Agent-based service policy determination method and apparatus, and agent

Abstract

The application provides an agent-based business policy determination method and device and an agent, relates to the technical field of artificial intelligence, and in particular to the technical fields of large models, agents, smart logistics and the like. The agent-based business policy determination method comprises: using a plurality of first agents to execute a plurality of tasks with a dependency relationship according to demand information about business demand, to determine a business policy script and a basic example; performing perturbation on example data related to model items of an algorithm model in the basic example to obtain a perturbed example; using a second agent to perform defect evaluation according to a perturbed example result to determine defect information related to a target task in the plurality of tasks, the perturbed example result being determined according to execution of the perturbed example by the business policy script; using a target agent corresponding to the defect information in the plurality of first agents to execute the target task according to the defect information to obtain a target business policy script; and determining a target business policy according to the target business policy script.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, particularly to the fields of large models, intelligent agents, and smart logistics, specifically to a method, apparatus, and intelligent agent for determining business strategies based on intelligent agents. Background Technology

[0002] An intelligent agent can be a software entity that can perceive its environment and make autonomous decisions to achieve specific goals. For example, an intelligent agent can perform specific tasks by calling various types of tools such as search tools, and can process information of any modality, such as text and images, by calling large models to output responses that meet user needs. Summary of the Invention

[0003] This application provides a method, apparatus, and intelligent agent for determining business strategies based on intelligent agents.

[0004] According to one aspect of this application, a business strategy determination method based on intelligent agents is provided, comprising: utilizing multiple first intelligent agents to execute multiple tasks with dependencies based on demand information regarding business requirements, determining a business strategy script and a basic computational instance, wherein the business strategy script represents an algorithm model matching the business requirements; perturbing the computational instance data related to model items of the algorithm model in the basic computational instance to obtain perturbed computational instances; utilizing a second intelligent agent to perform defect assessment based on the results of the perturbed computational instances, determining defect information related to a target task among the multiple tasks, wherein the results of the perturbed computational instances are determined by executing the perturbed computational instances according to the business strategy script; utilizing the target intelligent agent among the multiple first intelligent agents corresponding to the defect information to execute the target task based on the defect information, obtaining a target business strategy script; and determining a target business strategy based on the target business strategy script.

[0005] According to another aspect of this application, a business strategy determination device based on intelligent agents is provided, comprising: a first determination module, configured to utilize multiple first intelligent agents to execute multiple tasks with dependencies based on demand information regarding business requirements, and determine a business strategy script and a basic calculation example, wherein the business strategy script represents an algorithm model matching the business requirements; a perturbation calculation example acquisition module, configured to perturb the calculation example data related to model items of the algorithm model in the basic calculation example to obtain a perturbation calculation example; a second determination module, configured to utilize a second intelligent agent to perform defect assessment based on the results of the perturbation calculation example, and determine defect information related to a target task among the multiple tasks, wherein the results of the perturbation calculation example are determined according to the perturbation calculation example execution based on the business strategy script; a target business strategy script acquisition module, configured to utilize a target intelligent agent corresponding to the defect information among the multiple first intelligent agents to execute a target task based on the defect information, and obtain a target business strategy script; and a third determination module, configured to determine a target business strategy based on the target business strategy script.

[0006] According to another aspect of this application, an artificial intelligence agent is provided, comprising: an input module for receiving input information; a processing module for determining a target task based on the input information received by the input module, determining a large model based on the target task, and obtaining output information by calling the large model to execute the method provided in the embodiments of this application; and an output module for outputting the output information obtained by the processing module.

[0007] According to another aspect of this application, an electronic device is provided, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform a method provided according to an embodiment of this application.

[0008] According to another aspect of this application, a non-transitory computer-readable storage medium is provided storing computer instructions, wherein the computer instructions are used to cause the computer to perform a method provided according to an embodiment of this application.

[0009] According to another aspect of this application, a computer program product is provided, including a computer program that, when executed by a processor, implements the method provided according to embodiments of this application.

[0010] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of this application, nor is it intended to limit the scope of this application. Other features of this application will become readily apparent from the following description. Attached Figure Description

[0011] The accompanying drawings are provided for a better understanding of this solution and do not constitute a limitation of this application. Wherein:

[0012] Figure 1 The illustration shows an exemplary system architecture for applying an agent-based business strategy determination method and apparatus according to embodiments of this application.

[0013] Figure 2 A flowchart illustrating an agent-based business policy determination method according to an embodiment of this application is shown.

[0014] Figure 3 A flowchart illustrating an agent-based business strategy determination method according to another embodiment of this application is shown.

[0015] Figure 4 A schematic diagram of the structure of a service strategy orchestration system according to an embodiment of this application is shown.

[0016] Figure 5A block diagram of an agent-based business policy determination apparatus according to an embodiment of this application is illustrated schematically.

[0017] Figure 6 A schematic diagram illustrating the structure of an intelligent agent of artificial intelligence according to an embodiment of this application is shown.

[0018] Figure 7 A schematic block diagram of an example electronic device is shown that can be used to implement the agent-based business policy determination method of the present application. Detailed Implementation

[0019] The following description, in conjunction with the accompanying drawings, illustrates exemplary embodiments of this application, including various details to aid understanding. These should be considered merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of this application. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description.

[0020] In the technical solution of this application, the acquisition, storage and application of any type of information, such as user personal information, comply with the provisions of relevant laws and regulations, necessary confidentiality measures have been taken, and it does not violate public order and good morals.

[0021] In scenarios involving business strategy planning and execution, such as logistics operation scheduling, power dispatching, and industrial production planning, intelligent agents can be used to plan and adjust business strategy solutions based on business needs and knowledge, enabling the business strategies output by the agents to improve business efficiency more effectively. However, the business strategies output by intelligent agents are prone to defects such as business execution errors and significant discrepancies between strategy execution results and business objectives. Furthermore, it is difficult to conveniently prompt the agent to identify the causes of these defects, resulting in low efficiency in correcting business strategies and failing to meet actual business needs.

[0022] This application provides an agent-based business strategy determination method, apparatus, and agent. The agent-based business strategy determination method includes: utilizing multiple first agents to execute multiple tasks with dependencies based on demand information regarding business requirements, determining a business strategy script and basic computational examples, where the business strategy script represents an algorithm model matching the business requirements; perturbing computational example data related to model items in the basic computational examples to obtain perturbed computational examples; utilizing second agents to perform defect assessment based on the results of the perturbed computational examples, determining defect information related to a target task among the multiple tasks, the perturbed computational example results being determined based on the business strategy script; utilizing a target agent among the multiple first agents corresponding to the defect information to execute a target task based on the defect information, obtaining a target business strategy script; and determining a target business strategy based on the target business strategy script.

[0023] According to embodiments of this application, by utilizing multiple first agents to execute multiple tasks with dependencies based on demand information to determine the business strategy script and basic computational instance, multi-agent collaboration can be achieved to complete business strategy planning based on demand information, thereby improving the efficiency of business strategy planning. By perturbing the computing power data related to the model items of the algorithm model in the basic computational instance, and using a second whole to evaluate the defects in the perturbed computational instance results, it is possible to more accurately detect that defects such as errors in the execution results of the business strategy script are caused by defects in the execution results of the target task, thus accurately detecting the cause of errors in the business strategy. Therefore, by sending defect information to the target agent used to execute the target task, and enabling the target agent to execute the target task according to the prompts of the defect information, the target business strategy script can more accurately repair the task according to the description of the defect information, improving the accuracy of the business strategy script, thereby improving the matching degree of the target business strategy with business requirements, and improving the execution efficiency and accuracy of the business strategy.

[0024] Figure 1 The illustration shows an exemplary system architecture for applying an agent-based business strategy determination method and apparatus according to embodiments of this application.

[0025] It is important to note that Figure 1 The examples shown are merely examples of system architectures applicable to the embodiments of this application, intended to help those skilled in the art understand the technical content of this application. They do not imply that the embodiments of this application cannot be used in other devices, systems, environments, or scenarios. For instance, in another embodiment, an exemplary system architecture applicable to the agent-based business policy determination method and apparatus may include a terminal device. However, the terminal device can implement the agent-based business policy determination method and apparatus provided in the embodiments of this application without interacting with the server.

[0026] like Figure 1 As shown, the system architecture 100 according to this embodiment may include a first terminal device 101, a second terminal device 102, a third terminal device 103, a network 104, and a server 105. The network 104 serves as a medium for providing communication links between the first terminal device 101, the second terminal device 102, the third terminal device 103, and the server 105. The network 104 may include various connection types, such as wired and / or wireless communication links, etc.

[0027] Users can use the first terminal device 101, the second terminal device 102, and the third terminal device 103 to interact with the server 105 via the network 104 to receive or send messages, etc. Various communication client applications can be installed on the first terminal device 101, the second terminal device 102, and the third terminal device 103, such as knowledge reading applications, web browser applications, search applications, instant messaging tools, email clients, and / or social platform software, etc. (for example only).

[0028] The first terminal device 101, the second terminal device 102, and the third terminal device 103 can be various electronic devices with displays and support web browsing, including but not limited to smartphones, tablets, laptops, and desktop computers.

[0029] Server 105 can be a server that provides various services, such as a backend management server that supports the content browsed by users using the first terminal device 101, the second terminal device 102, and the third terminal device 103 (this is just an example). The backend management server can analyze and process data such as received user requests, and feed back the processing results (such as web pages, information, or data obtained or generated according to user requests) to the terminal devices.

[0030] It should be noted that the agent-based service policy determination method provided in this application embodiment can generally be executed by the first terminal device 101, the second terminal device 102, or the third terminal device 103. Correspondingly, the agent-based service policy determination device provided in this application embodiment can also be disposed in the first terminal device 101, the second terminal device 102, or the third terminal device 103.

[0031] Alternatively, the agent-based service policy determination method provided in this application embodiment can generally be executed by server 105. Correspondingly, the agent-based service policy determination device provided in this application embodiment can generally be located in server 105. The agent-based service policy determination method provided in this application embodiment can also be executed by a server or server cluster that is different from server 105 and capable of communicating with the first terminal device 101, the second terminal device 102, the third terminal device 103, and / or server 105. Correspondingly, the agent-based service policy determination device provided in this application embodiment can also be located in a server or server cluster that is different from server 105 and capable of communicating with the first terminal device 101, the second terminal device 102, the third terminal device 103, and / or server 105.

[0032] It should be understood that Figure 1 The number of first terminal devices, second terminal devices, third terminal devices, networks, and servers shown in the diagram is merely illustrative. Depending on implementation needs, any number of first terminal devices, second terminal devices, third terminal devices, networks, and servers can be included.

[0033] Figure 2 A flowchart illustrating an agent-based business policy determination method according to an embodiment of this application is shown.

[0034] like Figure 2 As shown, the agent-based business strategy determination method includes operations S210~S250.

[0035] In operation S210, multiple first agents execute multiple tasks with dependencies based on demand information about business requirements, and determine business strategy scripts and basic computational examples.

[0036] In operation S220, the data of the basic example related to the model items of the algorithm model are perturbed to obtain the perturbed example.

[0037] In operation S230, the second agent performs defect assessment based on the results of the perturbation calculation and determines defect information related to the target task among multiple tasks.

[0038] In operation S240, the target agent corresponding to the defect information among multiple first agents is used to execute the target task based on the defect information to obtain the target business strategy script.

[0039] When operating S250, the target business strategy is determined based on the target business strategy script.

[0040] According to embodiments of this application, the requirement information can be any modal information, such as text or images, input by the user regarding business objectives and execution conditions. For example, the requirement information could be "Schedule equipment maintenance based on the current number of maintenance personnel." Another example is "Plan the power generation plan and fuel procurement based on the current operating conditions of wind power and gas-fired power generation equipment."

[0041] Multiple first agents collaborate to execute multiple tasks with dependencies, which can be represented as the execution logic or execution order between the tasks. Each first agent uses the results of its preceding task to execute subsequent tasks and outputs the results of those subsequent tasks, until the last first agent executes a task and outputs the business strategy script.

[0042] In some embodiments, a business strategy script represents an algorithmic model that matches business requirements. This model processes business data to output computational results that satisfy those requirements. For example, a business strategy script might be program code representing a linear programming algorithm. By running this code to process business data, computational results relevant to the business requirements can be output. This allows users to execute business strategies based on the business strategy script to meet actual business optimization needs.

[0043] For example, the algorithm model is the execution result output by the first agent when it performs the algorithm model construction task. Another first agent performs a code editing task based on the algorithm model, outputting a business strategy script and a basic calculation example.

[0044] Basic calculation cases can be test cases used to test business strategy scripts. Basic calculation cases can include parameter values ​​and feasible domain information related to business information such as the business type and operating conditions indicated by the requirement information, which the algorithm model can process. By running basic calculation cases according to the business strategy script, the algorithm model processes the business data and outputs basic calculation case results, ensuring that the basic calculation case results match the business requirements.

[0045] For example, the algorithm model is an optimization function model for optimizing fuel consumption, the basic calculation case includes business data related to the business scenario indicated by the demand information, and the result of the basic calculation case is the optimized fuel consumption.

[0046] In some embodiments, model terms of an algorithm model can be represented as function terms combining coefficients and variables in the algorithm model. For example, if the algorithm model includes the function 2x + by > 0, the model term can be "2x" or "by", where b is a coefficient and y is a variable. The example data associated with the model term can be used to represent the data values ​​corresponding to the coefficients or variables. Thus, the basic example results can be obtained by running the business strategy script to execute the basic example.

[0047] In some embodiments, the basic computational data can be perturbed by modifying or deleting the computational data according to preset perturbation rules. The perturbed computational results are then determined by executing the perturbed computational data based on the business strategy script. This allows the perturbed computational results to represent the degree of difference between the perturbed computational results output by the business strategy script and business requirements in diverse business scenarios, thereby enabling a more accurate assessment of the defects in the algorithm model or the business strategy script.

[0048] The second agent can be an agent used to evaluate the execution results of the tasks performed by multiple first agents. The second agent analyzes whether there are defects in the execution results of each task by processing the results of perturbation calculations, business strategy scripts and algorithm models. The task corresponding to the execution result with defects is identified as the target task. The second agent calls the large language model to output defect information such as the type of defect and the description of the defect information in the execution result.

[0049] By sending defect information to the target agent used to execute the target task, the target agent can execute the target task more clearly based on the defect type, defect description and other information in the defect information, so that the output target business strategy script can meet business needs.

[0050] In some embodiments, the target business strategy can be a file containing the target business strategy script, a program application, a business strategy scheme, etc. The target object can achieve diverse business objectives such as logistics scheduling, energy and material procurement, and road planning based on the target business strategy to meet actual business needs.

[0051] In some embodiments, utilizing multiple first agents to perform multiple dependent tasks based on requirement information regarding business needs may include: utilizing an intent understanding agent to perform an intent understanding task based on the requirement information, and outputting structured business intent information; utilizing an algorithmic agent to perform an algorithm model generation task on the business intent information, and outputting an algorithmic model; and utilizing a scripting agent to perform a code editing task based on the business intent information and the algorithmic model, and outputting a business strategy script.

[0052] In some embodiments, multiple first agents run in mutually isolated task execution environments. Multiple stage workspaces within the task execution environment are used by each first agent to perform tasks. The multiple first agents include intent-understanding agents, algorithmic agents, and scripting agents. Mutually isolated stage workspaces are created for the multiple first agents via an orchestrator configured on the server, with each stage workspace copying only the input files that the first agent is authorized to read.

[0053] The intent-understanding agent processes requirement information (Q) and business knowledge information by invoking a large language model. This allows it to understand the original business requirement intent of the target object by combining business knowledge, outputting structured business intent information. This information includes structured data related to business requirements, such as the business problem type, the business objective to be optimized, the target direction, the business object, parameter names, parameter data ranges, parameter units, parameter index semantics, business constraints, constraint identifiers, and ambiguities corresponding to business terms. The intent-understanding agent writes this structured business intent information into a structured file. During the execution of the intent-understanding task, the agent does not define decision variables for business strategies or generate algorithm functions; it is only responsible for transforming the original requirement information into structured information that clearly represents the business requirement intent. This allows the business intent information recognition task and the algorithm model generation task to be executed by different first agents. This enables the second agent to analyze the structured documents and requirement information Q to assess whether there are any execution defects in the intent understanding task. If the structured business intent information and requirement information Q are inconsistent, the algorithm model generation task is determined as the target task. This enables accurate attribution of defects in the business strategy script, improving the efficiency and accuracy of repairing the business strategy script.

[0054] The algorithmic agent executes the algorithm model generation task by reading only the structured business intent information in the structured file. Thus, the algorithmic agent can design the decision variables, variable types, variable upper and lower bounds, auxiliary variables, objective function, and constraints of the algorithm model without accessing the original requirement information Q, generating a purely symbolic algorithm model M without specific numerical values.

[0055] For example, each constraint and objective function in the algorithm model is assigned a mapping index to the corresponding business intent information. This allows the second agent to evaluate the matching degree between model items in the algorithm model and the business intent information, as well as the coverage of the business intent information by the algorithm model, based on the mapping index. Since the algorithm agent does not read the requirement information Q during the execution of the algorithm model generation task, the second agent can identify errors in the algorithm model M when the business intent information meets the accuracy condition by processing the algorithm model M and the business intent information. This allows for accurate assessment of defects in the execution of the algorithm generation task, enabling the algorithm generation model to be identified as the target task and the error cause of the repair business strategy script to be accurately identified. Furthermore, by sending defect information to the algorithm agent, the efficiency of repairing the business strategy script is improved, avoiding hidden error states where the error cause of the business strategy script is difficult to find, leading to errors in the code execution or calculation results of the business strategy script.

[0056] The script agent executes code editing tasks by reading business intent information S and algorithm model M. The script agent obtains the business strategy script C by mapping the optimization objectives and constraints defined in the algorithm model M into executable program code, and interprets the business intent information to output the basic calculation examples containing specific parameter values, vector representations, matrix and index semantics, etc., used to solve the algorithm model. The phase work area restricts the script agent from reading the original requirement information Q; it can only use the business intent information S as the calculation example data source and the algorithm model M as the code editing data source. This avoids redundant data reading by the script agent, which could lead to confusion in understanding business requirements or knowledge. Furthermore, the second agent can perform defect assessment by analyzing the consistency between the algorithm model and the business intent information. This allows for a more accurate analysis of defects in the intent understanding task or algorithm model generation task, preventing irreparable errors in the code execution results caused by merely fixing the code editing task.

[0057] By using isolated phased work areas involving multiple first agents, the first agents performing different tasks can strictly execute their current tasks according to the execution results of previous, dependent tasks. This prevents subsequent first agents from bypassing the execution results of previous tasks, ensuring strict logical coherence among the multiple execution results used to generate business strategy scripts. Consequently, by utilizing a second agent to evaluate the execution results output by multiple first agents, defective target tasks can be detected more accurately, improving the uniqueness and evaluability of target task identification and avoiding the black-box effect caused by agent task execution that makes it difficult to discover the cause of defects.

[0058] Figure 3A flowchart illustrating an agent-based business strategy determination method according to another embodiment of this application is shown.

[0059] like Figure 3 As shown, the agent-based business strategy determination method includes operations S301 to S307.

[0060] When operating S301, the intent understanding agent is used to understand the business intent of the demand information and output the business intent information.

[0061] In operation S302, the algorithmic agent performs an algorithm model generation task based on structured business intent information to generate an algorithm model that matches the requirement information.

[0062] When operating S303, the code agent performs code editing tasks based on business intent information and algorithm models, outputs business strategy scripts, and outputs basic calculation examples.

[0063] When operating S304, the basic example is perturbed by the test agent, and the perturbed example is output.

[0064] When operating S305, the test agent executes basic and perturbation calculations by running business strategy scripts, and outputs the results of the basic and perturbation calculations to complete the testing of the business strategy scripts.

[0065] When operating S306, the second intelligent agent performs defect assessment on the business strategy script based on the results of the basic calculation and the results of the disturbance calculation, and outputs defect information.

[0066] In operation S307, the second agent sends defect information to the target agent related to the target task, so that the target agent can re-execute the target task and output the target business strategy script that meets the user's business needs.

[0067] The test agent receives a basic computational example D, structured business intent information S, an algorithm model M, and a business strategy script C. It then invokes a large language model to perform controlled adjustments on the computational power data values ​​related to model items in the basic computational example D, generating multiple deformed perturbation examples D'_j. This controlled adjustment process does not rewrite the business semantics of the computational data in the basic example. Instead, while maintaining the problem structure and constraint types, it performs rule-based modifications to the data capacity, business cost, upper and lower bounds of business data, business data indexes, or model parameter values ​​of the computational data to generate perturbation scripts used to test the correctness of the algorithm model or business strategy script.

[0068] In some embodiments, the algorithm model includes an objective function and constraints to meet business requirements. Perturbed computational examples are determined by perturbing the computational data corresponding to at least one of the model terms in the objective function and constraints. For example, perturbed computational examples can be obtained by adjusting the variable or parameter values ​​of the model terms in the constraints. Alternatively, perturbed computational examples can also be obtained by adjusting the computational data corresponding to the parameter values ​​in the objective function.

[0069] Using a second intelligent agent to perform defect assessment based on the results of disturbance calculations may include: using the second intelligent agent to process the results of disturbance calculations and the results of basic calculations based on the business optimization objectives represented by the objective function, and obtaining defect information.

[0070] In some embodiments, the objective function can represent a business optimization goal that is to maximize or minimize a business metric related to business needs. For example, the business optimization goal could be to maximize energy utilization, or it could be to minimize electricity consumption.

[0071] The second intelligent agent analyzes the differences between the results of the perturbation examples and the results of the basic examples by processing requirement information, algorithm models, and business strategy code. This allows the second intelligent agent to output defect information after understanding the differences in the computational results between the perturbation examples and the basic examples. The defect information describes the degree of difference between the objective function's target direction and the business requirement, or the degree of difference between the constraint direction of the constraint conditions and the business requirement.

[0072] For example, the objective function in the algorithm model is max(ax+by), where x and y represent the electricity consumption and fuel consumption, respectively. Since the optimization objective represented by the demand information is to minimize the energy consumption cost, the second agent can be used to understand the results of the disturbance case and the results of the basic case by interpreting the business optimization objective represented by the objective function. If the difference between the target direction of the output objective function and the business demand does not meet the preset conditions, it is determined that the target direction of the objective function of the defect information represents the algorithm model is wrong.

[0073] For example, the constraint in the algorithm model is a1x1+b1y1>R, where x1 and y1 represent the sorting operation time and the cargo transportation time, respectively, and R is the feasible region of cargo logistics time. Since the optimization objective of the demand information representation is to minimize the logistics time, the second intelligent agent can be used to understand the results of the disturbance example and the results of the basic example by understanding the business optimization objective represented by the objective function. The degree of difference between the constraint direction of the output constraint and the business demand does not meet the preset conditions, thus determining that the constraint direction of the defect information representation algorithm model constraint is incorrect.

[0074] The second intelligent agent assesses defects in demand information, business intent information, algorithm models, business strategy scripts, and the results of basic and perturbation calculations. It can accurately analyze the differences between the computational results obtained from perturbation and basic calculations in the execution of business strategy scripts, based on business requirements. This allows for a more precise determination of whether there are errors in the target or constraint directions of the algorithm model, or errors in the code of the business strategy script, based on the changes in computing power data represented by the perturbation rules and the differences between the perturbation and basic calculation results. Furthermore, it can more accurately identify errors between the objective function or constraints in complex algorithm models and the business demand information, preventing situations where the business strategy script runs normally but the cause of errors in the execution result is difficult to find. This improves the efficiency of repairing business strategy scripts and the efficiency of executing business optimization requirements.

[0075] In some embodiments, the perturbation case includes a first perturbation case, which is determined by adjusting the case data related to the business feasible domain item in the model item according to a preset adjustment direction.

[0076] The business feasibility domain term can represent the feasibility domain information in the objective function or constraints. For example, feasibility domain information can be the feasibility domain information in the constraints.

[0077] The constraint is: st a2x2+b2y2≤K, where x2 represents the number of screws, y2 represents the number of nuts, a2 and b2 represent the material consumption of the screws and nuts respectively, and K is the total material. K can be a business feasible domain term. In the basic example, the data related to the total material is 40, and in the perturbation example, the data related to the total material is 50. Therefore, the perturbation example can be determined by adjusting the data related to the business feasible domain term in the model according to the direction of numerical scaling.

[0078] It should be noted that adjusting the example data related to the business feasible domain items in the model items according to the preset adjustment direction can include expanding or shrinking the data value of the computing power data, thereby relaxing or tightening the business feasible domain items, so that the perturbation example results can be represented in the execution process of adjusting the algorithm model and business strategy script, and more accurately discover whether the target direction in the algorithm model meets the business requirements.

[0079] The second intelligent agent processes the results of the disturbance case and the results of the basic case based on the business optimization objective represented by the objective function. This includes the second intelligent agent performing the following operations: comparing the results of the disturbance case and the results of the basic case to obtain the comparison results; and using the comparison results to evaluate the defects of the objective function and constraints based on the prompt information describing the preset adjustment direction and business optimization objective to obtain defect information.

[0080] In one example, the objective function aims to minimize resource consumption. This is achieved by expanding the computational data corresponding to the feasible business domain, thereby relaxing the resource constraints of that domain. Therefore, the formula z(D'_relax) ≤ z(D) represents the comparison between the results of the perturbation computational instance and the results of the basic computational instance. Here, D'_relax represents the perturbation computational instance, D represents the basic computational instance, and z() represents the execution process of the business strategy script.

[0081] For the objective direction to minimize resource consumption, the result of the perturbation example should be less than the result of the basic example, i.e., z(D'_relax) ≤ z(D). If the comparison result does not satisfy z(D'_relax) ≤ z(D), it indicates that the code or model has a high probability of target direction error, constraint direction error, or parameter mapping error. Therefore, based on the comparison result, the second agent can determine whether the business strategy script or algorithm model has target direction error or constraint direction error.

[0082] For an algorithmic model representing a maximization problem, if resource constraints are relaxed by expanding the feasible region by increasing the example data corresponding to the business feasible region terms, thus expanding the feasible region, then z(D'_relax) ≥ z(D) should be satisfied. If the comparison result indicates that z(D'_relax) ≥ z(D) is not satisfied, the second agent can perform defect assessment to determine that the business strategy script or algorithmic model has a high probability of target direction error, constraint direction error, or parameter mapping error. This allows for a more accurate determination of the target task, instructing the target agent to repair the algorithmic model and business strategy script, or only repair the business strategy script, to improve the generation efficiency of the target strategy script.

[0083] In some embodiments, the second agent can also perform defect assessment based on the accuracy rates of the perturbation example results and the basic example results. If the accuracy rate of the perturbation example results decreases when the example data corresponding to the business feasible domain increases, it may indicate an error in the summation term in the objective function, causing a defect in the algorithm model described by the second agent's defect information. Alternatively, the second agent can also perform defect assessment by combining the accuracy comparison results with the constraints corresponding to the business feasible domain items. If a decrease in the accuracy rate of the perturbation example results is detected, it is determined that the constraint settings are defective. Thus, the defect information output by the monotonicity test can more accurately describe the defects of the perturbed constraints, prompting the algorithm agent or script agent to re-execute the target task, improving the accuracy and efficiency of generating the target business strategy.

[0084] In some embodiments, perturbing the example data related to the model items of the algorithm model in the basic example may further include: using a third agent to analyze the algorithm model based on demand information to determine the target parameters in the algorithm model; and scaling the parameter data related to the target parameters in the basic example based on a scaling factor to obtain the perturbed example.

[0085] The third agent can be an agent used to perturb the basic computational instance. For example, the third agent can be a test agent used to perturb the basic computational instance and execute the basic computational instance and the perturbation computational instance by running a business strategy script.

[0086] According to embodiments of this application, a scaling factor can be used to enlarge or reduce the data in a computational study. For example, a scaling factor k can be used to proportionally scale the parameter data related to the target parameter in the basic computational study. When k is greater than 1, the parameter data is enlarged; when k is less than 1 but greater than 0, the parameter data is reduced. This results in a perturbation computational study obtained after proportional scaling, which can be a second perturbation computational study.

[0087] In some embodiments, the defect information describes the ratio of the perturbation example result to the base example result, and the difference between the scaling factors.

[0088] For example, by scaling the target parameter data in the base case D proportionally to a scaling factor k greater than 1, resulting in a second perturbation case D'_scale = k×D, we obtain the second perturbation case D'_scale. Here, k is greater than 1. Theoretically, the result of the perturbation case should satisfy z(D'_scale) = k×z(D) with the result of the base case. Therefore, by using a second agent to analyze the ratio between the perturbation case result and the result of the base case, and the difference between the scaling factor and the ratio, we can perform defect assessment on the algorithm model or business strategy script. This allows us to determine whether the magnitude and direction of change in the perturbation case result are consistent with the magnitude and direction of the scaling factor's adjustment of the case data, thus more accurately identifying defects in the algorithm model or business strategy script. This allows for more precise identification of the target agent and target task based on defect information, and accurate description of the defective model item or code segment. This helps avoid errors in the execution results of business strategies due to defects in the target parameter settings in the algorithm model, or more accurately detects defects in the business strategy script such as omissions in the calculation of target parameters or incorrect units. This improves the efficiency of repairing business strategy scripts and avoids the inability to accurately output the target business strategy due to the difficulty in finding the cause of errors in the business strategy script.

[0089] In some embodiments, the second agent is used to perform defect assessment based on the results of the perturbation case, including: using the second agent to analyze the results of the perturbation case and the results of the base case based on the scaling factor, and outputting defect information.

[0090] According to embodiments of this application, the defect information indicates at least one of the following defects in the business strategy script: incorrect setting of the magnitude of business data; incorrect setting of the data unit attribute of business data; or code execution error in the code related to the target parameter.

[0091] An incorrect setting of the magnitude of business data, such as setting the target parameter for resource storage in a code segment of a business strategy script to 1000 times the base storage capacity, can lead to difficulties in accurately calculating the precise resource consumption resulting from the perturbation execution.

[0092] If the data unit attribute setting of business data is incorrect, for example, the storage unit attribute "ton" in the code segment of the business strategy script can be changed to the unit attribute "kilogram". This can be achieved by using a second intelligent agent to analyze the results of the disturbance case and the basic case, so as to more accurately identify the defective code segment in the business strategy script. This allows the script intelligent agent to accurately repair the business strategy script and improve the generation efficiency and execution accuracy of the target business strategy.

[0093] For code execution errors related to target parameters, such as omissions in multiplication of business parameters in code segments within the business strategy script, the second agent analyzes the differences between the perturbation example results and the base example results by examining the target parameters corresponding to the perturbation example results. This allows for precise detection of editing omissions or errors in the code related to target parameters within the business strategy script. The defect information then prompts the script agent to accurately correct these defects, improving the efficiency and accuracy of target business strategy generation.

[0094] It should be understood that defect information can also be used to describe one or more of the above defects, so that the script agent can accurately repair the business strategy script based on a more accurate algorithm model, thereby improving the execution efficiency of business optimization.

[0095] In some embodiments, perturbing the instance data related to the model items of the algorithm model in the basic instance may further include: updating the business data index related to the model items in the basic instance based on the perturbed business data index to obtain the perturbed instance.

[0096] The perturbation business index is used to map other business data that differs from the basic business data. There is a mapping relationship between the basic business data and the business data index. For example, the basic data in the basic example corresponds to the basic business data "x" in the model item ax. By replacing the business data index in the basic example with the perturbation business index, when the business strategy script executes the perturbation example, the perturbation business data corresponding to the perturbation business data "x1" can be processed through the business strategy script to obtain the perturbation example result. The defect information describes an error in the code used to map business data in the business strategy script.

[0097] In some embodiments, the business data indexes in the base example can be reordered or the matrix dimensions corresponding to the indexes can be modified. This allows the results of the perturbation example to be theoretically represented as z(D'_perm) = z(D), where D'_perm is the third perturbation example obtained by perturbing the business data indexes. The second agent detects whether there are errors in the business data index content, matrix identifier editing, or target parameter settings in the business strategy script by measuring the magnitude of data change between the perturbation example result of the third perturbation example and the result of the base example, thereby improving the accuracy of defect identification in the business strategy script.

[0098] In some examples, perturbation of the business data index is achieved by adding a row or column of data to the matrix-form case data in the base case, or by modifying the content of the matrix data in the case data, thus obtaining perturbed cases. A second agent can then analyze the results of the perturbed cases and the base case results to determine whether the code corresponding to the business data semantics in the business strategy script has been accurately edited. This allows for precise identification of errors in the code used to map business data in the business strategy script, avoiding omissions or misinterpretations, and improving the efficiency of generating the target business strategy.

[0099] In some examples, a test agent is used to automatically select the perturbation method for the basic test case based on the business type corresponding to the requirement information, and then executes the process of perturbing the basic test case to generate perturbed test cases. Simultaneously, the perturbation process, the perturbed test cases, and the results of the perturbed test cases are associated and stored as test evidence, which is then sent to a second agent. This allows the second agent to evaluate the execution results of multiple tasks based on the test evidence and output defect information. This avoids the hidden error where the business strategy script can pass the basic test case but not the diverse business test cases, improving the speed of fixing the business strategy script and enhancing the accuracy of the target business strategy.

[0100] In some embodiments, the defect information is determined by data compression of the disturbance case results, the basic case results, and the defect description information. The basic case results are determined by executing the basic case based on the business strategy script, and the defect description information is determined by the second intelligent agent by analyzing the disturbance case results and the basic case results based on the requirement information.

[0101] In some examples, the second agent can perform multi-level defect assessment by reading information related to the executed tasks, such as requirement information Q, business intent information S, algorithm model M, business strategy script C, perturbation cases, basic cases, and results of perturbation cases and basic cases. It can also read the perturbation mode summary and test process summary output by the test agent.

[0102] The first layer of evaluation assesses the consistency between the requirement information Q and the business intent information S, determining whether there are errors in target direction, key parameters, core constraints, or index semantic extraction. The second layer evaluates the consistency between the business intent information S and the algorithm model M, checking whether the objective function and constraints are correctly represented by the algorithm model M and whether they cover the business intent information S. The third layer evaluates the consistency between the algorithm model M, the business intent information S, and the business strategy script C, checking whether the code fully implements the objective function and constraints, whether the variable types are correct, and whether the parameter dimensions in the example data are consistent with the code representing the business data index in the business strategy script. The fourth layer of evaluation combines the results of solving the perturbation example, the results of the basic example, and other test evidence to determine the feasibility, integerity, target order of magnitude, and business rationality of the business strategy script. This allows the second agent to complete the defect assessment and output descriptive information describing the defect type, target task identifier, defective target parameters, constraints, etc. The defect information records a structured field indicating the defect type to indicate the target task that needs to be restarted, and executes the target task and related tasks that depend on the target task, thus avoiding the need to re-execute multiple tasks and thus re-execute the entire task chain.

[0103] In some examples, an orchestrator schedules multiple first agents, second agents, and test agents. The orchestrator sends defect information to the target agent corresponding to the target task by performing a routing process on the defect type field in the defect information.

[0104] For example, this routing process can be formally represented as: A_(t+1) = R(V_t), where A_(t+1) represents the target agent that needs to re-execute the target task in round t+1, and R() is the routing function. V_t represents the defect type identifier in the defect information. When the defect type represented by V_t is an error in business intent information, R(V_t) points to the intent understanding agent. When V_t represents an algorithm model defect, R(V_t) points to the algorithm agent. When V_t represents a code defect in the business strategy script, R(V_t) points to the script agent. Therefore, the reason for the defect in the business strategy script can be accurately represented based on the routing function, and the corresponding target agent can be instructed to perform precise repair.

[0105] Figure 4 A schematic diagram of the structure of a service strategy orchestration system according to an embodiment of this application is shown.

[0106] like Figure 4As shown, the business strategy orchestration system may include an orchestrator, intent-understanding agents, algorithm agents, script agents, testing agents, evaluation agents, and a memory module. The intent-understanding agents, algorithm agents, and script agents constitute multiple first agents. The evaluation agent is a second agent, and the testing agent is a third agent. The orchestrator is responsible for scheduling the multiple agents and the memory module.

[0107] Intent understanding agents can perform intent understanding tasks by calling rule parsers and table extraction programs, or by calling large language models and rule engines to collaborate on intent understanding tasks.

[0108] Algorithmic agents can perform algorithmic model generation tasks by calling the base large language model and the fine-tuned operations research expert large model. Alternatively, algorithmic agents can use template retrieval tools and rule completion tools to assist in calling the large model to perform algorithmic model generation tasks.

[0109] The script agent executes code editing tasks by calling a large language model to process business intent information and an algorithm model, and outputs business strategy scripts.

[0110] As a third-party intelligent agent, the test agent can perturb the base test case by invoking one or more large language models to obtain perturbation test cases. It then executes the business strategy scripts by invoking code execution tools to run both the perturbation test cases and the base test case. Furthermore, the test agent can also generate perturbation test cases by invoking rule-based transformation tools, programmatic data perturbation tools, or knowledge-based test case generation tools.

[0111] The evaluation agent can call tools such as large language models, rule review tools, theorem proving tools, and solvers to evaluate the consistency of the execution results of multiple tasks, as well as to evaluate the defects of the results of perturbation examples and the results of basic examples, and output defect information.

[0112] The memory module processes uncompressed defect information and outputs compressed defect information, such as defect description summaries. The compressed defect information can also be stored in file formats such as vector retrieval indexes, knowledge graphs, or structured JSON files. This facilitates the compression of historical errors into reusable short contexts, enabling the target agent to understand the defect content and perform the target task, thereby reducing the data computation overhead incurred by each first agent during the execution of the target task and related tasks.

[0113] For example, the memory module receives test reports, evaluation reports, and related perturbation examples and their execution results from the test agent, and compresses the defect description information into a reusable structured or semi-structured summary μ_t. The summary μ_t includes at least: defect type, defect triggering conditions, constraints or target model items containing the defect, defect examples, repair principles, and counterexample descriptions. The compression process can be formally represented as: μ_t = Compress(μ_(t-1), V_t, E_t), where E_t represents the set of test evidence generated in the t-th round of defect evaluation. An orchestrator can then be used to schedule the memory module to compress the summary μ_t along with the corresponding business intent information S and algorithm model M from the t-th round to obtain the context content for the target agent to execute the target task. The context content context_(t+1) is represented as an array [Q or S or M, μ_t]. This allows the target agent to execute the target task in the (t+1)-th round, facilitating the test agent's execution of the (t+1)-th round of defect evaluation. After introducing a memory module to compress defect information, the business strategy system forms a closed-loop execution process of "strategy generation—testing—evaluation—compression—regeneration". This closed-loop execution process replaces long historical information with short summaries, reducing the system's storage and large model inference burden, improving the utilization of context content, and enabling defect experience to be transferred across rounds and issues. It avoids the redundancy of context content information caused by directly accumulating complete historical dialogues during multiple rounds of task execution. By shortening the length of prompts used to execute the target task, the token consumption of context content is reduced, and redundant defect descriptions are decreased, thereby improving the targeting of the business strategy script for the next round of target task execution. Simultaneously, it enables cross-business type transfer, using defect information as error experience in the business strategy generation process of other business types to prompt the first agent to execute tasks, thereby suppressing the recurrence of similar defects in the execution results.

[0114] After the testing agent outputs a conclusion that the evaluation has passed, or when the number of defect evaluation rounds reaches a preset threshold, the orchestrator instructs the termination of the task corresponding to the business strategy determination method. The target business strategy output by the business strategy system includes at least: the target business strategy script, structured business intent information, symbolic algorithm model, test report, verification report, and compressed defect information. If the upper limit of defect evaluation rounds is reached but the evaluation is still not fully passed, a result with an uncertainty indicator can be output, allowing for manual intervention to review and modify the target strategy script, reducing the workload of manual repair.

[0115] The agent-based business strategy determination method of this application reduces unnecessary context information processing by each agent during task execution by utilizing multiple first agents to execute tasks in mutually isolated stage work areas. A second agent performs defect assessment based on perturbation example results to more accurately analyze defects occurring in the task execution of each first agent, facilitating precise location of errors in the business strategy script and reducing computational overhead redundancy caused by repeated execution across the entire chain. By utilizing the second agent and a test agent to collaboratively complete defect assessment, the code of the business strategy script of the algorithm model can be tested and analyzed more accurately, significantly improving the detection rate of hidden errors, index misalignment defects, order-of-magnitude errors, and constraint direction errors, thus enhancing the accuracy of defect attribution. This allows multiple agents to execute their respective tasks using the same large model, reducing the inference resource consumption cost required to generate the target business strategy, shortening the error correction path, and improving system stability.

[0116] In some embodiments, the demand information includes business requirements that represent optimizing energy utilization efficiency, and the target business strategy is used to execute an energy allocation strategy that allocates the amount of energy produced.

[0117] For example, business requirements represent minimizing the electricity consumption of the production workshop. The target business strategy includes a target business strategy script for executing the objective function and constraints. The target object inputs variables for the target business strategy, causing the target business strategy script to output the calculation results of control signal parameters such as operating power and operating time of multiple devices in the workshop, and output the electricity allocated to each device. This enables the target object to implement an energy allocation strategy for distributing electricity according to the target business strategy.

[0118] In some embodiments, the agent-based business strategy determination method provided in this application can be used in supply chain and logistics scenarios. It enables the input of demand information corresponding to business needs such as warehouses, factories, customers, transportation costs, inventory capacity, and energy demand constraints into a system based on the method provided in this application. The intent understanding agent outputs energy and material transportation scheduling data, the algorithm agent generates algorithmic models such as transportation route optimization models and site selection optimization models, and the script agent outputs business strategy scripts. By testing and evaluating the agent, the business strategy scripts are adjusted to further optimize transportation costs and energy consumption costs, thereby improving the effectiveness of the target business strategy for supply chain optimization.

[0119] In some embodiments, the agent-based business strategy determination method provided in this application can be used in production planning and scheduling scenarios. Work order information, machine performance, shift scheduling rules, employee attributes, time window constraints, etc., are input as demand information into a system based on the method provided in this application. The system can automatically generate a scheduling or scheduling model and use a test agent to perturb example data such as working hour limits, order priorities, and resource quantities to verify whether the business strategy script correctly handles various types of constraints such as time constraints and production capacity constraints.

[0120] In some embodiments, the agent-based business strategy determination method provided in this application can be used in facility site selection and resource allocation scenarios. Candidate sites, fixed construction costs, service coverage radius, budget constraints, and service requirements can be input as demand information into a system based on the method provided in this application. The system can automatically complete site selection modeling and check the correctness of the objective function and service coverage constraints through index rearrangement and budget relaxation tests.

[0121] In some embodiments, the agent-based business strategy determination method provided in this application can be used in energy, finance, and industrial optimization scenarios. It can input demand information such as power generation plans, energy storage constraints, investment ratios, risk constraints, and equipment maintenance strategies into a system based on the method provided in this application. The system can automatically identify different problems as network flow or constraint programming problems, and combine test agents and evaluation agents to improve the operational reliability of applications corresponding to the target business strategy.

[0122] In some embodiments, the agent-based business strategy determination method provided in this application can be used in education and R&D support scenarios. After teachers or researchers input a question, the system not only provides the final solution, but also outputs structured business intent information, the modeling process of the algorithm model, the editing process of the code script, and a defect description report, which can be used for teaching demonstrations, code reviews, and model debriefings.

[0123] Overall, this invention is applicable to a general process of "natural language input, formal modeling output, solver execution, automatic testing and verification, and error experience accumulation", and therefore has strong cross-industry transferability.

[0124] Figure 5 A block diagram of an agent-based business policy determination apparatus according to an embodiment of this application is illustrated schematically.

[0125] like Figure 5 As shown, the agent-based business strategy determination device 500 includes: a first determination module 510, a disturbance example acquisition module 520, a second determination module 530, a target business strategy script acquisition module 540, and a third determination module 550.

[0126] The first determining module 510 is used to utilize multiple first intelligent agents to execute multiple tasks with dependencies based on the demand information about business requirements, and to determine the business strategy script and basic calculation example. The business strategy script represents the algorithm model that matches the business requirements.

[0127] The perturbation case acquisition module 520 is used to perturb the case data related to the model items of the algorithm model in the basic case to obtain perturbation cases.

[0128] The second determination module 530 is used to perform defect assessment based on the results of the perturbation calculation using the second intelligent agent, and to determine the defect information related to the target task among multiple tasks. The results of the perturbation calculation are determined according to the perturbation calculation executed by the business strategy script.

[0129] The target business strategy script acquisition module 540 is used to utilize the target intelligent agent corresponding to the defect information among multiple first intelligent agents to execute the target task according to the defect information and obtain the target business strategy script.

[0130] The third determination module 550 is used to determine the target business strategy based on the target business strategy script.

[0131] According to an embodiment of this application, the algorithm model includes an objective function and constraints for meeting business requirements. The perturbation case is determined by perturbing the case data corresponding to at least one of the model items in the objective function and constraints. The second determination module includes a first obtaining unit.

[0132] The first obtaining unit is used by the second intelligent agent to process the disturbance case results and the basic case results according to the business optimization target represented by the objective function to obtain defect information. The basic case results are determined by executing the basic case according to the business strategy script. The defect information describes the degree of difference between the objective direction of the objective function and the business requirements, or describes the degree of difference between the constraint direction of the constraint conditions and the business requirements.

[0133] According to an embodiment of this application, the perturbation case includes a first perturbation case, which is determined by adjusting the case data related to the business feasible domain item in the model item according to a preset adjustment direction; the first obtaining unit is configured to use a second intelligent agent to perform the following operations: compare the perturbation case result with the basic case result to obtain a comparison result; and, based on the prompt information describing the preset adjustment direction and business optimization objective, use the comparison result to perform defect evaluation on the objective function and constraints to obtain defect information.

[0134] According to an embodiment of this application, the disturbance calculation module includes: a first determining unit and a second obtaining unit.

[0135] The first determining unit is used to analyze the algorithm model based on the demand information using the third intelligent agent and determine the target parameters in the algorithm model.

[0136] The second obtaining unit is used to scale the parameter data related to the target parameter in the basic calculation case based on the scaling factor to obtain the perturbation calculation case. The defect information describes the ratio between the perturbation calculation case result and the basic calculation case result, and the difference between the scaling factor. The basic calculation case result is determined by executing the basic calculation case according to the business strategy script.

[0137] According to an embodiment of this application, the second determining module includes a defect information output unit.

[0138] The defect information output unit is used to utilize the second intelligent agent to analyze the results of the disturbance calculation and the basic calculation based on the scaling factor, and output defect information. The defect information indicates at least one of the following defects in the business strategy script: incorrect setting of the magnitude of the business data; incorrect setting of the data unit attribute of the business data; code execution error in the code related to the target parameter.

[0139] According to an embodiment of this application, the perturbation example acquisition module includes an update unit.

[0140] The update unit is used to update the business data index related to the model item in the basic case based on the perturbation business data index to obtain the perturbation case. The perturbation business index is used to map other business data that are different from the basic business data. The basic business data and the business data index have a mapping relationship. The defect information describes that there is an error in the code used to map business data in the business strategy script.

[0141] According to an embodiment of this application, the first determining module includes: a business intent information output unit, an algorithm model output unit, and a business strategy script output unit.

[0142] The business intent information output unit is used to utilize the intent understanding agent to perform intent understanding tasks based on demand information and output structured business intent information.

[0143] The algorithm model output unit is used to perform the algorithm model generation task based on the business intent information by the algorithmic agent and output the algorithm model.

[0144] The business strategy script output unit is used to execute code editing tasks based on business intent information and algorithm models using a script agent, and output business strategy scripts.

[0145] According to an embodiment of this application, the defect information is determined by compressing the disturbance case results, the basic case results, and the defect description information. The basic case results are determined by executing the basic case based on the business strategy script. The defect description information is determined by the second intelligent agent by analyzing the disturbance case results and the basic case results based on the demand information.

[0146] According to an embodiment of this application, the demand information includes business requirements that represent optimizing energy utilization efficiency, and the target business strategy is used to execute an energy allocation strategy for allocating the amount of energy produced.

[0147] Figure 6 A schematic diagram illustrating the structure of an intelligent agent of artificial intelligence according to an embodiment of this application is shown.

[0148] In the embodiments of this application, such as Figure 6 As shown, the AI ​​agent 600 may include an input module 610, a processing module 620, and an output module 630.

[0149] Input module 610 is used to receive input information.

[0150] The processing module 620 is used to determine the target task based on the input information received by the input module, determine the large model based on the target task, and obtain output information by calling the large language model to execute the agent-based business strategy determination method provided in the embodiments of this application.

[0151] Output module 630 is used to output the output information obtained by the processing module.

[0152] According to embodiments of this application, the input module 610 is responsible for receiving or sensing information such as queries, requests, instructions, signals, or data from the outside world (e.g., users or the external environment), and converting it into a format that the AI ​​agent 600 can understand and process. The input module 610 is the primary link for the AI ​​agent 600 to interact with the outside world, enabling the AI ​​agent 600 to efficiently and accurately obtain necessary "sensory" information from the outside world and respond to this information.

[0153] In the example, input module 610 can input the requirement information, business strategy scripts, etc. described above.

[0154] In the example, processing module 620 is the core support for the AI ​​agent 600's ability to handle complex tasks. Processing module 620 can execute the agent-based business strategy determination method described above.

[0155] In the example, the performance of processing module 620 is closely related to the large model on which the AI ​​agent 600 is based. To fully leverage the capabilities of the large model, the internal structure of processing module 620 can be designed to be highly configurable and scalable to handle various types of tasks and requirements in real-world scenarios.

[0156] In the example, after the AI ​​agent 600 obtains the demand information in the form of voice, the processing module 620 can use the large voice recognition model to process the demand information to obtain the business strategy script and basic calculation example, and use the large model to process the perturbation calculation example results to obtain the defect information, and then pass the defect information to the output module 630.

[0157] Understandably, while large language models possess excellent language understanding and generation capabilities, like humans, their ability to solve tasks is limited without the aid of any tools. Once the AI ​​agent 600 is given the ability to invoke tools, it can perform tasks such as using a calculator to complete mathematical calculations, using Python to perform data analysis, and using a search engine to generate weather forecasts.

[0158] In the example, output module 630 can output the defect information described above.

[0159] The AI ​​agent 600 according to the embodiments of this application can simply and effectively improve the level of intelligence, and enhance flexibility and versatility.

[0160] According to embodiments of this application, this application also provides an electronic device, a readable storage medium, and a computer program product.

[0161] According to an embodiment of this application, an electronic device includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the method described above.

[0162] According to an embodiment of this application, a non-transitory computer-readable storage medium storing computer instructions, wherein the computer instructions are used to cause a computer to perform the method described above.

[0163] According to an embodiment of this application, a computer program product includes a computer program that, when executed by a processor, implements the method described above.

[0164] Figure 7A schematic block diagram of an example electronic device is shown that can be used to implement the agent-based business policy determination method of embodiments of this application. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the present application described and / or claimed herein.

[0165] like Figure 7 As shown, the electronic device 700 includes a computing unit 701, which can perform various appropriate actions and processes according to a computer program stored in a read-only memory (ROM) 702 or a computer program loaded from a storage unit 708 into a random access memory (RAM) 703. The RAM 703 may also store various programs and data required for the operation of the electronic device 700. The computing unit 701, ROM 702, and RAM 703 are interconnected via a bus 704. An input / output (I / O) interface 705 is also connected to the bus 704.

[0166] Multiple components in electronic device 700 are connected to input / output (I / O) interface 705, including: input unit 706, such as keyboard, mouse, etc.; output unit 707, such as various types of monitors, speakers, etc.; storage unit 708, such as disk, optical disk, etc.; and communication unit 709, such as network card, modem, wireless transceiver, etc. Communication unit 709 allows electronic device 700 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0167] The computing unit 701 can be various general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 701 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 701 performs the various methods and processes described above, such as an agent-based business policy determination method. For example, in some embodiments, the agent-based business policy determination method can be implemented as a computer software program tangibly contained in a machine-readable medium, such as storage unit 708. In some embodiments, part or all of the computer program can be loaded and / or installed on the electronic device 700 via ROM 702 and / or communication unit 709. When the computer program is loaded into RAM 703 and executed by the computing unit 701, one or more steps of the agent-based business policy determination method described above can be performed. Alternatively, in other embodiments, the computing unit 701 may be configured to perform an agent-based business policy determination method by any other suitable means (e.g., by means of firmware).

[0168] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), payload-programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0169] The program code used to implement the methods of this application may be written in any combination of one or more programming languages. This program code may be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing device, such that when executed by the processor or controller, the functions / operations specified in the flowcharts and / or block diagrams are implemented. The program code may be executed entirely on a machine, partially on a machine, as a standalone software package partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0170] In the context of this application, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. Machine-readable media can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0171] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device (e.g., a cathode ray tube or liquid crystal display) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the computer. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0172] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as a data server), or computing systems that include middleware components (e.g., an application server), or computing systems that include frontend components (e.g., a user computer with a graphical user interface or web browser through which a user can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., a communication network). Examples of communication networks include local area networks (LANs), wide area networks (WANs), and the Internet.

[0173] Computer systems can include clients and servers. Clients and servers are generally located far apart and typically interact via communication networks. Client-server relationships are created by computer programs running on the respective computers and having a client-server relationship with each other. Servers can be cloud servers, distributed system servers, or servers incorporating blockchain technology.

[0174] It should be understood that the various forms of processes shown above can be used to rearrange, add, or delete steps. For example, the steps described in this disclosure can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution disclosed in this application can be achieved, and this is not limited herein.

[0175] The specific embodiments described above do not constitute a limitation on the scope of protection of this application. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A method for determining business strategies based on intelligent agents, characterized in that, The method for determining the business strategy includes: Multiple first intelligent agents execute multiple tasks with dependencies based on demand information about business requirements, and determine business strategy scripts and basic computational examples. The business strategy scripts represent algorithm models that match the business requirements, and the business requirements include at least one of business objectives and business execution conditions. The data related to the model terms of the algorithm model in the basic example are perturbed to obtain perturbed examples; The second intelligent agent performs defect assessment based on the results of the perturbation calculation to determine defect information related to the target task among the multiple tasks. The results of the perturbation calculation are determined by executing the perturbation calculation according to the business strategy script. By utilizing the target intelligent agent among the multiple first intelligent agents that corresponds to the defect information, the target task is executed according to the defect information to obtain the target business strategy script; Determine the target business strategy based on the target business strategy script; In this system, multiple first-level intelligent agents operate in mutually isolated task execution environments. These agents execute multiple tasks with dependencies based on requirements information related to business needs, including: The intent-understanding agent performs intent-understanding tasks based on the aforementioned requirement information and outputs structured business intent information. The algorithmic agent performs an algorithmic model generation task on the business intent information and outputs the algorithmic model. The script agent executes code editing tasks based on the business intent information and the algorithm model, and outputs the business strategy script. The algorithm model includes an objective function and constraints to satisfy the business requirements. The perturbed computational examples are determined by perturbing the computational data corresponding to at least one of the model terms in the objective function and constraints. The defect assessment using a second intelligent agent based on the perturbed computational example results includes: The second intelligent agent processes the disturbance case results and the basic case results according to the business optimization objective represented by the objective function to obtain the defect information. The basic case results are determined by executing the basic case according to the business strategy script. The defect information describes the degree of difference between the objective direction of the objective function and the business requirement, or describes the degree of difference between the constraint direction of the constraint condition and the business requirement.

2. The method according to claim 1, characterized in that, The disturbance case includes a first disturbance case, which is determined by adjusting the case data related to the business feasible domain item in the model item according to a preset adjustment direction; The step of using the second intelligent agent to process the perturbation test results and the basic test results according to the business optimization objective represented by the objective function includes using the second intelligent agent to perform the following operations: The results of the disturbance case are compared with the results of the basic case to obtain the comparison results; Based on the prompt information describing the preset adjustment direction and the business optimization goal, the objective function and the constraints are evaluated for defects using the comparison results to obtain the defect information.

3. The method according to claim 1, characterized in that, The perturbation of the example data related to the model terms of the algorithm model in the basic example includes: A third-party intelligent agent analyzes the algorithm model based on the required information to determine the target parameters in the algorithm model; and The perturbation case is obtained by scaling the parameter data related to the target parameter in the base case based on the scaling factor. The defect information describes the ratio between the result of the perturbation case and the result of the base case, and the difference between them and the scaling factor. The result of the base case is determined by executing the base case according to the business strategy script.

4. The method according to claim 3, characterized in that, The defect assessment using a second intelligent agent based on the perturbation example results includes: The second agent analyzes the results of the perturbation test and the results of the base test based on the scaling factor, and outputs the defect information, which represents at least one of the following defects in the business strategy script: The magnitude of the business data was set incorrectly; The data unit attribute of the business data is set incorrectly; An execution error occurred in the code related to the target parameters.

5. The method according to any one of claims 1 to 3, characterized in that, The perturbation of the example data related to the model terms of the algorithm model in the basic example includes: Based on the perturbation business data index, the business data index related to the model item in the basic case is updated to obtain the perturbation case. The perturbation business data index is used to map other business data that are different from the basic business data. The basic business data has a mapping relationship with the business data index. The defect information describes that there is an error in the code used to map business data in the business strategy script.

6. The method according to claim 1, characterized in that, The defect information is determined by compressing the disturbance case results, the basic case results, and the defect description information. The basic case results are determined by executing the basic case according to the business strategy script. The defect description information is determined by the second agent by analyzing the disturbance case results and the basic case results according to the requirement information.

7. The method according to claim 1, characterized in that, The demand information includes business requirements that represent optimizing energy utilization efficiency, and the target business strategy is used to execute an energy allocation strategy that allocates the amount of energy produced.

8. A business strategy determination device based on intelligent agents, characterized in that, The business strategy determination device includes: The first determining module is used to utilize multiple first intelligent agents to execute multiple tasks with dependencies based on demand information about business requirements, and to determine a business strategy script and a basic calculation example. The business strategy script represents an algorithm model that matches the business requirements, and the business requirements include at least one of business objectives and business execution conditions. The perturbation case acquisition module is used to perturb the case data related to the model items of the algorithm model in the basic case to obtain perturbation cases; The second determination module is used to perform defect assessment based on the results of the perturbation calculation using a second intelligent agent, and determine defect information related to the target task among the multiple tasks. The results of the perturbation calculation are determined by executing the perturbation calculation according to the business strategy script. The target business strategy script acquisition module is used to utilize the target intelligent agent corresponding to the defect information among multiple first intelligent agents to execute the target task according to the defect information and obtain the target business strategy script; The third determining module is used to determine the target business strategy based on the target business strategy script. Among them, multiple first intelligent agents run in mutually isolated task execution environments, and the first determining module includes: The business intent information output unit is used to utilize the intent understanding agent to perform intent understanding tasks based on the demand information and output structured business intent information; An algorithm model output unit is used to perform an algorithm model generation task on the business intent information using an algorithmic agent, and output the algorithm model. The business strategy script output unit is used to utilize a script agent to execute code editing tasks based on the business intent information and the algorithm model, and output the business strategy script. The algorithm model includes an objective function and constraints to satisfy the business requirements. The perturbed computational instance is determined by perturbing the computational data corresponding to at least one of the model terms in the objective function and the constraints. The second determining module includes: The first obtaining unit is used to process the disturbance case results and the basic case results according to the business optimization objective represented by the objective function using the second intelligent agent to obtain the defect information. The basic case results are determined by executing the basic case according to the business strategy script. The defect information describes the degree of difference between the objective direction of the objective function and the business requirement, or describes the degree of difference between the constraint direction of the constraint condition and the business requirement.

9. An intelligent agent of artificial intelligence, characterized in that, The intelligent agent includes: The input module is used to receive input information; The processing module is configured to determine a target task based on the input information received by the input module, determine a large model based on the target task, and execute the method of any one of claims 1 to 7 by calling the large model to obtain output information; An output module is used to output the output information obtained by the processing module.

10. An electronic device, characterized in that, The electronic device includes: At least one processor; and A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the method of any one of claims 1 to 7.

11. A non-transitory computer-readable storage medium storing computer instructions, characterized in that, The computer instructions are used to cause the computer to perform the method according to any one of claims 1 to 7.

12. A computer program product, characterized in that, The computer program product includes a computer program that, when executed by a processor, implements the method according to any one of claims 1 to 7.

Images