a specific convolutional layer is assigned to an accelerator, the subsequent pooling layers will also be executed on the same accelerator, and the fully connected layer will be arranged to be executed on the same accelerator as the last convolutional layer;

[0101] Step S45: Configure the task scheduling rules. The user can specify that the task is executed at fixed minute intervals or hour intervals, and accurately specify the specific time point of each day to start the task, so that the user can make reasonable arrangements according to actual needs and business scenarios;

[0102] The scheduler will deliver tasks to the queue according to the task scheduling rules, and the consumer thread will wake up and execute tasks in the order of queue delivery. It supports disabling tasks during the scheduling process. When the task is called up again, it will no longer be scheduled due to the disabled state, completing the scheduling unloading. The task queue information will also be provided to the user in a visual display. If there is an abnormal execution of a task, it can be manually kicked out of the queue according to the actual situation to ensure that subsequent tasks can be scheduled normally. This part of the operation can also be automatically monitored and executed by the scheduler by configuring the task execution timeout.

[0103] Step S46: Record logs. Through task and node listeners, record task execution logs and node data, and store them persistently for use in subsequent troubleshooting.

[0104] By executing the above operations, a task allocation scheme that minimizes the overall delay is adopted, and reasonable allocation is made considering the performance of the accelerator and the characteristics of the task. Tasks are divided into layers and the execution locations are reasonably planned. The task queuing information is visually displayed, and disabling and calling up tasks are supported to ensure the normal progress of task scheduling. This solves the technical problems of inefficient data docking task allocation, insensitive change response, and inability to achieve effective task management and monitoring.

[0105] Example 6, see Figure 1 This embodiment is based on the above embodiment. In step S5, the update feedback is specifically real-time monitoring of template usage, task docking scheduling and the operation effect of the detection model. The relevant data and information collected based on the monitoring results are fed back to the staff for timely adjustment and optimization.

[0106] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0107] While the embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that many changes, modifications, substitutions and variations can be made to the embodiments without departing from the principles and spirit of the invention.

[0108] The present invention and its embodiments are described above, and such description is not restrictive. The drawings show only one embodiment of the present invention, and the actual structure is not limited thereto. In short, if ordinary technicians in the field are inspired by it, without departing from the purpose of the invention, they can design a structure and embodiment similar to the technical solution without creativity, which should belong to the protection scope of the present invention.

Claims

1. A low-code data docking detection method based on artificial intelligence, Features: The method comprises the following steps: Step S1: component arrangement; Step S2: template test, specifically obtaining simulation data for template test, and indicating the error level through background color; Step S3: Establish a data docking detection model, specifically define sample neighborhoods and neighborhood granularity, measure feature similarity, use local gram matrix modeling to obtain the relationship between data features, and finally introduce an attention mechanism layer to transfer feature similarity and relationship, mark isomorphic and heterogeneous features, and realize the establishment of a data docking detection model; Step S4: Task docking scheduling, specifically to clarify the task scheduling goal is to find a task allocation plan that minimizes the overall delay, calculate the delay when the accelerator allocates tasks, allocate tasks in sequence, divide the CNN task into subtasks, plan the execution position of each layer, and then configure the task scheduling rules and record logs; Step S5: Update feedback.

2. According to the artificial intelligence-based low-code data docking detection method of claim 1, Features: In step S4, the task docking scheduling includes the following steps: Step S41: clarify the task scheduling goal and find a task allocation solution that minimizes the overall delay. The formula used is as follows: ； In the formula, U represents the overall goal of minimization, that is, minimizing the overall delay, k represents different computing nodes, represents the set of tasks assigned to the kth computing node, Indicates a specific task. Indicates different tasks, Indicates execution of tasks Assign to task Delay time on Step S42: Calculate the delay when the accelerator assigns tasks, sort the CNN tasks from large to small, traverse the sorted task list, and for each task, calculate the delay when assigning the task to each accelerator in turn, while considering the existing task queues on each accelerator, select the accelerator with the smallest delay and the smallest impact on the overall performance for assignment, and repeat the steps until all tasks are assigned; Step S43: Divide the CNN task into subtasks by layer, determine the characteristics and requirements of each subtask, evaluate the performance of each accelerator and the task type suitable for processing, and select the most suitable accelerator for allocation for each subtask; Step S44: Plan the execution location of each layer, and only assign the convolutional layer to different accelerators. When a specific convolutional layer is assigned to an accelerator, the subsequent pooling layers will also be executed on the same accelerator, and the fully connected layer will be arranged to be executed on the same accelerator as the last convolutional layer; Step S45: Configure the task scheduling rules. The user specifies that the task is executed at fixed minute intervals or hour intervals, and accurately specifies the specific time point of each day to start the task, so that the user can make reasonable arrangements according to actual needs and business scenarios. The scheduler will deliver tasks to the queue according to the task scheduling rules, and the consumer thread will wake up and execute tasks in the order in which the tasks are delivered to the queue. It supports disabling tasks during the scheduling process. When the tasks are called up again, they will no longer be scheduled due to the disabled state, completing the scheduling unloading. The task queue information will also be provided to users in a visual display mode. Step S46: Record logs. Through task and node listeners, record task execution logs and node data, and store them persistently for use in subsequent troubleshooting.

3. According to the artificial intelligence-based low-code data docking detection method of claim 1, Features: In step S3, the establishment of the data docking detection model includes the following steps: Step S31: define the sample neighborhood, the neighborhood granularity of the sample on the feature set is defined, and the formula used is as follows: ； In the formula, Representation sample neighborhood, U1 is the set containing all features, x represents the sample point in the feature set, θ is the threshold of the neighborhood granularity, represents the distance metric, Represents the sample point x and the sample The distance measure between them; Step S32: measure feature similarity, model the relationship according to the neighborhood granularity of the sample, and measure feature similarity by calculating the absolute value of the elements in the local gram matrix. The formula used is as follows: ； In the formula, represents the element in the i-th row and j-th column of the local gram matrix, represents the i-th eigenvector in the sample neighborhood granularity, represents the transpose of the jth eigenvector in the sample neighborhood granularity; Step S33: Use the local gram matrix to model and obtain the relationship between data features. The formula used is as follows: ； ； In the formula, n represents the number of neighborhood granularities, g represents the index of the specific neighborhood granularity, is the jth row of the coefficient matrix, represents the i-th row of the coefficient matrix, N represents the number of rows of the coefficient matrix, that is, the number dimension of elements involved in the calculation, represents the Laplacian matrix of the neighborhood granularity g, It is a diagonal matrix with a neighborhood granularity of g, which is used to encode the similarity between pairs of features, and tr() represents the trace of the matrix; Step S34: Introduce the attention mechanism layer, pass the relationship between feature similarity and data features to the attention mechanism layer, assign different attention weights to each feature, mark isomorphic and heterogeneous features according to the similarity threshold, evaluate the importance and weight the heterogeneous features, and then realize the establishment of the data docking detection model.

4. According to the artificial intelligence-based low-code data docking detection method of claim 1, Features: In step S2, the template test includes the following steps: Step S21: Acquire simulation data, obtain partial slice data obtained by the node pair test preview, configure corresponding generation rules according to own needs, create partial simulation data, and allow users to customize data according to actual conditions and specific requirements; Step S22: Performing a test, including the following steps: Step S221: Create a test instance and select simulation data, and enable the template test instance; Step S222: According to the operation result, the visualization component on the interface will accurately indicate the step where the data flow error occurs through the background color, and the prompt information is divided into three levels: green normal, yellow warning, and red error; Step S23: Modify the template. Based on the modification suggestions provided by this test, the user can directly adjust the template in the test instance until the template logic is normal.

5. According to the artificial intelligence-based low-code data docking detection method of claim 1, Features: In step S1, the component arrangement includes the following steps: Step S11: Generate a docking template, drag the component into the workbench through visual interaction, and use it as the basic unit of scheduling in the docking template; Step S12: Complete node configuration, click on the component, and complete the node configuration according to the interface configuration items. The node includes a data reading node, a data cleaning node, and a data writing node; Step S13: Perform orchestration. The low-code platform sets the execution order of components through visual operations, and determines the flow direction of subsequent steps based on the execution results, thereby orchestrating the relationships and processes between components and building a business flow and logical architecture that meets various needs.

6. According to the artificial intelligence-based low-code data docking detection method of claim 1, Features: In step S5, the update feedback is specifically real-time monitoring of template usage, task docking scheduling and the running effect of the detection model, and the relevant data and information collected based on the monitoring results are fed back to the staff so as to make timely adjustments and optimizations.

Images