AI Orchestration Methods and Systems for Load Balancing Across All Business Scenarios

Abstract

This invention discloses an AI orchestration method and system for load balancing across all business scenarios, relating to the field of business data processing technology. The invention collects user demand parameters from business scenarios in real time and processes these parameters using data processing methods. After processing, features of the user demand parameters are extracted using feature extraction. A recognition template library is established, and recognition is performed using graph neural network subgraph matching based on the extracted user demand parameter features. When no suitable template is found in the recognition template library, a template is automatically adjusted and generated using parameterized constraint solving and intelligent combination of atomic template fragments. Finally, based on the adjusted and generated template, the user demand parameters collected in real time are translated into code, and the translated code is tested and verified, improving the rationality of business data processing.

Description

Technical Field

[0001] This invention relates to the field of business data processing technology, specifically to an AI orchestration method and system for load balancing across all business scenarios. Background Technology

[0002] With the development of mobile internet, businesses need to meet the deployment requirements of multiple brands, multiple data centers, multiple operators, multiple protocol stacks, and multiple architectures. This requirement is currently mainly achieved through load balancing equipment. For such a complex business deployment scenario, the configuration of load balancers has brought great challenges. Often, the orchestration of a business scenario can take more than 3 hours to configure, and there is still no guarantee that it will be correct.

[0003] Existing technology, such as the invention patent application with publication number CN116719622A, discloses a business process orchestration method and system. The method includes: obtaining a template corresponding to the business process; parsing the template to generate a runtime task stack containing multiple tasks; for each task in the runtime task stack, a thread pool submits the task to a task queue for execution; calling the API of the resources required by the task to execute the API's interface logic; if the execution rule is an asynchronous MQ message rule, then according to the configured MQ queue information, listening for MQ asynchronous callback logic and responding to the detected asynchronous message, scheduling and refreshing whether there are any unexecuted tasks in the runtime task stack; if there are unexecuted tasks, returning to the step of submitting the task to the task queue by the thread pool for execution; if there are no unexecuted tasks, obtaining the business process orchestration result. Based on the above method, the business process orchestration work is scheduled according to the template, reducing the workload of staff.

[0004] As can be seen from the above solutions, most current service orchestration methods focus on orchestrating existing business processes and services within the operator, neglecting the analysis and orchestration of user demand data, thus having certain limitations. Summary of the Invention

[0005] The purpose of this invention is to provide an AI orchestration method and system for load balancing across all business scenarios, which solves the problems existing in the background technology.

[0006] To solve the above technical problems, the present invention adopts the following technical solution: The present invention provides an AI orchestration method for load balancing across all business scenarios, specifically including the following steps:

[0007] S1. Collect user demand parameters in business scenarios in real time, and process the collected user demand parameters in real time through data processing methods to obtain processed user demand parameters.

[0008] S2. Based on the processed user demand parameters, extract the user demand parameter features using feature extraction methods.

[0009] S3. Set up a recognition template library, and based on the extracted user demand parameter features, determine whether there is a suitable template in the recognition template library through graph neural network subgraph matching recognition method;

[0010] S4. When no suitable template is found in the template library, the template is automatically adjusted and generated by using parametric constraint solving and intelligent combination of atomic template fragments.

[0011] S5. Based on the adjusted and generated template, translate and write user requirement parameters in real-time collected business scenarios, convert user requirement parameters into code, and test and verify the converted code.

[0012] Preferably, the real-time collection of user demand parameters in the business scenario, and the processing of the real-time collected user demand parameters to obtain the processed user demand parameters, includes the following steps:

[0013] S11. Construct user requirement parameter standards, and filter the collected user requirement parameters based on the constructed user requirement parameter standards to obtain filtered user requirement parameters.

[0014] Set a set of standard user requirement parameters, including: user requirement generation time, user address, user requirement content, and business deadline.

[0015] Iterate through the collected user requirement parameters, and when missing data is detected in the collected user requirement parameters, delete the corresponding user requirement parameters;

[0016] Summarize the deleted user requirement parameters to obtain the filtered user requirement parameters;

[0017] S12. The filtered user demand parameters are filtered a second time based on the Bloom filter algorithm to obtain the processed user demand parameters.

[0018] Create an array of length m, and select... Each hash function iterates through each set of data in the filtered user requirement parameters and stores the results in an array.

[0019] During the traversal using the hash function, if two sets of data have the same traversal result, each bit in these two sets of data is compared.

[0020] When the comparison results are consistent, the two sets of data are set as the same data, and the filtered user requirement parameters are generated after deletion based on the user requirement generation time in the user requirement parameters.

[0021] After the traversal is complete, the filtered user requirement parameters stored in the array during the traversal are summarized to obtain the processed user requirement parameters.

[0022] Preferably, the step of extracting user demand parameter features based on the obtained processed user demand parameters includes the following steps:

[0023] The processed user demand parameters are analyzed using feature analysis methods to determine their characteristics.

[0024] Based on the processed user demand parameters input, the features of the processed user demand parameters are extracted through the convolution operation of a convolutional neural network and saved.

[0025] The formula for calculating convolution is as follows:

[0026] ;

[0027] in, This represents the processed user requirement parameters input. This represents the weights of the corresponding convolution kernel. Indicates the bias value. This represents the extracted and processed user demand parameters.

[0028] The processed user demand parameters are classified using a similarity comparison method;

[0029] A similarity threshold is set. When the similarity value between the processed user demand parameter features is greater than the set similarity threshold, it indicates that the two sets of processed user demand parameters are of the same type of data. Otherwise, they are of different types of data.

[0030] Preferably, the step of setting up a recognition template library and determining whether there is a suitable template in the recognition template library based on the extracted user demand parameter features using a graph neural network subgraph matching recognition method includes the following steps:

[0031] S31. Based on the extracted user demand parameter features, user demand keywords are extracted in real time and a user undirected graph is constructed using the keyword weight calculation method.

[0032] S32. After the user's undirected graph is constructed, the graph neural network subgraph matching method is used to determine whether there is a suitable template in the template library.

[0033] Preferably, the step of extracting user demand keywords and constructing a user undirected graph in real time based on the extracted user demand parameter features and using a keyword weight calculation method includes the following steps:

[0034] Based on the extracted user demand parameter features, various user demand parameters are summarized, and the central data of various user demand parameters are calculated based on the mean calculation method.

[0035] By selecting keywords, all keywords from the core data of various user demand parameters are selected to construct a keyword set;

[0036] Keyword weight is determined by keyword position. The formula for calculating keyword weight is as follows:

[0037] ;

[0038] in, This represents the weight of the i-th keyword group. This represents the frequency of the i-th keyword group. This indicates the position where the i-th keyword appears. This indicates the proportion of the i-th group of keywords. These represent the weighting factors for frequency, position, and proportion, respectively.

[0039] Based on the calculated keyword weight values, the top 30 keywords from the calculated keyword weights are selected as the topic words;

[0040] Construct an undirected graph of users based on the frequency and location of keywords;

[0041] In the user's undirected graph, each node represents one and only one topic word, and each edge in the user's undirected graph represents that the current topic word and the connected topic word are in the same sentence.

[0042] Preferably, after the user undirected graph is constructed, determining whether there is a suitable template in the template library through graph neural network subgraph matching and recognition includes the following steps:

[0043] Each recognition template in the recognition template library is converted into a template undirected graph, and the converted template undirected graph is input into a graph neural network. The convolution operation in the graph neural network is used to recognize the converted template undirected graph and output the features of the corresponding template undirected graph.

[0044] The convolution operation is as follows:

[0045] ;

[0046] in, Represents the transformed template undirected graph , Represents a convolutional neural network. Indicates the filter parameters. Represents an undirected graph with respect to a template. Fourier transform, This represents the Fourier transform of the filter parameters. Indicates Fourier transform, This represents the convolution operation. Represents matrix operations. This represents the characteristics of the undirected graph corresponding to the template;

[0047] The constructed user undirected graph is input into a graph neural network to obtain the features of the user undirected graph. ;

[0048] The features of user undirected graphs and template undirected graphs are analyzed and matched using subgraph matching.

[0049] The subgraph matching analysis formula is:

[0050] ;

[0051] in, Features of the user's undirected graph This represents the analysis and matching results of user undirected graph features and template undirected graph features;

[0052] A threshold is set for the analysis and matching results. When the analysis and matching results of the user's undirected graph features and the template's undirected graph features are greater than or equal to the set threshold, it is determined that there is a suitable template in the current recognition template library; when the analysis and matching results of the user's undirected graph features and the template's undirected graph features are less than the threshold, it is determined that there is no suitable template in the recognition template library.

[0053] Preferably, when no suitable template is found in the template library, the method of automatically adjusting and generating templates through parametric constraint solving and intelligent combination of atomic template fragments includes the following steps:

[0054] S41. When no suitable template is found in the template library, chromosome encoding is performed on the constructed user undirected graph and template undirected graph to obtain the chromosome encoding sequence.

[0055] Select vertices from the user undirected graph and the template undirected graph, and set the number of the user undirected graph vertices and the template undirected graph vertices to 1. Based on the set number of the user undirected graph vertices and the template undirected graph vertices, use the breadth-first search algorithm to number the vertices of the user undirected graph and the template undirected graph.

[0056] The user-defined undirected graph and the template undirected graph are numbered using chromosome coding sequences;

[0057] S42. Intelligent combination of chromosome coding sequences corresponding to user undirected graphs is performed by parametric constraint solving and atomic template fragment method to obtain intelligently combined chromosome coding sequences;

[0058] Based on the vertices and chromosome coding sequences of the user's undirected graph, the chromosome coding sequences are intelligently combined through parameterized constraint solving and atomic template fragments.

[0059] Encoding constraints are set: there must be connecting edges between the generated chromosome coding sequences;

[0060] S43. Calculate the fitness of the chromosome coding sequence after intelligent combination using a mixed integer algorithm, and realize the automatic adjustment and generation of templates based on the calculated fitness.

[0061] Preferably, the step of calculating the fitness of the chromosome coding sequence after intelligent combination using a mixed integer algorithm, and automatically adjusting and generating the template based on the calculated fitness, includes the following steps:

[0062] The subgraph matching analysis formula is set as the fitness objective function;

[0063] Fitness function: ;

[0064] in, Represents the chromosome coding sequence after intelligent combination. And template undirected graph chromosome coding sequence Analysis and matching results between them;

[0065] Based on the calculated fitness values ​​of the chromosome coding sequences after each intelligent combination, the chromosome coding sequences after intelligent combination are selected.

[0066] The probability of the chromosome coding sequence after intelligent combination being selected is set as follows: ;

[0067] ;

[0068] in, This indicates the number of chromosome coding sequences after intelligent combination. This represents the number of chromosome coding sequences in the template undirected graph. Represents the chromosome coding sequence after intelligent combination. And template undirected graph chromosome coding sequence Analysis and matching results between them This represents the probability that the chromosome coding sequence after the nth intelligent combination is selected;

[0069] Set the maximum number of iterations Y and the iteration threshold H. Set the algorithm to terminate the mixed integer algorithm when the output fitness value converges after B iterations or when the number of iterations reaches the set maximum number of iterations. Use the user requirement parameters corresponding to the chromosome coding sequence of the last output as the template for automatic adjustment and generation.

[0070] Preferably, the process of translating and writing user requirement parameters from real-time collected business scenarios based on the adjusted and generated template, converting the user requirement parameters into code, and testing and verifying the converted code includes the following steps:

[0071] Based on the adjusted and generated template, the user requirement parameters in the real-time collected business scenarios are translated and written to generate compilable code, and the compilable code is compiled.

[0072] During the compilation process, the administrator checks the compilable code and uses data testing to determine whether the generated compilable code meets the user's requirements.

[0073] The present invention also provides an AI orchestration system for load balancing across all business scenarios, implementing an AI orchestration method for load balancing across all business scenarios. The system includes: a data acquisition module, a data processing module, a feature analysis module, a template matching module, a template adjustment module, and a requirement verification module.

[0074] The data acquisition module is used to collect user demand parameters in business scenarios in real time.

[0075] The data processing module is used to process user demand parameters collected in real time business scenarios to obtain processed user demand parameters.

[0076] The feature analysis module is used to extract features from the processed user demand parameters;

[0077] The template matching module is used to determine whether there are suitable templates in the template library based on the extracted user demand parameter features and through graph neural network subgraph matching recognition method.

[0078] The template adjustment module is used to achieve automatic template adjustment and generation through a method of parametric constraint solving and intelligent combination of atomic template fragments;

[0079] The requirement verification module is used to translate and write user requirement parameters collected in real-time business scenarios, convert user requirement parameters into code, and test and verify the converted code.

[0080] The beneficial effects of this invention are as follows:

[0081] (1) This invention collects user demand parameters in business scenarios in real time and processes the collected user demand parameters through data processing. After processing, based on the processed user demand parameters, the user demand parameter features are extracted through feature extraction. At the same time, a recognition template library is set, and based on the extracted user demand parameter features, recognition is performed through graph neural network subgraph matching. When there is no suitable template in the recognition template library, the template is automatically adjusted and generated through parameterized constraint solving and intelligent combination of atomic template fragments. Finally, based on the adjusted and generated template, the user demand parameters in the real-time collected business scenarios are translated into code, and the converted code is tested and verified, which improves the rationality of business data processing.

[0082] (2) This invention constructs a standard for user demand parameters and filters the collected user demand parameters twice based on the constructed standard, thereby ensuring the accuracy of user demand parameter collection and processing.

[0083] (3) The present invention extracts user demand keywords in real time through keyword weight calculation method. At the same time, based on the real-time extraction of user demand keywords, it constructs user undirected graph. After the construction is completed, it judges whether there is a suitable template in the template library through graph neural network subgraph matching and recognition method. When there is no suitable template, it automatically adjusts and generates a new template, thereby improving the adaptability of business data matching. Attached Figure Description

[0084] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0085] Figure 1 This is a schematic diagram of the AI ​​orchestration method for load balancing across all business scenarios according to the present invention. Detailed Implementation

[0086] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0087] In a specific embodiment of the present invention,

[0088] Reference Figure 1As shown, this invention provides an AI orchestration method for load balancing across all business scenarios, comprising the following steps:

[0089] S1. Collect user demand parameters in business scenarios in real time, and process the collected user demand parameters in real time through data processing methods to obtain processed user demand parameters.

[0090] S2. Based on the processed user demand parameters, extract the user demand parameter features using feature extraction methods.

[0091] S3. Set up a recognition template library, and based on the extracted user demand parameter features, determine whether there is a suitable template in the recognition template library through graph neural network subgraph matching recognition method;

[0092] S4. When no suitable template is found in the template library, the template is automatically adjusted and generated by using parametric constraint solving and intelligent combination of atomic template fragments.

[0093] S5. Based on the adjusted and generated template, translate and write user requirement parameters in real-time collected business scenarios, convert user requirement parameters into code, and test and verify the converted code.

[0094] Furthermore, referring to Figure 1 As shown, user demand parameters in the business scenario are collected in real time, and these parameters are processed using data processing methods to obtain the processed user demand parameters. The process includes the following steps:

[0095] S11. Construct user requirement parameter standards, and filter the collected user requirement parameters based on the constructed user requirement parameter standards to obtain filtered user requirement parameters.

[0096] Set a set of standard user requirement parameters, including: user requirement generation time, user address, user requirement content, and business deadline.

[0097] Iterate through the collected user requirement parameters, and when missing data is detected in the collected user requirement parameters, delete the corresponding user requirement parameters;

[0098] Summarize the deleted user requirement parameters to obtain the filtered user requirement parameters;

[0099] S12. The filtered user demand parameters are filtered a second time based on the Bloom filter algorithm to obtain the processed user demand parameters.

[0100] Create an array of length m, and select... Each hash function iterates through each set of data in the filtered user requirement parameters and stores the results in an array.

[0101] During the traversal using the hash function, if two sets of data have the same traversal result, each bit in these two sets of data is compared.

[0102] When the comparison results are consistent, the two sets of data are set as the same data, and the filtered user requirement parameters are generated after deletion based on the user requirement generation time in the user requirement parameters.

[0103] After the traversal is completed, the filtered user requirement parameters stored in the array during the traversal are summarized to obtain the processed user requirement parameters.

[0104] Furthermore, referring to Figure 1 As shown, based on the processed user demand parameters, the extraction of user demand parameter features through feature extraction includes the following steps:

[0105] The processed user demand parameters are analyzed using feature analysis methods to determine their characteristics.

[0106] Furthermore, based on the processed user demand parameters input, the features of the processed user demand parameters are extracted through the convolution operation of a convolutional neural network and saved;

[0107] The formula for calculating convolution is as follows:

[0108] ;

[0109] in, This represents the processed user requirement parameters input. This represents the weights of the corresponding convolution kernel. Indicates the bias value. This represents the extracted and processed user demand parameters.

[0110] Furthermore, the processed user demand parameters are classified using a similarity comparison method;

[0111] The similarity comparison formula is as follows:

[0112] ;

[0113] in, Indicates the characteristics of processed user demand parameters and processed user requirement parameters characteristics Covariance between This represents the standard deviation of the processed user demand parameter characteristics. This represents the standard deviation of the processed user demand parameter characteristics. Indicates the characteristics of processed user demand parameters and processed user requirement parameters characteristics The similarity between them;

[0114] Furthermore, a similarity threshold is set when the processed user demand parameters feature and processed user requirement parameters characteristics When the similarity value between the two sets of user demand parameters is greater than the set similarity threshold, it means that the two sets of user demand parameters are of the same type; otherwise, they are of different types.

[0115] Furthermore, referring to Figure 1 As shown, the process of setting up a recognition template library and determining whether there are suitable templates in the library based on the extracted user demand parameter features using a graph neural network subgraph matching recognition method includes the following steps:

[0116] S31. Based on the extracted user demand parameter features, user demand keywords are extracted in real time and a user undirected graph is constructed using the keyword weight calculation method.

[0117] Based on the extracted user demand parameter features, various user demand parameters are summarized, and the central data of various user demand parameters are calculated based on the mean calculation method.

[0118] Furthermore, by selecting keywords, all keywords from the core data of various user demand parameters are selected to construct a keyword set;

[0119] Furthermore, keyword weight is determined by keyword position. The keyword weight calculation formula is as follows:

[0120] ;

[0121] in, This represents the weight of the i-th keyword group. This represents the frequency of the i-th keyword group. This indicates the position where the i-th keyword appears. This indicates the proportion of the i-th group of keywords. These represent the weighting factors for frequency, position, and proportion, respectively.

[0122] Based on the calculated keyword weight values, the top 30 keywords from the calculated keyword weights are selected as the topic words;

[0123] Furthermore, an undirected graph of users is constructed based on the frequency and location of the keywords.

[0124] In the user undirected graph, each node represents one and only one topic word, and each edge in the user undirected graph represents that the current topic word and the connected topic word are in the same sentence.

[0125] S32. After the user's undirected graph is constructed, the graph neural network subgraph matching and recognition method is used to determine whether there is a suitable template in the template library.

[0126] Each recognition template in the recognition template library is converted into a template undirected graph, and the converted template undirected graph is input into a graph neural network. The convolution operation in the graph neural network is used to recognize the converted template undirected graph and output the features of the corresponding template undirected graph.

[0127] The convolution operation is as follows:

[0128] ;

[0129] in, Represents the transformed template undirected graph , Represents a convolutional neural network. Indicates the filter parameters. Represents an undirected graph with respect to a template. Fourier transform, This represents the Fourier transform of the filter parameters. Indicates Fourier transform, This represents the convolution operation. Represents matrix operations. This represents the characteristics of the undirected graph corresponding to the template;

[0130] Furthermore, the constructed user undirected graph is input into a graph neural network to obtain the features of the user undirected graph. ;

[0131] The features of user undirected graphs and template undirected graphs are analyzed and matched using subgraph matching.

[0132] The subgraph matching analysis formula is:

[0133] ;

[0134] in, Features of the user's undirected graph This represents the analysis and matching results of user undirected graph features and template undirected graph features;

[0135] Set a threshold for the analysis and matching results. When the analysis and matching results of the user's undirected graph features and the template's undirected graph features are greater than or equal to the set threshold, it is determined that there is a suitable template in the current recognition template library; when the analysis and matching results of the user's undirected graph features and the template's undirected graph features are less than the threshold, it is determined that there is no suitable template in the recognition template library.

[0136] Furthermore, referring to Figure 1As shown, when no suitable template is found in the template library, the automatic adjustment and generation of templates is achieved through a method of parametric constraint solving and intelligent combination of atomic template fragments, including the following steps:

[0137] S41. When no suitable template is found in the template library, chromosome encoding is performed on the constructed user undirected graph and template undirected graph to obtain the chromosome encoding sequence.

[0138] Select vertices from the user undirected graph and the template undirected graph, and set the number of the user undirected graph vertices and the template undirected graph vertices to 1. Based on the set number of the user undirected graph vertices and the template undirected graph vertices, use the breadth-first search algorithm to number the vertices of the user undirected graph and the template undirected graph.

[0139] The user-defined undirected graph and the template undirected graph are numbered using chromosome coding sequences;

[0140] S42. Intelligent combination of chromosome coding sequences corresponding to user undirected graphs is performed by parametric constraint solving and atomic template fragment method to obtain intelligently combined chromosome coding sequences;

[0141] Based on the vertices and chromosome coding sequences of the user's undirected graph, the chromosome coding sequences are intelligently combined through parameterized constraint solving and atomic template fragments.

[0142] Encoding constraints are set: there must be connecting edges between the generated chromosome coding sequences;

[0143] S43. Calculate the fitness of the chromosome coding sequence after intelligent combination using a mixed integer algorithm, and realize the automatic adjustment and generation of templates based on the calculated fitness.

[0144] The subgraph matching analysis formula is set as the fitness objective function;

[0145] Fitness function: ;

[0146] in, Represents the chromosome coding sequence after intelligent combination. And template undirected graph chromosome coding sequence Analysis and matching results between them;

[0147] Furthermore, based on the calculated fitness values ​​of the chromosome coding sequences after each intelligent combination, the chromosome coding sequences after intelligent combination are selected.

[0148] The probability of the chromosome coding sequence after intelligent combination being selected is set as follows: ;

[0149] ;

[0150] in, This indicates the number of chromosome coding sequences after intelligent combination. This represents the number of chromosome coding sequences in the template undirected graph. Represents the chromosome coding sequence after intelligent combination. And template undirected graph chromosome coding sequence Analysis and matching results between them This represents the probability that the chromosome coding sequence after the nth intelligent combination is selected;

[0151] Furthermore, the maximum number of iterations Y and the iteration threshold H are set. The mixed integer algorithm is terminated when the output fitness value converges after B iterations or when the number of iterations reaches the set maximum number of iterations. The user requirement parameters corresponding to the chromosome coding sequence of the last output are used as the template for automatic adjustment and generation.

[0152] Furthermore, referring to Figure 1 As shown, based on the adjusted and generated template, the user requirement parameters in the real-time collected business scenario are translated and written, the user requirement parameters are converted into code, and the converted code is tested and verified, including the following steps:

[0153] Based on the adjusted and generated template, the user requirement parameters in the real-time collected business scenarios are translated and written to generate compilable code, and the compilable code is compiled.

[0154] Furthermore, during the compilation process, the administrator checks the compilable code and uses data testing to determine whether the generated compilable code meets the user's requirements.

[0155] In one specific embodiment, the AI ​​orchestration system for load balancing across all business scenarios implements an AI orchestration method for load balancing across all business scenarios. The system includes: a data acquisition module, a data processing module, a feature analysis module, a template matching module, a template adjustment module, and a requirement verification module.

[0156] The data acquisition module is used to collect user demand parameters in business scenarios in real time.

[0157] The data processing module is used to process user demand parameters collected in real time business scenarios to obtain processed user demand parameters.

[0158] The feature analysis module is used to extract features from the processed user demand parameters;

[0159] The template matching module is used to determine whether there are suitable templates in the template library based on the extracted user demand parameter features and through graph neural network subgraph matching recognition method.

[0160] The template adjustment module is used to achieve automatic template adjustment and generation through a method of parametric constraint solving and intelligent combination of atomic template fragments;

[0161] The requirement verification module is used to translate and write user requirement parameters collected in real-time business scenarios, convert user requirement parameters into code, and test and verify the converted code.

[0162] It should be noted that,

[0163] The above content is merely an example and illustration of the concept of the present invention. Those skilled in the art can make various modifications or additions to the specific embodiments described, or use similar methods to replace them, as long as they do not deviate from the concept of the invention or exceed the scope defined by the present invention, and all such modifications and additions should fall within the protection scope of the present invention.

Claims

1. An AI orchestration method for load balancing across all business scenarios, characterized in that, Includes the following steps: S1. Collect user demand parameters in business scenarios in real time, and process the collected user demand parameters in real time through data processing methods to obtain processed user demand parameters. S2. Based on the processed user demand parameters, extract the user demand parameter features using feature extraction methods. S3. Set up a recognition template library, and based on the extracted user demand parameter features, determine whether there is a suitable template in the recognition template library through graph neural network subgraph matching recognition method; S4. When no suitable template is found in the template library, the template is automatically adjusted and generated by using parametric constraint solving and intelligent combination of atomic template fragments. S5. Based on the adjusted and generated template, translate and write user requirement parameters in real-time collected business scenarios, convert user requirement parameters into code, and test and verify the converted code.

2. The AI ​​orchestration method for load balancing across all business scenarios according to claim 1, characterized in that, The process of collecting user demand parameters in real-time from business scenarios and processing these parameters to obtain the processed user demand parameters includes the following steps: S11. Construct user requirement parameter standards, and filter the collected user requirement parameters based on the constructed user requirement parameter standards to obtain filtered user requirement parameters. Set a set of standard user requirement parameters, including: user requirement generation time, user address, user requirement content, and business deadline. Iterate through the collected user requirement parameters, and when missing data is detected in the collected user requirement parameters, delete the corresponding user requirement parameters; Summarize the deleted user requirement parameters to obtain the filtered user requirement parameters; S12. The filtered user demand parameters are filtered a second time based on the Bloom filter algorithm to obtain the processed user demand parameters. Create an array of length m, and select... Each hash function iterates through each set of data in the filtered user requirement parameters and stores the results in an array. During the traversal using the hash function, if two sets of data have the same traversal result, each bit in these two sets of data is compared. When the comparison results are consistent, the two sets of data are set as the same data, and the filtered user requirement parameters are generated after deletion based on the user requirement generation time in the user requirement parameters. After the traversal is complete, the filtered user requirement parameters stored in the array during the traversal are summarized to obtain the processed user requirement parameters.

3. The AI ​​orchestration method for load balancing across all business scenarios according to claim 1, characterized in that, The process of extracting user demand parameter features based on the obtained processed user demand parameters includes the following steps: The processed user demand parameters are analyzed using feature analysis methods to determine their characteristics. Based on the processed user demand parameters input, the features of the processed user demand parameters are extracted through the convolution operation of a convolutional neural network and saved. The formula for calculating convolution is as follows: ； in, This represents the processed user requirement parameters input. This represents the weights of the corresponding convolution kernel. Indicates the bias value. This represents the extracted and processed user demand parameters. The processed user demand parameters are classified using a similarity comparison method; A similarity threshold is set. When the similarity value between the processed user demand parameter features is greater than the set similarity threshold, it indicates that the two sets of processed user demand parameters are of the same type of data. Otherwise, they are of different types of data.

4. The AI ​​orchestration method for load balancing across all business scenarios according to claim 1, characterized in that, The process of setting up a recognition template library and determining whether there are suitable templates in the library based on extracted user demand parameter features using a graph neural network subgraph matching recognition method includes the following steps: S31. Based on the extracted user demand parameter features, user demand keywords are extracted in real time and a user undirected graph is constructed using the keyword weight calculation method. S32. After the user's undirected graph is constructed, the graph neural network subgraph matching method is used to determine whether there is a suitable template in the template library.

5. The AI ​​orchestration method for load balancing across all business scenarios according to claim 4, characterized in that, The process of extracting user demand parameters based on extracted user demand features, and then using a keyword weight calculation method to extract user demand keywords in real time and construct an undirected user graph includes the following steps: Based on the extracted user demand parameter features, various user demand parameters are summarized, and the central data of various user demand parameters are calculated based on the mean calculation method. By selecting keywords, all keywords from the core data of various user demand parameters are selected to construct a keyword set; Keyword weight is determined by keyword position. The formula for calculating keyword weight is as follows: ； in, This represents the weight of the i-th keyword group. This represents the frequency of the i-th keyword group. This indicates the position where the i-th keyword appears. This indicates the proportion of the i-th group of keywords. These represent the weighting factors for frequency, position, and proportion, respectively. Based on the calculated keyword weight values, the top 30 keywords from the calculated keyword weights are selected as the topic words; Construct an undirected graph of users based on the frequency and location of keywords; In the user's undirected graph, each node represents one and only one topic word, and each edge in the user's undirected graph represents that the current topic word and the connected topic word are in the same sentence.

6. The AI ​​orchestration method for load balancing across all business scenarios according to claim 4, characterized in that, After the user undirected graph is constructed, the method of graph neural network subgraph matching and recognition is used to determine whether there is a suitable template in the template library, which includes the following steps: Each recognition template in the recognition template library is converted into a template undirected graph, and the converted template undirected graph is input into a graph neural network. The convolution operation in the graph neural network is used to recognize the converted template undirected graph and output the features of the corresponding template undirected graph. The convolution operation is as follows: ； in, Represents the transformed template undirected graph , Represents a convolutional neural network. Indicates the filter parameters. Represents an undirected graph with respect to a template. Fourier transform, This represents the Fourier transform of the filter parameters. Indicates Fourier transform, This represents the convolution operation. Represents matrix operations. This represents the characteristics of the undirected graph corresponding to the template; The constructed user undirected graph is input into a graph neural network to obtain the features of the user undirected graph. ; The features of user undirected graphs and template undirected graphs are analyzed and matched using subgraph matching. The subgraph matching analysis formula is: ； in, Features of the user's undirected graph This represents the analysis and matching results of user undirected graph features and template undirected graph features; A threshold is set for the analysis and matching results. When the analysis and matching results of the user's undirected graph features and the template's undirected graph features are greater than or equal to the set threshold, it is determined that there is a suitable template in the current recognition template library; when the analysis and matching results of the user's undirected graph features and the template's undirected graph features are less than the threshold, it is determined that there is no suitable template in the recognition template library.

7. The AI ​​orchestration method for load balancing across all business scenarios according to claim 1, characterized in that, When no suitable template is found in the template library, the method of automatically adjusting and generating templates through parametric constraint solving and intelligent combination of atomic template fragments includes the following steps: S41. When no suitable template is found in the template library, chromosome encoding is performed on the constructed user undirected graph and template undirected graph to obtain the chromosome encoding sequence. Select vertices from the user undirected graph and the template undirected graph, and set the number of the user undirected graph vertices and the template undirected graph vertices to 1. Based on the set number of the user undirected graph vertices and the template undirected graph vertices, use the breadth-first search algorithm to number the vertices of the user undirected graph and the template undirected graph. The user-defined undirected graph and the template undirected graph are numbered using chromosome coding sequences; S42. Intelligent combination of chromosome coding sequences corresponding to user undirected graphs is performed by parametric constraint solving and atomic template fragment method to obtain intelligently combined chromosome coding sequences; Based on the vertices and chromosome coding sequences of the user's undirected graph, the chromosome coding sequences are intelligently combined through parameterized constraint solving and atomic template fragments. Encoding constraints are set: there must be connecting edges between the generated chromosome coding sequences; S43. Calculate the fitness of the chromosome coding sequence after intelligent combination using a mixed integer algorithm, and realize the automatic adjustment and generation of templates based on the calculated fitness.

8. The AI ​​orchestration method for load balancing across all business scenarios according to claim 7, characterized in that, The process of calculating the fitness of the chromosome coding sequence after intelligent combination using a mixed integer algorithm, and automatically adjusting and generating templates based on the calculated fitness, includes the following steps: The subgraph matching analysis formula is set as the fitness objective function; Fitness function: ; in, Represents the chromosome coding sequence after intelligent combination. And template undirected graph chromosome coding sequence Analysis and matching results between them; Based on the calculated fitness values ​​of the chromosome coding sequences after each intelligent combination, the chromosome coding sequences after intelligent combination are selected. The probability of the chromosome coding sequence after intelligent combination being selected is set as follows: ; ； in, This indicates the number of chromosome coding sequences after intelligent combination. This represents the number of chromosome coding sequences in the template undirected graph. Represents the chromosome coding sequence after intelligent combination. And template undirected graph chromosome coding sequence Analysis and matching results between them This represents the probability that the chromosome coding sequence after the nth intelligent combination is selected; Set the maximum number of iterations Y and the iteration threshold H. Set the algorithm to terminate the mixed integer algorithm when the output fitness value converges after B iterations or when the number of iterations reaches the set maximum number of iterations. Use the user requirement parameters corresponding to the chromosome coding sequence of the last output as the template for automatic adjustment and generation.

9. The AI ​​orchestration method for load balancing across all business scenarios according to claim 1, characterized in that, The process of translating and writing user requirement parameters from real-time collected business scenarios based on the adjusted and generated template, converting user requirement parameters into code, and testing and verifying the converted code includes the following steps: Based on the adjusted and generated template, the user requirement parameters in the real-time collected business scenarios are translated and written to generate compilable code, and the compilable code is compiled. During the compilation process, the administrator checks the compilable code and uses data testing to determine whether the generated compilable code meets the user's requirements.

10. A system for implementing the AI ​​orchestration method for load balancing across all business scenarios as described in claim 1, characterized in that, include: The module includes a data acquisition module, a data processing module, a feature analysis module, a template matching module, a template adjustment module, and a requirements verification module. The data acquisition module is used to collect user demand parameters in business scenarios in real time. The data processing module is used to process user demand parameters collected in real time business scenarios to obtain processed user demand parameters. The feature analysis module is used to extract features from the processed user demand parameters; The template matching module is used to determine whether there are suitable templates in the template library based on the extracted user demand parameter features and through graph neural network subgraph matching recognition method. The template adjustment module is used to achieve automatic template adjustment and generation through a method of parametric constraint solving and intelligent combination of atomic template fragments; The requirement verification module is used to translate and write user requirement parameters collected in real-time business scenarios, convert user requirement parameters into code, and test and verify the converted code.

Images