Intelligent agent automatic reasoning method and system based on task dependence planning

Abstract

The invention provides an agent automatic reasoning method and system based on task dependence planning, and belongs to the technical field of model automatic reasoning and systems. The method comprises the steps that according to state information of a current task, a forward reasoning module is adopted to predict a subtask needing to be executed, and a dependency relationship between the subtask and the state is generated; according to the dependency relationship between the subtasks and the state, generating a confidence coefficient score of the dependency relationship between each subtask and the state by adopting reverse reasoning; performing parameter uncertainty statistical analysis on the predicted and output sub-tasks, generating probability distribution of results, and quantifying the uncertainty of the sub-tasks; decision uncertainty analysis is carried out on historical behaviors and current states of an executor, and the decision credibility of the executor is evaluated. The problem that a predefined task dependency relationship cannot adapt to different task inputs and topological structures is solved, and the problem that a large language model is too confident when expressing verbal uncertainty is avoided.

Description

Technical Field

[0001] The present invention belongs to the technical field of model automatic reasoning, and particularly relates to an intelligent agent automatic reasoning method and system based on task-dependent planning. Background Art

[0002] The statements in this part only provide background technical information related to the present invention and do not necessarily constitute prior art.

[0003] Task-dependent planning aims to plan subtasks to complete a goal, decompose the goal task into individual tasks, and model the dependencies between tasks. Existing planning methods mainly use predefined dependency structures to model task dependencies and evaluate the reliability of intermediate results through the verbalized uncertainty generated by the model. However, the predefined dependency structures are not common in practice, and the uncertainty of verbal expressions faces the challenge of overconfidence.

[0004] In addition, task-dependent planning supported by large language models still faces two challenges: First, the predefined task dependencies are not common in practice. Most methods use reasoning ideas to model task dependencies as predefined structures, such as chains, trees, and graphs. Although recent research has explored dynamic planning of tasks to improve interpretability, this cannot dynamically simulate the appropriate dependencies between the tasks to be executed and the states, which record the user input and the history of task execution. For example, ReAct uses a large language model to reason about the chain of thought, obtain observations, and update the task plan simultaneously. The dependency topology of the task is linear. However, under different task inputs, the target task can be decomposed into subtasks with different numbers and topological dependency structures. Some tasks only need to select appropriate logical rules to prove the goal, while other tasks need to be decomposed into subtasks with linear or graphical dependencies respectively. Therefore, the predefined task dependencies cannot solve all problems in specific task domains.

[0005] Second, most autonomous planning methods evaluate the reliability of intermediate results through the reliability generated by the model. However, when expressing verbal uncertainty, large language models tend to be overconfident. Summary of the Invention

[0006] To overcome the deficiencies of the above-mentioned prior art, the present invention provides an intelligent agent automatic reasoning method and system based on task-dependent planning, which is used to adaptively model the dependencies between planned tasks and states and estimate the uncertainty of intermediate results during the planning process, solve the limitations of existing large language models in task-dependent planning and uncertainty evaluation, and thus improve the accuracy, flexibility, and reliability of decision-making.

[0007] To achieve the above object, one or more embodiments of the present invention provide the following technical solutions: The first aspect of the present invention provides an intelligent agent automatic reasoning method based on task dependency planning; The intelligent agent automatic reasoning method based on task dependency planning includes: Step S1, according to the status information of the current task, use forward reasoning to predict the subtasks to be executed, and generate the dependency relationship between the subtasks and the status information; Step S2, according to the dependency relationship between the subtasks and the status, use backward reasoning to generate the confidence score of the dependency relationship between each subtask and the status information, and eliminate the subtasks that do not meet the task dependency relationship according to the set belief score threshold; Step S3, use the executor to execute the remaining subtasks and output the intermediate results of the subtasks; Step S4, through the uncertainty analysis of the intermediate results of the subtasks, obtain the uncertainty score of the intermediate results of the subtasks; wherein, the uncertainty analysis includes parameter uncertainty analysis and decision uncertainty analysis; The parameter uncertainty statistical analysis quantifies the uncertainty of the subtasks by generating the probability distribution of the intermediate results of the subtasks; The decision uncertainty analysis evaluates the credibility of the decision by performing uncertainty analysis on the intermediate results of the subtasks and the tasks to be executed in the subsequent reasoning process; Step S5, iterate steps S1 to S4, stop iterating when the maximum number of iterations is reached or the target task is completed, and summarize and output the final reasoning result according to all the intermediate results generated during the iteration process.

[0008] As a further technical solution, the status information of the current task includes user input information, historical task execution records, task definitions and their results.

[0009] As a further technical solution, the process of using the forward reasoning module to predict the subtasks to be executed and generate the dependency relationship between the subtasks and the status is as follows: Analyze the current status information, including user input information, historical task execution records and task definitions; Based on the results of the status analysis, decompose the target task into subtasks with different dependency relationships; Perform priority sorting according to the importance and urgency of the subtasks to determine the order of task execution.

[0010] As a further technical solution, the process of using backward reasoning to generate the confidence score of the dependency relationship between each subtask and the status information is as follows: According to the task definitions of each state and subtask, determine whether there is a dependency relationship between the subtask and each state information and the probability of the existence of the dependency relationship; Calculate the reverse inference probability, construct a confidence score calculation formula, and obtain the confidence score of the dependency relationship between each subtask and the state information.

[0011] As a further technical solution, the confidence score calculation formula is:

[0012] In the formula, is the confidence score; represents a linear transformation; is the state; is the subtask; is the probability of the forward inference result, which is used to represent the confidence of the subtask depending on all historical tasks; is the probability of the reverse inference result, which is used to represent the confidence of the subtask depending on only one historical task.

[0013] As a further technical solution, the parameter uncertainty analysis and the decision uncertainty analysis are respectively evaluated by information entropy; wherein, the uncertainty of the subtask intermediate result is evaluated by calculating the information entropy of the subtask intermediate result; By calculating the information entropy of the inference result obtained in the previous inference process and the task planned by the current planner, the credibility of the decision is evaluated.

[0014] As a further technical solution, perform uncertainty analysis on the intermediate result of the subtask to obtain the uncertainty score of the subtask intermediate result, including: By calculating the arithmetic mean of the parameter uncertainty and the decision uncertainty analysis results, the uncertainty score of the subtask intermediate result is obtained, as shown in the following formula:

[0015] In the formula, is the uncertainty score of the subtask intermediate result; represents the result of the parameter uncertainty analysis; represents the result of the decision uncertainty analysis.

[0016] The second aspect of the present invention provides an intelligent agent automatic reasoning system based on task dependency planning.

[0017] The intelligent agent automatic reasoning system based on task dependency planning includes: A forward inference module, configured to: according to the state information of the current task, use forward inference to predict the subtasks to be executed, and generate the dependency relationship between the subtasks and the state; The backward reasoning module is configured to: according to the dependency relationship between the subtasks and the states, perform backward reasoning to generate a confidence score for the dependency relationship between each subtask and the state information, and eliminate the subtasks that do not satisfy the task dependency relationship according to the set belief score threshold; The intermediate result obtaining module is configured to: use the executor to execute the remaining subtasks and output the intermediate results of the subtasks; The uncertainty analysis module is configured to: perform uncertainty analysis on the intermediate results of the subtasks to obtain the uncertainty score of the intermediate results of the subtasks; wherein, the uncertainty analysis includes parameter uncertainty analysis and decision uncertainty analysis; The parameter uncertainty statistical analysis quantifies the uncertainty of the subtasks by generating the probability distribution of the intermediate results of the subtasks; The decision uncertainty analysis evaluates the credibility of the decision by performing uncertainty analysis on the intermediate results of the subtasks and the tasks to be executed in the subsequent reasoning process; The inference result generation module is configured to: iterate the above process, stop iterating when the maximum number of iterations is reached or the target task is completed, and summarize and output the final inference result according to all the intermediate results generated during the iteration process.

[0018] The third aspect of the present invention provides a computer-readable storage medium, on which a program is stored, and when the program is executed by a processor, it implements the steps in the agent automatic reasoning method based on task dependency planning as described in the first aspect of the present invention.

[0019] The fourth aspect of the present invention provides an electronic device, including a memory, a processor, and a program stored on the memory and executable on the processor, and when the processor executes the program, it implements the steps in the agent automatic reasoning method based on task dependency planning as described in the first aspect of the present invention.

[0020] The above one or more technical solutions have the following beneficial effects: (1) The present invention dynamically models the dependency relationship between tasks and states through a bidirectional reasoning module (forward reasoning and backward reasoning), and solves the problem that the predefined task dependency relationship cannot adapt to different task inputs and topological structures. Forward reasoning predicts subtasks based on the current state, and backward reasoning evaluates the confidence of the dependency relationship between subtasks and states, so as to be able to flexibly meet the requirements of different task domains and improve the adaptability and accuracy of task planning.

[0021] (2) By introducing an uncertainty estimation module, the present invention can evaluate the reliability of the intermediate results generated during the planning process. By considering parameter uncertainty and decision uncertainty, the system can more accurately judge the credibility of the generated results, avoiding the problem that large language models are too confident when expressing verbal uncertainty. This helps to improve the reliability and safety of decision-making and reduce incorrect decisions caused by overconfidence of the model.

[0022] (3) The present invention does not require model training or rely on demonstration examples, making it more practical and convenient in practical applications. This method reduces the complexity of deployment and maintenance, can be quickly applied to various fields, reduces the dependence on external data or complex training processes, and improves the scalability and usability of the system.

[0023] The advantages of additional aspects of the present invention will be partly given in the following description, partly will become apparent from the following description, or will be learned through the practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] The specification drawings forming a part of the present invention are used to provide a further understanding of the present invention. The schematic embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation to the present invention.

[0025] Figure 1 It is a flowchart of the method for the first embodiment.

[0026] Figure 2 It is a system structure diagram for the second embodiment. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0027] It should be noted that the following detailed description is exemplary and is intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by those of ordinary skill in the technical field to which the present invention belongs.

[0028] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present invention.

[0029] Without conflict, the embodiments in the present invention and the features in the embodiments can be combined with each other.

[0030] The two-way reasoning module in the present invention explicitly models the dependency relationship between tasks and states through two processes of forward reasoning and backward reasoning; an uncertainty estimation module is proposed to quantitatively evaluate the reliability of intermediate results.

[0031] Embodiment 1 This embodiment discloses an intelligent agent automatic reasoning method based on task - dependency planning. Task - dependency relationship modeling is a core component of the intelligent agent automatic reasoning framework, aiming to dynamically capture the complex dependency relationships between tasks and states, thereby supporting flexible and efficient task planning. Traditional task - planning methods usually rely on predefined task - dependency structures, such as chain - like, tree - like, or graph - like structures. However, these methods have obvious limitations in practical applications because the dependency relationships of tasks often change dynamically with the variation of input conditions. To solve this problem, a bidirectional reasoning module is proposed in this embodiment. This module explicitly models the dependency relationships between tasks and states through two processes: forward reasoning and backward reasoning.

[0032] As Figure 1 shown, the intelligent agent automatic reasoning method based on task - dependency planning includes: Step S1, according to the status information of the current task, use forward reasoning to predict the subtasks to be executed, and generate the dependency relationships between the subtasks and the states; In step S1, the status information of the current task includes user input information, historical task execution records, task definitions, and the execution results of historical tasks. Directly predicting the subtasks to be executed based on the above - mentioned status information can dynamically adapt to different task inputs and context environments. For example, when facing complex tasks, forward reasoning can flexibly generate subtask dependency relationships with different topological structures according to the specific requirements of the tasks, rather than being limited to fixed chain - like or tree - like structures.

[0033] Furthermore, the forward - reasoning process is as follows: Step S11, state analysis. The system first analyzes the current status information, including user input, historical task execution records, and task definitions. Through state analysis, comprehensively understand the context environment of the task.

[0034] Step S12, task decomposition. Based on the results of state analysis, the system decomposes the target task into multiple subtasks. These subtasks may have different dependency relationships, such as linear dependency, parallel dependency, or graphical dependency.

[0035] Step S13, priority ranking. The system ranks the subtasks according to their importance and urgency to determine the order of task execution.

[0036] Through forward reasoning, the system can quickly generate subtask dependency relationships that adapt to different task inputs, thereby improving the flexibility and adaptability of task planning.

[0037] Step S2: According to the dependency relationship between the subtasks and the states, use backward reasoning to generate the confidence scores of the dependency relationship between each subtask and the state information, and eliminate the subtasks that do not meet the task dependency relationship according to the set belief score threshold. Backward reasoning is used to evaluate the confidence scores of the dependency relationship between each subtask and the state. Through backward reasoning, the system can quantify the importance of each subtask in the current state and its relevance to other tasks. This process not only enhances the transparency of task planning but also provides reliable support for subsequent decision-making. By combining forward reasoning and backward reasoning, it can dynamically adapt to the requirements of different task domains and avoid the limitations caused by relying on predefined structures in traditional methods. This adaptive task dependency modeling method significantly improves the flexibility and scalability of task planning.

[0038] During the forward reasoning process, the planner used for planning the task execution order predicts the subtask , with a probability of , which represents the belief of the subtask in and depends on all states S.

[0039] To explicitly simulate the dependency relationship between the state and the subtask, the confidence score of the task dependency between the subtask and the state is quantitatively calculated through backward reasoning.

[0040] During the backward reasoning process, the planner first determines whether there is a dependency relationship between the subtask and and each state in S according to the task definitions of . The probability that there is a dependency relationship between the subtask and the state is . Subsequently, calculate , which represents the probability of backward reasoning, that is, depends on the states in S, but is excluded, which represents the belief that the target task depends on . The difference reveals that the belief of only depends on the probability value of .

[0041] Therefore, the confidence score of the task dependency between the subtask and the state is defined as:

[0042] In the formula, is the confidence score; represents a linear transformation to ensure that the range of the confidence score is [0, 1], is the state; is the subtask; is the belief of the subtask; is the backward inference probability; by setting a belief score threshold , to determine and whether there is a dependency relationship between them. When , and there is a task dependency relationship between them, which indicates that depends on .

[0043] Step S3, use the executor to execute the remaining subtasks and output the intermediate results of the subtasks.

[0044] The executor executes the subtasks through , where Executor is an executor based on a large model, inputting the task description and the current inference state , and outputting the intermediate result of the subtask .

[0045] Step S4, by performing uncertainty analysis on the intermediate results of the subtasks, obtain the uncertainty score of the intermediate results of the subtasks.

[0046] During the task planning process, the reliability of the intermediate results directly affects the quality of the final decision. However, existing large language models tend to be overconfident when expressing uncertainty, resulting in potential risks that may be overlooked during the decision-making process. To address this issue, an uncertainty estimation module is also proposed in this embodiment to quantitatively evaluate the reliability of the intermediate results. Among them, the uncertainty analysis process includes parameter uncertainty analysis and decision uncertainty analysis.

[0047] Parameter uncertainty mainly focuses on the credibility of the model-generated results. By performing statistical analysis on the model output, this module can estimate the probability distribution of the generated results, thereby quantifying its uncertainty. This process evaluates the uncertainty by calculating the information entropy of the intermediate results of the subtasks. The higher the information entropy, the greater the uncertainty of the subtask results. The information entropy is as follows:

[0048] In the formula, is the parameter uncertainty, used to evaluate the uncertainty of the intermediate results of the subtasks; is the information entropy; is the intermediate result of the subtask.

[0049] For example, when the model generates multiple possible subtasks, the parameter uncertainty module can evaluate the probability of each subtask and select the most reliable option as the basis for the next step of planning. Through parameter uncertainty estimation, the system can comprehensively evaluate the credibility of the intermediate results, thereby reducing the risk of incorrect decisions caused by the model's overconfidence.

[0050] Decision uncertainty focuses on the reliability of the speech or instructions generated by the executor during the planning process. This module evaluates the credibility of its decisions by analyzing the executor's historical behavior and current state. Specifically, uncertainty analysis is performed on the intermediate results of the subtasks and the tasks to be executed in the subsequent reasoning process, and its information entropy is calculated to evaluate the credibility of the decision. The information entropy is as follows:

[0051] In the formula, is the decision uncertainty; D represents the result of the decision, which is used to determine whether the result generated by the decision is reliable.

[0052] The higher the decision uncertainty, the less reliable the executor's decision in generating the subtask result. For example, when the executor proposes a high-risk suggestion during the planning process, the decision uncertainty module can identify its potential uncertainty and remind the system to take additional verification steps. By combining parameter uncertainty and decision uncertainty, the reliability of the intermediate results can be comprehensively evaluated, thereby reducing the risk of incorrect decisions caused by the model's overconfidence. This function is particularly important in complex task planning because it can help the system make more cautious and reliable decisions at critical nodes.

[0053] During the specific execution process, when the executor executes the subtask , it is based on the dependent state . Then, the executor generates a reliability label D for . The uncertainty score of the intermediate result is calculated as the arithmetic mean of two parts, which includes two aspects of the uncertainty score: parameter uncertainty and decision uncertainty. The specific calculation method is as follows:

[0054] Where represents the parameter uncertainty, which is obtained by calculating the information entropy based on the output probability of the executor. represents the decision uncertainty, which is obtained by calculating the information entropy based on the probability output by the executor along with the production of D.

[0055] Example 2 This embodiment discloses an intelligent agent automatic reasoning system based on task - dependent planning; As Figure 2 shown, the intelligent agent automatic reasoning system based on task - dependent planning includes: The intelligent agent automatic reasoning system based on task - dependent planning includes: A forward reasoning module, configured to: according to the status information of the current task, use forward reasoning to predict the subtasks to be executed, and generate the dependency relationship between the subtasks and the status; A backward reasoning module, configured to: according to the dependency relationship between the subtasks and the status, use backward reasoning to generate the confidence score of the dependency relationship between each subtask and the status information, and eliminate the subtasks that do not meet the task dependency relationship according to the set belief score threshold; An intermediate result acquisition module, configured to: use an executor to execute the remaining subtasks and output the intermediate results of the subtasks; An uncertainty analysis module, configured to: perform uncertainty analysis on the intermediate results of the subtasks to obtain the uncertainty score of the subtask intermediate results; wherein, the uncertainty analysis includes parameter uncertainty analysis and decision - making uncertainty analysis; The parameter uncertainty statistical analysis quantifies the uncertainty of the subtasks by generating the probability distribution of the subtask intermediate results; The decision - making uncertainty analysis evaluates the credibility of the decision by performing uncertainty analysis on the intermediate results of the subtasks and the tasks to be executed in the subsequent reasoning process; An inference result generation module, configured to: iterate the above process, stop the iteration when the maximum number of iterations is reached or the target task is completed, and summarize and output the final inference result according to all the intermediate results generated during the iteration.

[0056] Example 3 The purpose of this embodiment is to provide a computer - readable storage medium.

[0057] A computer - readable storage medium, on which a computer program is stored, and when the program is executed by a processor, it implements the steps in the intelligent agent automatic reasoning method based on task - dependent planning as described in Example 1.

[0058] Example 4 The purpose of this embodiment is to provide an electronic device.

[0059] An electronic device, including a memory, a processor, and a program stored on the memory and executable on the processor, and when the processor executes the program, it implements the steps in the intelligent agent automatic reasoning method based on task - dependent planning as described in Example 1.

[0060] In the devices of the above Second, Third, and Fourth Embodiments, the steps involved correspond to those of the First Method Embodiment. For the specific implementation manners, reference may be made to the relevant description part of the First Embodiment. The term "computer-readable storage medium" should be understood to include a single medium or multiple media containing one or more instruction sets; it should also be understood to include any medium that can store, encode, or carry an instruction set for execution by a processor and enable the processor to execute any method in the present invention.

[0061] Those skilled in the art should understand that the above-mentioned modules or steps of the present invention can be implemented by a general-purpose computer device. Optionally, they can be implemented by program codes executable by a computing device, so that they can be stored in a storage device for execution by the computing device, or they can be separately fabricated into individual integrated circuit modules, or multiple modules or steps among them can be fabricated into a single integrated circuit module for implementation. The present invention is not limited to any specific combination of hardware and software.

[0062] Although the specific implementation manners of the present invention have been described above in conjunction with the accompanying drawings, this is not a limitation on the protection scope of the present invention. Those skilled in the art should understand that, based on the technical solutions of the present invention, various modifications or variations that can be made without creative efforts by those skilled in the art are still within the protection scope of the present invention.

Claims

1. An agent automatic reasoning method based on task-dependent planning, characterized in that: include: Step S1, based on the status information of the current task, forward reasoning is used to predict the subtasks to be executed, and the dependency relationship between the subtasks and the status information is generated; Step S2, based on the dependency relationship between the subtasks and the state, reverse reasoning is used to generate a confidence score for the dependency relationship between each subtask and the state information, and subtasks that do not meet the task dependency relationship are eliminated according to a set belief score threshold; Step S3, using the executor to execute the remaining subtasks and output the intermediate results of the subtasks; Step S4, performing uncertainty analysis on the intermediate results of the subtask to obtain an uncertainty score of the intermediate results of the subtask; wherein the uncertainty analysis includes parameter uncertainty analysis and decision uncertainty analysis; The parameter uncertainty statistical analysis quantifies the uncertainty of the subtask by generating a probability distribution of the intermediate results of the subtask; The decision uncertainty analysis evaluates the credibility of the decision by performing uncertainty analysis on the intermediate results of the subtasks and the tasks that need to be performed in the subsequent reasoning process; Step S5, iterate step S1 to step S4, stop iteration when the maximum number of iterations is reached or the target task is completed, and summarize and output the final inference result based on all intermediate results generated during the iteration process.

2. The agent automatic reasoning method based on task-dependent planning as claimed in claim 1, characterized in that: The status information of the current task includes user input information, historical task execution records, task definition and its results.

3. The agent automatic reasoning method based on task-dependent planning as claimed in claim 1, characterized in that: The process of using the forward reasoning module to predict the subtasks to be executed and generating the dependency relationship between the subtasks and the states is as follows: Analyze current status information, including user input information, historical task execution records, and task definitions; Based on the results of state analysis, the target task is decomposed into subtasks with different dependencies; Prioritize subtasks according to their importance and urgency and determine the order in which tasks should be performed.

4. The agent automatic reasoning method based on task-dependent planning as claimed in claim 1, characterized in that: The process of using reverse reasoning to generate the confidence score of the dependency relationship between each subtask and the state information is as follows: According to the task definition of each state and subtask, determine whether there is a dependency relationship between the subtask and each state information and the probability of the dependency relationship; Calculate the reverse reasoning probability, construct the confidence score calculation formula, and obtain the confidence score of the dependency relationship between each subtask and the state information.

5. The agent automatic reasoning method based on task-dependent planning as claimed in claim 4, characterized in that: The confidence score calculation formula is: In the formula, is the confidence score; represents a linear transformation; for status; For subtasks; is the probability of the forward reasoning result, which is used to indicate the confidence that the subtask depends on all historical tasks; is the probability of the reverse reasoning result, which is used to indicate the confidence that the subtask depends only on a certain historical task.

6. The agent automatic reasoning method based on task-dependent planning as claimed in claim 1, characterized in that: The parameter uncertainty analysis and decision uncertainty analysis are evaluated by information entropy respectively; wherein the uncertainty is evaluated by calculating the information entropy of the intermediate results of the subtask; The credibility of the decision is evaluated by calculating the information entropy of the reasoning results obtained in the previous reasoning process and the tasks planned by the current planner.

7. The agent automatic reasoning method based on task-dependent planning as claimed in claim 1, characterized in that: Perform uncertainty analysis on the intermediate results of the subtask to obtain uncertainty scores of the intermediate results of the subtask, including: By calculating the arithmetic mean of the parameter uncertainty and decision uncertainty analysis results, the uncertainty score of the intermediate result of the subtask is obtained, as shown in the following formula: In the formula, Score the uncertainty of the intermediate results of the subtask; represents the results of parameter uncertainty analysis; Represents the results of decision uncertainty analysis.

8. An agent-based automatic reasoning system based on task-dependent planning, characterized by: include: The forward reasoning module is configured to: predict the subtasks to be executed by using forward reasoning according to the state information of the current task, and generate the dependency relationship between the subtasks and the state; The reverse reasoning module is configured to: generate a confidence score of the dependency relationship between each subtask and the state information by reverse reasoning according to the dependency relationship between the subtask and the state, and eliminate the subtasks that do not meet the task dependency relationship according to a set belief score threshold; The intermediate result acquisition module is configured to: execute the remaining subtasks using the executor and output the intermediate results of the subtasks; The uncertainty analysis module is configured to: obtain an uncertainty score of the intermediate result of the subtask by performing uncertainty analysis on the intermediate result of the subtask; wherein the uncertainty analysis includes parameter uncertainty analysis and decision uncertainty analysis; The parameter uncertainty statistical analysis quantifies the uncertainty of the subtask by generating a probability distribution of the intermediate results of the subtask; The decision uncertainty analysis evaluates the credibility of the decision by performing uncertainty analysis on the intermediate results of the subtasks and the tasks that need to be performed in the subsequent reasoning process; The inference result generation module is configured to iterate the above process, stop iteration when the maximum number of iterations is reached or the target task is completed, and summarize and output the final inference result based on all intermediate results generated during the iteration process.

9. A computer-readable storage medium having a program stored thereon, characterized in that: When the program is executed by a processor, the steps in the intelligent agent automatic reasoning method based on task-dependent planning as described in any one of claims 1 to 7 are implemented.

10. An electronic device comprising a memory, a processor, and a program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the steps in the intelligent agent automatic reasoning method based on task-dependent planning as described in any one of claims 1-7 are implemented.

Images