Task processing method and device based on multi-agent cooperation, equipment and medium

Abstract

The invention relates to the technical field of artificial intelligence, can be applied to business system platforms of financial science and technology, medical treatment and health and the like, and discloses a task processing method, device and equipment based on multi-agent cooperation and a medium. And constructing a sub-task dependency graph, sequentially scheduling and executing the sub-tasks, and performing parameter completion to obtain the completed sub-tasks. And tracking the execution progress and the historical record of the completion subtask through a preset progress agent, and generating an executable operation decision by using a preset decision agent. If the task execution does not reach the expected effect, feeding back difference information and a correction suggestion, adjusting an operation decision and generating an updating operation; if the task achieves the expected effect, task completion information is sent to the progress agent, and the state is updated to be task completion. According to the method, the perception accuracy is improved, a task dependence tracking and feedback correction mechanism is provided, and a cross-application automatic task is successfully realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of artificial intelligence, and in particular to a task processing method and device based on multi-agent cooperation, equipment and a medium. BACKGROUND

[0002] Currently, multi-modal large language models (MLLMs) have made some progress in agent research in graphical user interface (GUI) environments, but mainly focus on relatively simple interaction scenarios such as mobile phones and web pages, and can perform basic operations such as opening applications, clicking buttons, and searching for information. However, in more complex personal computer (PC) environments, existing technologies still have significant shortcomings: computer interface interaction elements are dense and diverse, text layout is complex and highly unstructured, resulting in low accuracy in element recognition and semantic understanding for agents, making it difficult to accurately execute fine-grained instructions such as "click the center alignment button on the toolbar" or "select the last paragraph of text"; computer tasks often involve multiple steps and software collaboration, and there are clear dependencies between sub-tasks, but existing methods generally lack cross-task data transmission and dependency perception mechanisms, which can easily cause task chain breaks or inconsistent execution; existing systems lack a perfect progress tracking and dynamic error correction mechanism, so even advanced models such as GPT-4o have a low overall success rate of about 8% when faced with complex computer tasks, which is much lower than the completion rate of a single sub-task.

[0003] In the field of medical health, medical interfaces often contain a large number of tables, examination images, and multi-level menus, with complex and highly differentiated element layouts. Existing MLLMs lack accuracy in identifying medical terminology and accurately positioning interface elements, which can easily result in misselection or information omission. Secondly, cross-department, multi-step diagnosis and treatment processes rely on highly rigorous data linkage, but existing systems lack dynamic tracking and verification mechanisms for dependencies between sub-tasks, which can cause process breaks or error transmission, affecting the safety and reliability of diagnosis and treatment.

[0004] In the field of financial technology business, agents can be used for financial data extraction, risk assessment, and report generation, but their application is also subject to various limitations. Financial system interfaces often contain highly structured but significantly different tables and transaction controls across platforms, and existing technologies are not accurate in multi-software collaboration and fine-grained element recognition, making it difficult to ensure compliance for complex operations. In addition, financial business process chains are long and have very low fault tolerance, but existing agents lack a perfect execution progress tracking and dynamic feedback correction mechanism, making it difficult to detect and correct errors in data entry or operation in a timely manner, thereby increasing operational risk and making it difficult to meet the strict requirements of the financial sector for robustness and compliance.

[0005] Therefore, in the prior art, the multi-modal large language model-based agent generally has low perception accuracy, lacks a task-dependent tracking and feedback correction mechanism, and is difficult to perform multi-step cross-application automated tasks in a complex computer environment. SUMMARY

[0006] The application provides a multi-agent cooperation-based task processing method, device, equipment and medium, which mainly aims to solve the problem that the multi-modal large language model-based agent generally has low perception accuracy, lacks a task-dependent tracking and feedback correction mechanism, and is difficult to perform multi-step cross-application automated tasks in a complex computer environment.

[0007] In a first aspect, to achieve the above-mentioned purpose, the application provides a multi-agent cooperation-based task processing method, which comprises the following steps: obtaining a task instruction of a target user, performing high-level semantic analysis and decomposition on the task instruction by using a preset management agent, and constructing a subtask dependency graph according to a subtask queue obtained by decomposition; sequentially scheduling and executing the subtask queue according to the subtask dependency graph, and using a task execution result of an executed task to complete parameter completion of a next subtask in the subtask queue to obtain a completed subtask; tracking execution progress and historical records of the completed subtask by using a preset progress agent to obtain an execution state and an execution historical track; generating an executable operation decision according to the execution state and the execution historical track by using a preset decision agent; judging whether a task execution result of the operation decision achieves an expected effect by using a preset reflection agent; if the task execution result of the operation decision does not achieve the expected effect, feeding back result difference information and correction suggestions generated to the preset decision agent, adjusting the operation decision by using the preset decision agent after receiving the feedback, generating an operation update decision, executing the completed subtask by using the operation update decision, and obtaining a task completion result; if the task execution result of the operation decision achieves the expected effect, sending task completion information to the preset progress agent, and updating the execution state of the completed subtask to task completion by using the preset progress agent after receiving the information.

[0008] In a second aspect, the application further provides a multi-agent cooperation-based task processing device, which comprises: A dependency graph construction module is configured to obtain a task instruction of a target user, perform high-level semantic analysis and decomposition on the task instruction by using a preset management agent, and construct a subtask dependency graph according to a subtask queue obtained by decomposition; A subtask parameter completion module is configured to sequentially schedule and execute the subtask queue according to the subtask dependency graph, and use the task execution result of the executed subtask to complete the parameters of the next subtask in the subtask queue to obtain a completed subtask. An execution progress tracking module is configured to track the execution progress and history of the completed subtask by using a preset progress agent to obtain an execution state and an execution history trajectory. An operation decision generation module is configured to generate an executable operation decision according to the execution state and the execution history trajectory by using a preset decision agent. An execution effect judgment module is configured to judge whether the task execution result of the operation decision achieves an expected effect by using a preset reflection agent. An operation decision adjustment module is configured to, if the task execution result of the operation decision does not achieve the expected effect, feed back result difference information and correction suggestions to the preset decision agent, adjust the operation decision by using the preset decision agent receiving the feedback, generate an operation update decision, execute the completed subtask by using the operation update decision, and obtain a task completion result. An execution state updating module is configured to, if the task execution result of the operation decision achieves the expected effect, send task completion information to the preset progress agent, and update the execution state of the completed subtask to task completion by using the preset progress agent receiving the information.

[0009] In a third aspect, the present application further provides an electronic device, which comprises: at least one processor; and a memory connected with the at least one processor in communication; wherein the memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor to enable the at least one processor to execute the task processing method based on multi-agent cooperation described above.

[0010] In a fourth aspect, the present application further provides a computer readable storage medium, which stores at least one computer program, and the at least one computer program is executed by a processor in an electronic device to implement the task processing method based on multi-agent cooperation described above.

[0011] The application can effectively identify the sequence and dependency relationship between tasks by obtaining the task instructions of the target user, performing high-level semantic analysis and decomposition on the task instructions by a preset management agent, and constructing a subtask dependency graph according to the subtask queue obtained by decomposition, thereby realizing optimized scheduling of tasks and efficient use of resources, enhancing the automation, intelligence and accuracy of task execution, scheduling and executing the subtask queue in turn according to the subtask dependency graph, and using the executed task execution result to complete parameter completion of the next subtask in the subtask queue to obtain a completed subtask. A centralized data management platform is provided by the communication hub storage structure, so that the task execution result can be timely transmitted to the subsequent subtask, ensuring smooth data flow between parameters and tasks. The execution progress and historical record of the completed subtask are tracked by a preset progress agent to obtain an execution state and an execution history track. An executable operation decision is generated by a preset decision agent according to the execution state and the execution history track. The system can automatically adjust the execution path by screening the optimal operation action and converting it into an executable decision, ensuring that the task is promoted in the most effective way. The preset reflection agent is used to judge whether the task execution result of the operation decision reaches the expected effect. If the task execution result of the operation decision does not reach the expected effect, the generated result difference information and correction suggestion are fed back to the preset decision agent, and the operation decision is adjusted by the preset decision agent receiving the feedback to generate an operation update decision. The completed subtask is executed by the operation update decision to obtain a task completion result. The operation decision can be identified and adjusted in real time to ensure that the execution path and strategy of the task are always consistent with the target. This closed-loop feedback mechanism ensures the efficiency of task execution, reduces risks, and improves the adaptive ability of decision-making. If the task execution result of the operation decision reaches the expected effect, the task completion information is sent to the preset progress agent, and the execution state of the completed subtask is updated to task completion by the preset progress agent receiving the information. The task is executed by the operation decision to generate task completion information, which is transmitted to the preset progress agent to realize automatic updating of the task execution progress, improve the perception accuracy, and have a task dependency tracking and feedback correction mechanism to successfully realize cross-application automated tasks. BRIEF DESCRIPTION OF DRAWINGS

[0012] In order to more clearly illustrate the technical solutions of the embodiments of the application, the following will briefly introduce the drawings needed to be used in the description of the embodiments of the application. Obviously, the drawings in the following description are only some embodiments of the application, and other drawings can be obtained by those skilled in the art without creative labor.

[0013] Figure 1An application environment schematic diagram of a task processing method based on multi-agent cooperation according to an embodiment of the present application; Figure 2 A flow schematic diagram of a task processing method based on multi-agent cooperation according to an embodiment of the present application; Figure 3 A flow schematic diagram of a subtask execution progress tracking process of a task processing method based on multi-agent cooperation according to an embodiment of the present application; Figure 4 A module schematic diagram of a task processing device based on multi-agent cooperation according to an embodiment of the present application; Figure 5 A structure schematic diagram of an electronic device for implementing a task processing method based on multi-agent cooperation according to an embodiment of the present application; Figure 6 Another structure schematic diagram of an electronic device for implementing a task processing method based on multi-agent cooperation according to an embodiment of the present application.

[0014] The object, function characteristics and advantages of the present application will be further described with reference to the embodiments and the accompanying drawings. DETAILED DESCRIPTION

[0015] In order to make the person in the art better understand the technical scheme of the present disclosure, and understand the implementation process of how to apply technical means to solve the technical problems of the present disclosure and achieve the corresponding technical effects, and fully understand and implement the present disclosure, the technical scheme in the embodiments of the present disclosure will be described clearly and completely in conjunction with the drawings in the embodiments of the present disclosure. Obviously, the described embodiments are only a part of the embodiments of the present disclosure, not all. The embodiments of the present disclosure and each feature in the embodiments can be combined with each other without conflict, and the formed technical scheme is within the protection scope of the present disclosure. Based on the embodiments in the present disclosure, all other embodiments obtained by those skilled in the art without creative labor should be within the protection scope of the present disclosure.

[0016] It should be noted that the terms "first", "second", and the like in the description and claims of the present disclosure and above drawings are used to distinguish between similar objects and not necessarily describe a particular chronological or sequential order. It should be understood that the use of such terms is arbitrary and the embodiments of the present disclosure described herein can be practiced in orders other than those illustrated or described herein. Furthermore, the terms "comprise" and "have", and any variations thereof, are intended to cover a non-exclusive inclusion, for example, a process, method, device, product, or apparatus that comprises a list of steps or units is not necessarily limited to those steps or units that are clearly listed, but can include other steps or units that are not clearly listed or inherent to such processes, methods, products, or apparatuses.

[0017] The embodiment of the present application provides a task processing method based on multi-agent cooperation. The execution subject of the task processing method based on multi-agent cooperation includes but is not limited to at least one of electronic devices that can be configured to execute the device provided by the embodiment of the present application, such as a server and a terminal. In other words, the task processing method based on multi-agent cooperation can be executed by software or hardware installed in a terminal device or a server device. The server includes but is not limited to a single server, a server cluster, a cloud server, or a cloud server cluster. The server can be a stand-alone server, or a cloud server that provides cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content distribution networks (CDN), and big data and artificial intelligence platforms, and other basic cloud computing services.

[0018] The embodiment of the present application provides a task processing method based on multi-agent cooperation. The execution subject of the task processing method based on multi-agent cooperation includes but is not limited to at least one of electronic devices that can be configured to execute the device provided by the embodiment of the present application, such as a server and a terminal. In other words, the task processing method based on multi-agent cooperation can be executed by software or hardware installed in a terminal device or a server device. The server includes but is not limited to a single server, a server cluster, a cloud server, or a cloud server cluster. The server can be a stand-alone server, or a cloud server that provides cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content distribution networks (CDN), and big data and artificial intelligence platforms, and other basic cloud computing services. Figure 1In the application environment, the client communicates with the server through the network. The server can obtain the task instruction of the target user through the client, perform high-level semantic analysis and decomposition on the task instruction by using the preset management agent, construct a subtask dependency graph according to the subtask queue obtained by decomposition, effectively identify the sequence and dependency relationship between tasks, and thus realize the optimized scheduling of the task and the efficient use of resources, enhance the automation, intelligence and accuracy of the task execution, sequentially schedule and execute the subtask queue according to the subtask dependency graph, and use the executed task execution result to complete the parameter completion of the next subtask in the subtask queue to obtain a completed subtask. The communication hub storage structure provides a centralized data management platform, so that the task execution result can be timely transmitted to the subsequent subtask, ensuring smooth data flow between parameters and tasks. The preset progress agent tracks the execution progress and historical record of the completed subtask to obtain the execution state and execution history track. The preset decision agent generates executable operation decisions according to the execution state and the execution history track. By screening the optimal operation action and converting it into an executable decision, the system can automatically adjust the execution path to ensure that the task is pushed forward in the most effective way. The preset reflection agent judges whether the task execution result of the operation decision reaches the expected effect. If the task execution result of the operation decision does not reach the expected effect, the generated result difference information and correction suggestion are fed back to the preset decision agent, and the preset decision agent after receiving the feedback is used to adjust the operation decision to generate an operation update decision. The completed subtask is executed by using the operation update decision to obtain a task completion result. The operation decision can be identified and adjusted in real time to ensure that the execution path and strategy of the task are always consistent with the target. This closed-loop feedback mechanism ensures the efficiency of task execution, reduces risks, and improves the adaptive ability of decision-making. If the task execution result of the operation decision reaches the expected effect, the task completion information is sent to the preset progress agent, and the preset progress agent after receiving the information updates the execution state of the completed subtask to task completion. The task is executed by the operation decision to generate task completion information, which is transmitted to the preset progress agent to realize the automatic update of the task execution progress, improve the perception accuracy, and have the task dependency tracking and feedback correction mechanism to successfully realize the cross-application automated task. Finally, the task completion state is output and fed back to the user client. The client can be, but is not limited to, various personal computers, notebook computers, smart phones, tablet computers and portable wearable devices. The server can be implemented by an independent server or a server cluster composed of multiple servers. The application will be described in detail through specific embodiments.

[0019] The following is explained in the description of the present application, the present application realizes high-level semantic analysis and dependency modeling of task instructions through management agents, ensures reasonable task decomposition and orderly execution; the progress agent provides real-time tracking and state recording, ensures transparent and controllable execution process; the decision agent generates optimal operation actions based on the state and history, realizes intelligent scheduling and execution; the reflection agent is responsible for result verification and difference analysis, and timely provides correction suggestions, forms a self-correction and continuous optimization mechanism. This multi-agent collaborative architecture can not only improve the accuracy and robustness of task execution, but also dynamically adjust the strategy in complex environment to ensure efficient and stable target achievement.

[0020] Referring to Figure 2 Fig. 1 is a flowchart of a task processing method based on multi-agent cooperation provided by an embodiment of the present application. In the embodiment, the task processing method based on multi-agent cooperation comprises the following steps. S1, obtaining the task instruction of a target user, using a preset management agent to perform high-level semantic analysis and decomposition on the task instruction, and constructing a subtask dependency graph according to the subtask queue obtained by decomposition.

[0021] In the embodiment of the present application, by extracting the language features and key elements of the task instruction, performing semantic labeling and generating labeled instructions, using a preset management agent to perform context association, context semantic information is obtained. Combined with the task key elements and context information, the logical structure of the task instruction is generated, the labeled instructions are decomposed, the subtask queue is generated, and the dependency relationship between each subtask is identified, and a subtask dependency graph is constructed.

[0022] In the specific scene of medical health, it can be applied to an intelligent diagnosis system. Through high-level semantic analysis of task instructions, the symptoms and needs of patients are quickly identified, and diagnosis tasks and related subtasks are automatically generated. The system can perform subtask scheduling and execution according to medical record data and medical knowledge, track the progress of diagnosis and treatment in real time, ensure efficient connection and feedback correction of each link, thereby improving the diagnosis accuracy and reducing misdiagnosis and missed diagnosis.

[0023] In the specific scene of financial technology, it can be applied to an intelligent risk control system. By analyzing the transaction instructions and financial needs of customers, the system automatically decomposes them into multiple risk assessment and monitoring tasks. The system can monitor transaction behavior in real time, identify potential risk points, and make feedback corrections based on historical data to optimize the decision-making process. Through fine management of task scheduling and execution, the accuracy of risk prediction can be effectively improved, and the safety of financial transactions can be ensured.

[0024] In the embodiment of the present application, the use of a preset management agent to perform high-level semantic analysis and decomposition on the task instruction, and to construct a subtask dependency graph according to the subtask queue obtained by decomposition, comprises: extracting instruction language features of the task instruction, and extracting task key elements from the instruction language features; performing semantic labeling on the task instruction using the task key elements to obtain labeled instructions; performing context association on the labeled instructions using a preset management agent to obtain context semantic information; generating a logical structure of the task instruction according to the task key elements and the context semantic information; converting the logical structure into a high-level semantic representation; performing decomposition on the labeled instructions based on the high-level semantic representation to generate a sub-task queue; identifying a dependency relationship between the sub-task queue, and mapping the sub-task queue into a sub-task dependency graph according to the dependency relationship.

[0025] In detail, by analyzing the language structure of the task instruction, key syntax features such as verbs, nouns, time, and conditions are extracted, and the core elements of the task are identified, usually including task objectives, execution subjects, time constraints, and execution conditions. Through natural language processing technology, instruction language features are converted into operational task elements, providing clear basic data support for subsequent task decomposition and execution.

[0026] By assigning clear semantic labels to each key element, such as "task objective", "execution subject", "time requirement", etc., the ambiguous information in the instruction is clarified. For example, if the task instruction contains "analyze medical records and generate a report", "analyze medical records" is labeled as "task objective", and "generate a report" is labeled as "result output", making the task instruction more structured, providing a clear semantic basis for subsequent task analysis and execution.

[0027] The preset management agent associates and analyzes the context according to the labeled information of the task instruction. The agent identifies the front-back relationship, condition dependency, and other potential connections in the task instruction, and interfaces the labeled instruction with related background information, historical records, or other context data, which helps to understand the logical relationships implied in the task instruction, such as time sequence, priority, and dependency, thereby generating accurate context semantic information.

[0028] By identifying the logical relationships between tasks, such as sequence, condition dependency, and interaction between tasks, the execution process and the interrelation of each step of the task are depicted, and the logical structure is converted into a high-level semantic representation. This representation not only abstracts the core objectives and execution paths of the task, but also captures the global dependency relationships and constraints of the task.

[0029] Based on the high-level semantic representation, the preset management agent further decomposes the task target and logical structure in the annotated instruction, identifies specific subtasks, and organizes them into a subtask queue. The tasks in the subtask queue are independently defined according to the dependency relationship, execution order and condition requirements, and the smooth connection between the tasks in the subtask queue is ensured. The management agent converts the high-level target in the instruction into an operable small unit according to the complexity and execution requirement of the task, thereby providing clear steps and priority ranking for the specific execution of the task.

[0030] The preset management agent analyzes the execution dependency between each subtask queue, identifies which subtasks need to be completed before other tasks, and which tasks can be executed in parallel. By evaluating the order relationship and condition constraints between tasks, the management agent constructs a subtask dependency graph, maps these subtasks according to the dependency relationship, and displays the order, dependency level and execution priority between each subtask, providing visual structural support for task scheduling and execution, and ensuring that the task proceeds smoothly as expected.

[0031] Through accurate task instruction analysis and decomposition, efficient and automated execution of tasks can be achieved. By extracting and annotating the key elements of the instruction, the task target and requirements are clearly defined, avoiding human understanding bias. Combined with the generation of context information and logical structure, the accuracy and rationality of task execution are ensured. By constructing a subtask dependency graph, the order and dependency relationship between tasks can be effectively identified, thereby realizing optimized scheduling and efficient resource utilization of tasks, enhancing the automation, intelligence and accuracy of task execution, and helping to improve work efficiency, reduce errors, and ensure that the task is completed as expected.

[0032] S2, according to the subtask dependency graph, the subtask queue is sequentially scheduled and executed, and the execution result of the executed task is used to complete the parameters of the next subtask in the subtask queue to obtain a completed subtask.

[0033] In the embodiment of the application, the execution order and dependency relationship of the task are identified by the subtask dependency graph, the scheduling execution sequence is generated, and the communication hub storage structure is established to manage the data flow. During execution, the target subtask without previous dependency is selected according to the scheduling sequence, the input parameters are obtained and written into the communication hub storage structure. The target subtask is executed and the execution result is obtained, the result is stored in the communication hub and passed to the subsequent dependent task. The subsequent subtask is filled with the completed parameters to realize continuous execution and automatic completion of the task.

[0034] In the medical health specific scenario, it can be applied to help the medical information system to automatically process the diagnosis and treatment tasks of patients. By analyzing the medical records and instructions of patients, the system can identify key medical information and generate a sub-task queue according to the disease and treatment plan, such as laboratory examination, diagnosis analysis and report generation, etc. The dependency relationship between tasks is clearly marked to ensure that each step is executed at the right time, avoiding omission or repetition. During the task execution process, the system fills in the missing parameters in real time, such as automatically adjusting the subsequent diagnosis and treatment tasks according to the newly obtained detection results of the patient, thereby improving the diagnosis and treatment efficiency and accuracy.

[0035] In the financial technology specific scenario, it can be applied to intelligent risk management and investment decision system. By high-level semantic analysis of customer transaction instructions and market data, the system can identify key elements of each transaction task, such as market fluctuations, transaction risks, etc., and generate a series of sub-tasks, including risk assessment, fund scheduling and investment suggestions, etc. The system automatically schedules these tasks, adjusts the execution order according to the dependency relationship between tasks, and fills in the missing data in real time, such as supplementing the latest market data for risk assessment, thereby optimizing investment decisions and improving the safety and efficiency of financial business.

[0036] In the embodiment of the application, the sub-task queue is sequentially scheduled and executed according to the sub-task dependency graph, and the execution result of the executed task is used to complete the parameters of the next sub-task in the sub-task queue to obtain a completed sub-task, comprising: identifying the sub-task execution order and the sub-task dependency relationship according to the sub-task dependency graph; generating a scheduling execution sequence according to the sub-task execution order and the sub-task dependency relationship; screening the target sub-task without previous dependency from the sub-task queue according to the scheduling execution sequence; establishing a communication hub storage structure for the sub-task queue; scheduling the target sub-task and obtaining the input parameters of the target sub-task, and writing the input parameters into the communication hub storage structure to convert them into execution parameters; executing the target sub-task using the execution parameters to obtain a target execution result; storing the target execution result in the communication hub storage structure; obtaining a subsequent sub-task that has a dependency relationship with the target sub-task and reading the target execution result of the target sub-task from the communication hub storage structure; using the target execution result as a completed parameter and filling the input parameters of the subsequent sub-task with the completed parameter to obtain a completed sub-task.

[0037] In detail, the execution order of each subtask and the dependency relationship between them are identified according to the subtask dependency graph. By analyzing the sequence of each subtask and other tasks, a scheduling execution sequence is generated based on the identified order and dependency relationship, which accurately arranges the execution order of each subtask and ensures that dependent tasks are completed in the correct order.

[0038] By analyzing the dependency relationship between each subtask and the previous task, target subtasks that have no previous dependency are filtered out. These tasks can be executed independently and do not need to wait for the completion of other tasks to start. This can efficiently identify subtasks that can be executed immediately and prepare for subsequent scheduling and execution, avoiding unnecessary waiting and resource waste.

[0039] The communication hub storage structure is used to centrally store and manage parameters and data during task execution. According to the scheduling sequence, the target subtask is selected for execution, and the input parameters required by the target subtask are obtained. These input parameters are written into the communication hub storage structure to form execution parameters. The communication hub storage structure ensures that various data required by the task can be read and transmitted in a timely manner during execution.

[0040] According to the execution parameters obtained from the communication hub storage structure, the execution of the target subtask is started. During execution, these execution parameters are used as input to complete the specific operations of the subtask according to the task requirements and logical rules. Through real-time calculation and processing, the execution result of the target subtask is finally obtained.

[0041] The execution result of the target subtask is stored in the communication hub storage structure to ensure that the result can be accessed and used by subsequent tasks. Subsequent subtasks that have a dependency relationship with the target subtask are identified and the stored execution result is read from the communication hub storage structure. The subsequent subtasks use these execution results as input parameters to ensure that the data flow and dependency relationship between tasks are correctly transmitted and processed.

[0042] The execution result of the target subtask is used as a completion parameter to fill in the input parameters of the subsequent subtasks that have a dependency relationship. By using these completion parameters, the subsequent subtasks obtain the complete input data required, ensuring that they can be executed smoothly. The completed subtasks will contain all the necessary parameters, eliminating the execution obstacles caused by missing data, and ensuring that each link in the task chain can be efficiently and accurately advanced.

[0043] Through precise task scheduling and execution management, efficient automated task flow can be realized. By identifying the dependency relationship between sub-tasks and generating a scheduling execution sequence, it can ensure that each sub-task is executed in the correct order, avoiding conflicts or errors caused by unordered execution. The communication hub storage structure provides a centralized data management platform, so that the task execution result can be timely delivered to the subsequent sub-tasks, ensuring parameter completion and smooth data flow between tasks, dynamically adjusting the task execution path, ensuring that the dependency relationship between tasks is effectively managed, ultimately improving execution efficiency, reducing delay, and improving the accuracy and reliability of task completion.

[0044] S3, the execution progress and history record of the completed sub-task are tracked by using a preset progress agent, to obtain an execution state and an execution history track.

[0045] In the embodiment of the application, the state of the completed sub-task is initialized by the preset progress agent, and a corresponding state machine model is established in the progress agent. According to the finite state machine transition rule, the execution state of the completed sub-task is updated and tracked in real time, and a state transition sequence is generated. By analyzing these state transitions, detailed execution state and execution history track can be generated, ensuring that every step in the task execution process is accurately tracked and recorded.

[0046] In the medical health specific scenario, it can be applied to help the intelligent diagnosis and treatment system to accurately track the patient treatment process. By monitoring and updating the execution state of each treatment step in real time, it can ensure that each diagnosis and treatment sub-task (such as drug administration, examination appointment, treatment plan adjustment, etc.) proceeds smoothly according to the predetermined order and dependency relationship. If a problem occurs in a certain step, the system will update the execution state and feedback the history track in time, helping doctors to make corresponding adjustments, thereby improving the efficiency and accuracy of diagnosis and treatment.

[0047] In the financial technology specific scenario, it can be applied to the intelligent risk control and transaction monitoring system. The system can track the progress and history record of transaction execution in real time, analyze each sub-task in the transaction process, such as risk assessment, fund audit, transaction execution, etc. By accurately recording the execution state and history track of each sub-task, the system can detect potential risks, adjust transaction strategies in time, ensure the safety and compliance of transaction process, and avoid loss and fraud events.

[0048] Figure 3 A flowchart of the sub-task execution progress tracking process in the task processing method based on multi-agent cooperation provided by an embodiment of the application is shown.

[0049] In the embodiment of the application, the execution progress and history record of the completed sub-task are tracked by using a preset progress agent, to obtain an execution state and an execution history track, including: state initialization is performed on the completion subtask to obtain a subtask initial state; A state machine model is established in a preset progress agent according to the subtask initial state; A finite state machine transition rule of the state machine model is obtained, and the execution state of the completion subtask is updated and tracked in real time according to the finite state machine transition rule to generate a state transition sequence; The execution state is generated according to the state transition sequence; The execution history track is obtained by tracking the history record of the completion subtask.

[0050] In detail, the initial state of the completion subtask is set, such as basic information of to-be-executed, being executed, or completed. Based on the initial state, a corresponding state machine model is established in a preset progress agent to describe different states and transition relationships of the subtask in the execution process. The state machine model defines the possible states of the subtask in the execution process and how to trigger the state transition according to the task progress or external factors, thereby realizing dynamic monitoring and management of the execution process of the subtask.

[0051] The finite state machine transition rule of the state machine model defines the transition conditions and trigger events between different states of the subtask. According to these rules, the execution process of the completion subtask is monitored in real time, and the execution state is dynamically updated according to the task progress or external input. When the execution conditions of the subtask meet the transition rule, the state change is automatically triggered, and the state transition sequence is generated, recording every change of the task from one state to another, providing detailed execution track for subsequent task optimization, adjustment, and backtracking.

[0052] According to the generated state transition sequence, the specific time and trigger condition of each state change are analyzed and extracted, thereby generating the execution state of the subtask, reflecting different stages of the subtask in the execution process, such as to-be-executed, being executed, and completed. The history record of the completion subtask is arranged in chronological order to form a complete execution history track, recording the execution process, state change, and key decision points of each subtask, providing detailed data support for subsequent task review, optimization, and problem troubleshooting.

[0053] The execution progress and history record of each completion subtask are accurately tracked and managed. Through the initialization of the task state, the establishment of the state machine model, and the real-time update of the execution state, each stage of the task can be dynamically monitored to ensure that the task is executed according to the predetermined steps and conditions. The generation of the state transition sequence makes the execution path of each subtask be clearly recorded, providing complete historical data for subsequent analysis and optimization, not only improving the transparency of task execution, but also enabling problems to be discovered and adjusted in a timely manner during the execution process, thereby improving the efficiency and accuracy of overall task management.

[0054] S4, generating an executable operation decision by a preset decision agent according to the execution state and the execution history track.

[0055] In the embodiment of the application, the current operation environment and environmental interaction information are obtained, and the information is extracted, filtered and normalized by an active perception module. The content perception of the completion subtask is performed by a text perception branch, the text range is located, and candidate task targets are generated. The information is fused with the execution state and the history track to form decision prompt information. The decision agent analyzes the prompt information, selects the optimal operation action in the limited action space, and converts it into an executable operation decision.

[0056] In the medical health specific scenario, it can be applied to an intelligent diagnosis and treatment decision support system. The system can perceive and analyze the patient's medical record data, treatment history and clinical environment in real time, identify the key treatment targets and tasks through text perception and environmental interaction information extraction, generate the optimal treatment plan by combining the patient's execution state and treatment history, automatically adjust the treatment plan, and provide operation decisions to help doctors accurately diagnose and treat. The automation and intelligence of this process improve the efficiency and accuracy of medical decision-making and reduce human errors.

[0057] In the financial technology specific scenario, it can be applied to an intelligent risk assessment and transaction decision system. The system perceives the interaction information of market data and customer transaction behavior, identifies potential risks and transaction targets, and generates decision prompts according to historical transaction records and real-time state. The system optimizes risk control and transaction decisions by combining transaction strategies and market dynamics, and automatically selects the optimal operation action. This technology can improve the decision-making speed of financial transactions, reduce operational errors, and improve the safety and yield of transactions.

[0058] In the embodiment of the application, the use of a preset decision agent to generate an executable operation decision according to the execution state and the execution history track comprises: Obtain the current operation environment, and extract environmental interaction information of the current operation environment using an interaction element perception branch of a preset active perception module; Filter the environmental interaction information to obtain filtered interaction information; Perform semantic normalization on the filtered interaction information to obtain interaction element information; Perform text content perception on the completion subtask using a text perception branch of a preset active perception module to obtain a completion task text range; Locate and identify the completion task text range in the current operation environment to obtain a text location result; Generate a set of candidate task targets according to the interaction element information and the text location result; fuse the candidate task target, the execution state and the execution history track to generate decision prompt information; perform decision analysis on the decision prompt information by using a preset decision agent, and filter out an optimal operation action in a preset limited action space according to an analysis result; convert the optimal operation action into an executable operation decision.

[0059] In detail, the interactive element perception branch of the active perception module is responsible for extracting interactive information related to the current operating environment, including external devices, sensors or user input data. These interactive information can cover dynamic changes of the environment, user behavior, device status, etc.

[0060] The extracted environmental interaction information is filtered to remove noise and irrelevant data, ensuring that only valid interaction information is retained. The filtered interaction information is subjected to semantic normalization processing, converting it into a standardized format or category, making the information more uniform, easy to understand and process.

[0061] The text perception branch in the active perception module is responsible for extracting task-related text information, identifying key content and context in the task, and determining the text range of the task completion. It can accurately locate the core information area of the task, ensuring that subsequent processing and decision-making focus on the most relevant part, providing more accurate task completion and execution support.

[0062] The identified text range of the task completion is matched with the current operating environment, and the specific location of the text range is identified in the environment through positioning algorithms, ensuring the correct association between the text content and the environment state, providing accurate reference for subsequent task execution and data processing.

[0063] The candidate task target is based on the accurate positioning of environmental interaction data and task text, representing the target that the current task may need to execute. These candidate targets are fused and analyzed with the execution state and history track of the task, considering the current progress of the task, past execution records and environmental factors, to generate decision prompt information.

[0064] The generated decision prompt information is analyzed by a preset decision agent to evaluate different schemes for task execution, and the optimal operation action is filtered out in the preset limited action space, i.e. the action scheme that best meets the task target and current environmental requirements, which is converted into a specific executable operation decision, ensuring that the task is executed in the most effective way.

[0065] By deeply integrating environmental perception, task execution status and historical data, intelligent decision-making and task optimization are realized, which can perceive the operation environment and task text information in real time, accurately identify the task target and analyze the execution status and historical trajectory, and provide comprehensive data support. By screening the optimal operation action and converting it into an executable decision, the system can automatically adjust the execution path to ensure that the task is pushed forward in the most effective way, which not only improves the efficiency and accuracy of task execution, but also dynamically adapts to environmental changes, reduces manual intervention, and optimizes the overall operation process.

[0066] S5, using a preset reflection agent to determine whether the task execution result of the operation decision achieves the expected effect.

[0067] In the embodiment of the application, the system interface state is compared with the task expected target, the execution historical trajectory and the context constraint are combined to determine whether the current operation makes the task execution result achieve the expected effect; if the determination result does not achieve the expected effect, the deviation reason is output and fed back to the decision-making agent for generating a new operation decision, so as to realize self-correction and iterative optimization of the task execution result; if the determination result achieves the expected effect, the task completion information is fed back to the progress agent to confirm the task completion.

[0068] If the task execution result of the operation decision does not achieve the expected effect, s6, the generated result difference information and correction suggestion are fed back to the preset decision-making agent, and the preset decision-making agent after receiving the feedback is used to adjust the operation decision, generate an operation update decision, and execute the completion subtask using the operation update decision to obtain a task completion result.

[0069] In the embodiment of the application, the generated result difference information and correction suggestion are fed back to the preset decision-making agent, and the preset decision-making agent after receiving the feedback is used to adjust the operation decision, generate an operation update decision, and execute the completion subtask using the operation update decision to obtain a task completion result, which comprises: According to the operation decision, the completion subtask is executed to obtain a task execution result; Using a preset reflection agent to analyze the difference between the task execution result and the expected target, and generating result difference information; According to the result difference information, the operation decision execution process of the completion subtask is adjusted to generate alternative operation actions and an adjusted execution path; According to the alternative operation action and the adjusted execution path, a correction suggestion is generated; The result difference information and the correction suggestion are fed back to the preset decision-making agent; Using the preset decision-making agent after receiving the feedback, the operation decision is adjusted based on the correction suggestion to generate an operation update decision; The operation decision is used to update the decision to perform the completion subtask, and the final execution result is taken as the task completion result.

[0070] In detail, the completion subtask is executed according to the generated operation decision, and the actual execution result of the task is obtained, and the preset reflection agent identifies the deviation or deficiency in the execution process. By analyzing these differences, the reflection agent generates result difference information, which details the parts that do not meet the expected target in the task execution.

[0071] According to the result difference information, the operation decision execution process of the completion subtask is analyzed, the part that needs to be adjusted is identified, and alternative operation actions and optimized execution paths are generated. By exploring different operation schemes and adjusting steps, a more effective execution scheme is provided for the task, and a correction suggestion is generated, which helps to better achieve the task target and ensure that the task execution is more smooth and efficient.

[0072] The generated result difference information and correction suggestion are fed back to the preset decision agent, and the decision agent analyzes the relationship between the current operation decision and the correction suggestion according to the received feedback information, identifies the part that needs to be adjusted, optimizes and adjusts the operation decision, and generates an updated operation decision, so that the system can self-improve and make accurate decision adjustment when there is a difference.

[0073] By continuously feeding back and optimizing the decision, the continuous improvement and accuracy of the task execution are ensured. By analyzing the difference between the execution result and the expectation through the reflection agent, the operation decision can be identified and adjusted in real time to ensure that the execution path and strategy of the task are always consistent with the target. The generated correction suggestion and difference information are fed back to the decision agent, which further adjusts the operation decision, thereby avoiding the accumulation of decision errors or deviations and improving the flexibility and accuracy of task execution. This closed-loop feedback mechanism ensures the efficiency of task execution, reduces risks, and improves the adaptive ability of decision-making.

[0074] If the task execution result of the operation decision reaches the expected effect, s7, the task completion information is sent to the preset progress agent, and the execution state of the completion subtask is updated to task completion by the preset progress agent after receiving the information.

[0075] In the embodiment of the present application, the task completion information is sent to the preset progress agent, and the execution state of the completion subtask is updated to task completion by the preset progress agent after receiving the information, which comprises: The operation decision is used to update the decision to perform the completion subtask, and the final execution result is taken as the task completion result. The operation decision is used to update the decision to perform the completion subtask, and the final execution result is taken as the task completion result. Send the task completion information to the preset progress agent, and trigger the finite state machine transition rule of the preset progress agent after receiving the information; According to the finite state machine transition rule, update the execution state of the completed subtask to task completion.

[0076] In detail, the completed subtask is executed according to the operation decision, and the actual execution result of the task is obtained. The corresponding task completion information is generated according to the task execution result, which records whether the task is successfully completed as expected and whether any problem or deviation occurs in the execution process. Not only the execution of the task is summarized, but also data support is provided for the scheduling, optimization and evaluation of subsequent tasks, ensuring that the system can make appropriate adjustments according to the actual execution situation.

[0077] The task completion information is sent to the preset progress agent as feedback of the task execution. After receiving the information, the preset progress agent processes according to the finite state machine transition rule, analyzes the completion state of the task, and triggers the state transition. According to these transition rules, the execution state of the completed subtask is updated to 'task completion', indicating that the task has been successfully completed and meets the expected goal. This update ensures real-time monitoring of the task execution progress and provides accurate state information for the continuation or scheduling of subsequent tasks.

[0078] Through accurate task state update and progress monitoring, the accuracy and real-time feedback of task execution are ensured. The system executes the task through operation decision and generates task completion information, which is delivered to the preset progress agent to realize automatic update of the task execution progress. Through the finite state machine transition rule, the progress agent can automatically identify the completion state of the task and update the execution state to 'task completion', ensuring the smooth progress of the task and the reasonable scheduling of subsequent tasks. Not only the efficiency of task management is improved, but also each link in the task execution process is timely tracked and adjusted.

[0079] It should be understood that the size of the serial number of each step in the above embodiment does not mean the order of execution, and the execution order of each process should be determined by its function and inherent logic, and should not constitute any limitation on the implementation process of the embodiment of the present application.

[0080] As Figure 4 shown is a functional module diagram of a task processing device based on multi-agent cooperation provided by an embodiment of the present application.

[0081] In the embodiment of the present disclosure, a task processing device based on multi-agent cooperation is provided, which corresponds one-to-one with the above-mentioned task processing method based on multi-agent cooperation. As Figure 4As shown, the task processing device 100 based on multi-agent cooperation can be installed in an electronic device. According to the implemented functions, the task processing device 100 based on multi-agent cooperation includes a dependency graph construction module 101, a subtask parameter completion module 102, an execution progress tracking module 103, an operation decision generation module 104, an execution effect judgment module 105, an operation decision adjustment module 106, and an execution state updating module 107. The detailed description of each functional module is as follows: The dependency graph construction module 101 is configured to obtain a task instruction of a target user, perform high-level semantic analysis and decomposition on the task instruction by using a preset management agent, and construct a subtask dependency graph according to a subtask queue obtained by decomposition. The subtask parameter completion module 102 is configured to sequentially schedule and execute the subtask queue according to the subtask dependency graph, and use a task execution result of execution completion to complete the parameter of a next subtask in the subtask queue to obtain a completed subtask. The execution progress tracking module 103 is configured to track the execution progress and historical record of the completed subtask by using a preset progress agent to obtain an execution state and an execution history track. The operation decision generation module 104 is configured to generate an executable operation decision according to the execution state and the execution history track by using a preset decision agent. The execution effect judgment module 105 is configured to judge whether the task execution result of the operation decision achieves an expected effect by using a preset reflection agent. The operation decision adjustment module 106 is configured to, if the task execution result of the operation decision does not achieve the expected effect, feed back result difference information and correction suggestions generated to the preset decision agent, adjust the operation decision by using the preset decision agent after receiving the feedback, generate an operation update decision, execute the completed subtask by using the operation update decision, and obtain a task completion result. The execution state updating module 107 is configured to, if the task execution result of the operation decision achieves the expected effect, send task completion information to the preset progress agent, and update the execution state of the completed subtask to task completion by using the preset progress agent after receiving the information.

[0082] In an embodiment, the dependency graph construction module 101 performs high-level semantic analysis and decomposition on the task instruction by using a preset management agent, and constructs a subtask dependency graph according to a subtask queue obtained by decomposition, including: Extracting instruction language features of the task instruction, and extracting task key elements from the instruction language features; Performing semantic labeling on the task instruction by using the task key elements to obtain a labeled instruction. correlate the labeling instruction with context by using a preset management intelligent agent to obtain context semantic information; generate a logical structure of the task instruction according to the task critical element and the context semantic information; convert the logical structure into a high-level semantic representation; decompose the labeling instruction based on the high-level semantic representation to generate a subtask queue; identify a dependency relationship between the subtask queue, and map the subtask queue into a subtask dependency relationship graph according to the dependency relationship.

[0083] In an embodiment, the subtask parameter completion module 102 sequentially schedules and executes the subtask queue according to the subtask dependency relationship graph, and uses the task execution result of the execution to complete parameter completion of the next subtask in the subtask queue to obtain a completed subtask, including: identify a subtask execution order and a subtask dependency relationship according to the subtask dependency relationship graph; generate a scheduling execution sequence according to the subtask execution order and the subtask dependency relationship; perform task screening on the subtask queue according to the scheduling execution sequence to screen out a target subtask without a previous dependency; establish a communication hub storage structure for the subtask queue; schedule the target subtask and obtain an input parameter of the target subtask, write the input parameter into the communication hub storage structure, and then convert it into an execution parameter; execute the target subtask by using the execution parameter to obtain a target execution result; store the target execution result in the communication hub storage structure; obtain a subsequent subtask that has a dependency relationship with the target subtask, and read the target execution result of the target subtask from the communication hub storage structure; use the target execution result as a completion parameter, and fill the input parameter of the subsequent subtask with the completion parameter to obtain a completed subtask.

[0084] In an embodiment, the execution progress tracking module 103 tracks the execution progress and historical record of the completed subtask by using a preset progress intelligent agent to obtain an execution state and an execution history track, including: initialize the state of the completed subtask to obtain a subtask initial state; establish a state machine model in the preset progress intelligent agent according to the subtask initial state; acquire a finite state machine transition rule of the state machine model, update and track the execution state of the completion subtask in real time according to the finite state machine transition rule, and generate a state transition sequence; generate an execution state according to the state transition sequence; track a history record of the completion subtask to obtain an execution history track.

[0085] In an embodiment, the operation decision generation module 104 executes the operation decision generation using a preset decision agent according to the execution state and the execution history track, including: acquire a current operation environment, and extract environmental interaction information of the current operation environment using an interaction element perception branch of a preset active perception module; filter the environmental interaction information to obtain filtered interaction information; perform semantic normalization on the filtered interaction information to obtain interaction element information; perform text content perception on the completion subtask using a text perception branch of the preset active perception module to obtain a completion task text range; perform positioning recognition on the completion task text range in the current operation environment to obtain a text positioning result; generate a set of candidate task targets according to the interaction element information and the text positioning result; fuse the candidate task targets, the execution state, and the execution history track to generate decision prompt information; perform decision analysis on the decision prompt information using a preset decision agent, and filter an optimal operation action in a preset limited action space according to an analysis result; convert the optimal operation action into an executable operation decision.

[0086] In an embodiment, the operation decision adjustment module 106 executes the generation of result difference information and correction suggestions to a preset decision agent, and adjusts the operation decision using the preset decision agent after receiving the feedback to generate an operation update decision, executes the completion subtask using the operation update decision, and obtains a task completion result, including: execute the completion subtask according to the operation decision to obtain a task execution result; perform analysis on the difference between the task execution result and an expected target using a preset reflection agent to generate result difference information; adjust the operation decision execution process of the completion subtask according to the result difference information to generate a replaceable operation action and an adjustment execution path; generate a correction suggestion according to the replaceable operation action and the adjustment execution path; feedback the result difference information and the correction suggestion to a preset decision agent; adjust the operation decision based on the correction suggestion by using the preset decision agent receiving the feedback, to generate an operation update decision; execute the completion subtask by using the operation update decision, and take the final execution result as the task completion result.

[0087] In an embodiment, the execution state updating module 107 executes sending the task completion information to a preset progress agent, and updates the execution state of the completion subtask to task completion by using the preset progress agent receiving the information, including: execute the completion subtask according to the operation decision, to obtain a task execution result; generate task completion information of the completion subtask according to the task execution result; send the task completion information to a preset progress agent, and trigger the finite state machine transition rule of the preset progress agent after receiving the information; update the execution state of the completion subtask to task completion according to the finite state machine transition rule.

[0088] In the present application, for a task processing device based on multi-agent cooperation, first, the present application obtains the task instruction of the target user, uses the preset management agent to perform high-level semantic analysis and decomposition on the task instruction, and constructs a subtask dependency graph according to the subtask queue obtained by decomposition, which can effectively identify the sequence and dependency relationship between tasks, thereby realizing optimized scheduling of tasks and efficient use of resources, enhancing the automation, intelligence and accuracy of task execution, scheduling and executing the subtask queue according to the subtask dependency graph, and using the executed task execution result to complete the parameter completion of the next subtask in the subtask queue to obtain a completed subtask. The communication hub storage structure provides a centralized data management platform, so that the task execution result can be timely transmitted to the subsequent subtask, ensuring smooth data flow between parameters and tasks. The preset progress agent tracks the execution progress and historical record of the completed subtask to obtain the execution state and execution history track, then the preset decision agent generates executable operation decisions according to the execution state and the execution history track, filters the optimal operation action and converts it into executable decisions, the system can automatically adjust the execution path to ensure that the task is pushed forward in the most effective way, finally, the preset reflection agent judges whether the task execution result of the operation decision reaches the expected effect, if the task execution result of the operation decision does not reach the expected effect, the generated result difference information and correction suggestion are fed back to the preset decision agent, and the preset decision agent after receiving the feedback is used to adjust the operation decision to generate an operation update decision, and the completed subtask is executed by using the operation update decision to obtain a task completion result. It can identify and adjust the operation decision in real time to ensure that the execution path and strategy of the task are always consistent with the target, this closed-loop feedback mechanism ensures the efficiency of task execution, reduces the risk, and improves the adaptive ability of decision making. If the task execution result of the operation decision reaches the expected effect, the task completion information is sent to the preset progress agent, and the execution state of the completed subtask is updated to task completion by using the preset progress agent after receiving the information. The operation decision executes the task and generates task completion information, which is transmitted to the preset progress agent to realize automatic updating of the task execution progress, improve the perception accuracy, and have a task dependency tracking and feedback correction mechanism, and successfully realize cross-application automated tasks. The specific limitations of the task processing device based on multi-agent cooperation can be referred to the limitations of the task processing method based on multi-agent cooperation in the above, which will not be repeated here. The modules in the above task processing device based on multi-agent cooperation can be realized by software, hardware and their combinations. The above modules can be embedded in or independent of the processor in the computer device in hardware form, or stored in the memory in the computer device in software form, so as to call and execute the operations of the above modules by the processor.

[0089] In one embodiment, a computer device is provided, which can be a server, and an internal structure diagram thereof can be as shown in Figure 5 The computer device includes a processor, a memory, a network interface and a database connected through a system bus. The processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device includes a non-volatile and / or volatile storage medium, an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operating system and the computer program in the non-volatile storage medium to run. The network interface of the computer device is configured to communicate with an external client through a network connection. The computer program is executed by the processor to implement functions or steps of a server side of a task processing method based on multi-agent cooperation.

[0090] In one embodiment, a computer device is provided, which can be a client, and an internal structure diagram thereof can be as shown in Figure 6 The computer device includes a processor, a memory, a network interface, a display screen and an input device connected through a system bus. The processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operating system and the computer program in the non-volatile storage medium to run. The network interface of the computer device is configured to communicate with an external server through a network connection. The computer program is executed by the processor to implement functions or steps of a client side of a task processing method based on multi-agent cooperation.

[0091] In one embodiment, a computer device is provided, which includes a memory, a processor and a computer program stored in the memory and executable on the processor, and the processor implements the following steps when executing the computer program: obtaining a task instruction of a target user, performing high-level semantic analysis and decomposition on the task instruction by using a preset management agent, and constructing a subtask dependency graph according to a subtask queue obtained by decomposition; sequentially scheduling and executing the subtask queue according to the subtask dependency graph, and using a task execution result of an executed task to complete parameters of a next subtask in the subtask queue to obtain a completed subtask; tracking execution progress and historical records of the completed subtask by using a preset progress agent to obtain an execution state and an execution historical track; generating an executable operation decision according to the execution state and the execution historical track by using a preset decision agent; The preset reflection agent is used to judge whether the task execution result of the operation decision reaches the expected effect or not. If the task execution result of the operation decision does not reach the expected effect, the generated result difference information and the correction suggestion are fed back to the preset decision agent, the preset decision agent after receiving the feedback is used to adjust the operation decision, an operation update decision is generated, the operation update decision is used to execute the completion subtask, and a task completion result is obtained. If the task execution result of the operation decision reaches the expected effect, task completion information is sent to the preset progress agent, and the preset progress agent after receiving the information is used to update the execution state of the completion subtask to task completion.

[0092] In several embodiments provided by the present application, it should be understood that the disclosed devices and apparatuses can be implemented in other manners. For example, the above described system embodiments are merely illustrative. For example, the division of the modules is merely logical function division. In actual implementation, another division manner can be used.

[0093] In addition, each function module in each embodiment of the present application can be integrated in a processing unit, or each unit can exist physically, or two or more units can be integrated in one unit. The above integrated unit can be realized in the form of hardware, or in the form of hardware plus software function module.

[0094] Therefore, no matter from which point of view, the embodiments should be regarded as exemplary and non-limiting, and the scope of the present application is defined by the appended claims rather than the above description, and therefore all changes falling within the meaning and scope of the equivalent elements of the claims are intended to be included in the present application. Any reference signs in the claims should not be regarded as limiting the claims involved.

[0095] It is obvious for those skilled in the art that the present application is not limited to the details of the above exemplary embodiments, and the present application can be implemented in other specific forms without departing from the spirit or essential characteristics of the present application.

[0096] In some embodiments of the present embodiment, a computer readable storage medium is provided, and a computer program is stored on the computer readable storage medium, wherein the computer program is executed by a processor to implement the steps of the method in the above embodiments.

[0097] The readable storage medium of the present application stores a computer program, and the computer program can implement the following when executed by a processor of an electronic device: Obtain a task instruction of a target user, perform high-level semantic analysis and decomposition on the task instruction by using a preset management agent, and construct a subtask dependency graph according to a subtask queue obtained by decomposition; According to the subtask dependency graph, the subtask queue is sequentially scheduled and executed, and the task execution result of the executed task is used for parameter completion of a next subtask in the subtask queue, to obtain a completed subtask; The execution progress and history record of the completed subtask are tracked by using a preset progress agent, to obtain an execution state and an execution history track; An executable operation decision is generated by using a preset decision agent according to the execution state and the execution history track; It is judged by using a preset reflection agent whether a task execution result of the operation decision reaches an expected effect; If the task execution result of the operation decision does not reach the expected effect, result difference information and a correction suggestion generated are fed back to the preset decision agent, the preset decision agent after receiving the feedback is used to adjust the operation decision, an operation update decision is generated, the completed subtask is executed by using the operation update decision, and a task completion result is obtained; If the task execution result of the operation decision reaches the expected effect, task completion information is sent to the preset progress agent, and the execution state of the completed subtask is updated to task completion by using the preset progress agent after receiving the information.

[0098] It should be noted that the functions or steps of the computer readable storage medium or the computer device described above can be referred to the related descriptions of the server side and the client side in the foregoing method embodiments. To avoid repetition, they will not be described one by one here.

[0099] The computer readable storage medium can also store at least one computer executable program / instruction, such as computer readable instructions. The computer readable storage medium includes, but is not limited to, for example, volatile memory and / or non-volatile memory. The volatile memory may, for example, include random access memory (RAM) and / or cache memory, etc. The computer readable storage medium may, for example, include read-only memory (ROM), hard disk, flash memory, etc. For example, the non-transitory computer readable storage medium can be connected to a computing device such as a computer, and then when the computing device runs the computer readable instructions stored on the computer readable storage medium, the various methods described above can be performed.

[0100] In addition, the computer device can also include (but not limited to) a data bus, an input / output (I / O) bus, a display, and an input / output device (for example, a keyboard, a mouse, a speaker, etc.), etc.

[0101] The processor can communicate with the external device through a wired or wireless network through an I / O bus.

[0102] In one embodiment, the at least one computer-executable instruction can also be compiled or composed into a software product / computer program product, wherein the one or more computer-executable instructions are executed by the processor to perform the steps of the various functions and / or methods described in the embodiments of the present technology.

[0103] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be completed by a computer program instructing related hardware, and the computer program can be stored in a non-volatile computer readable storage medium. When the computer program is executed, it can include the processes of the above-mentioned embodiments. Any reference to memory, storage, database, or other medium used in the embodiments provided by the present application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration but not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0104] Those skilled in the art can clearly understand that, for the convenience and brevity of description, only the above-mentioned division of functional units and modules is exemplified, and in actual application, the above-mentioned functions can be completed by different functional units and modules according to needs, that is, the internal structure of the device is divided into different functional units or modules to complete all or part of the above-described functions.

[0105] In the embodiments provided by the present disclosure, it should be understood that the disclosed apparatus and method can also be implemented in other manners. The embodiments described above are merely exemplary for describing the present disclosure. For example, the flowcharts and block diagrams in the accompanying drawings show the possible implementation architectures, functions and operation of the apparatus, method and computer program product according to the embodiments of the present disclosure. In this regard, each block in the flowcharts and block diagrams can represent a module, a program segment or a part of code, which contains one or more executable instructions for implementing the specified logic function. It should also be noted that, in some alternative implementations, the functions noted in the blocks can occur in different orders from those noted in the accompanying drawings. For example, two consecutive blocks can actually be executed substantially in parallel, and sometimes they can be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and the combination of blocks in the block diagrams and / or flowcharts, can be implemented by a special-purpose hardware-based system for implementing the specified functions or actions, or can be implemented by a combination of special-purpose hardware and computer instructions.

[0106] It should be noted that, in the present disclosure, the term "comprising" or "including" or any other variant thereof is intended to cover non-exclusive inclusion, so that the process, method, article or device including a series of elements not only includes those elements, but also includes other elements not explicitly listed, or further includes elements inherent to such process, method, article or device. Without more limitations, the element limited by the statement "including a" does not exclude the presence of additional same elements in the process, method, article or device including the element.

[0107] The above-described embodiments are merely used to illustrate the technical solutions of the present disclosure, rather than limit them; although the present disclosure has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: the technical solutions recorded in the foregoing embodiments can be modified, or some technical features can be replaced by equivalent replacements; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present disclosure, and should be included in the protection scope of the present disclosure.

[0108] It should be noted that, in the embodiments of the present disclosure, if non-company software tools or components appear, they are only used for example introduction, and do not represent actual use.

Claims

1. A task processing method based on multi-agent cooperation, characterized in that, The method comprises: obtaining a task instruction of a target user, performing high-level semantic analysis and decomposition on the task instruction by using a preset management agent, and constructing a subtask dependency graph according to a subtask queue obtained by decomposition; sequentially scheduling and executing the subtask queue according to the subtask dependency graph, and using a task execution result of execution completion to complete parameter completion of a next subtask in the subtask queue, to obtain a completed subtask; tracking execution progress and historical records of the completed subtask by using a preset progress agent, to obtain an execution state and an execution historical track; generating an executable operation decision according to the execution state and the execution historical track by using a preset decision agent; judging whether a task execution result of the operation decision achieves an expected effect by using a preset reflection agent; if the task execution result of the operation decision does not achieve the expected effect, feeding back result difference information and a correction suggestion generated to the preset decision agent, and adjusting the operation decision by using the preset decision agent receiving the feedback, to generate an operation update decision, and executing the completed subtask by using the operation update decision, to obtain a task completion result; if the task execution result of the operation decision achieves the expected effect, sending task completion information to the preset progress agent, and updating the execution state of the completed subtask to task completion by using the preset progress agent receiving the information. 2.The method of claim 1, wherein, The method comprises: extracting instruction language features of the task instruction, and extracting task key elements from the instruction language features; performing semantic labeling on the task instruction by using the task key elements, to obtain labeled instructions; performing context association on the labeled instructions by using a preset management agent, to obtain context semantic information; generating a logical structure of the task instruction according to the task key elements and the context semantic information; converting the logical structure into a high-level semantic representation; performing decomposition on the labeled instructions based on the high-level semantic representation, to generate a subtask queue; identifying dependency relationships between the subtask queue, and mapping the subtask queue into a subtask dependency graph according to the dependency relationships. 3.The method of claim 1, wherein, The method comprises: identifying a subtask execution order and a subtask dependency relationship according to the subtask dependency graph; generating a scheduling and execution sequence according to the subtask execution order and the subtask dependency relationship; performing task screening on the subtask queue according to the scheduling and execution sequence, to screen out a target subtask without a precedent dependency; establishing a communication hub storage structure for the subtask queue; scheduling the target subtask, and obtaining input parameters of the target subtask, converting the input parameters into execution parameters after writing the input parameters into the communication hub storage structure; Perform the target subtask by using the execution parameter to obtain a target execution result; Store the target execution result in the communication hub storage structure; Obtain a subsequent subtask that has a dependency relationship with the target subtask, and read the target execution result of the target subtask from the communication hub storage structure; Use the target execution result as a completion parameter, and fill the input parameter of the subsequent subtask by using the completion parameter to obtain a completion subtask. 4.The method of claim 1, wherein, The execution progress and history record of the completion subtask are tracked by using a preset progress agent to obtain an execution state and an execution history track, including: State initialization is performed on the completion subtask to obtain a subtask initial state; A state machine model is established in the preset progress agent according to the subtask initial state; A finite state machine transition rule of the state machine model is obtained, and the execution state of the completion subtask is updated and tracked in real time according to the finite state machine transition rule to generate a state transition sequence; An execution state is generated according to the state transition sequence; The history record of the completion subtask is tracked to obtain an execution history track. 5.The method of claim 1, wherein, An executable operation decision is generated according to the execution state and the execution history track by using a preset decision agent, including: A current operation environment is obtained, and environment interaction information of the current operation environment is extracted by using an interactive element perception branch of a preset active perception module; The environment interaction information is filtered to obtain filtered interaction information; The filtered interaction information is semantically normalized to obtain interactive element information; Text content perception is performed on the completion subtask by using a text perception branch of the preset active perception module to obtain a completion task text range; The completion task text range is positioned and recognized in the current operation environment to obtain a text positioning result; A group of candidate task targets are generated according to the interactive element information and the text positioning result; The candidate task targets, the execution state, and the execution history track are fused to generate decision prompt information; The decision prompt information is analyzed by using the preset decision agent, and the optimal operation action is selected in a preset limited action space according to the analysis result; The optimal operation action is converted into an executable operation decision. 6.The method of claim 1, wherein, The generated result difference information and the correction suggestion are fed back to the preset decision agent, and the operation decision is adjusted by using the preset decision agent receiving the feedback to generate an operation update decision, and the completion subtask is executed by using the operation update decision to obtain a task completion result, including: The completion subtask is executed according to the operation decision to obtain a task execution result; The difference between the task execution result and the expected target is analyzed by using a preset reflection agent to generate result difference information; The operation decision execution process of the completion subtask is adjusted according to the result difference information to generate an alternative operation action and an adjustment execution path; A correction suggestion is generated according to the alternative operation action and the adjustment execution path; The result difference information and the correction suggestion are fed back to the preset decision agent; The preset decision agent after receiving the feedback is utilized to adjust the operation decision based on the correction suggestion, and an operation update decision is generated; The operation update decision is utilized to execute the completion subtask, and a final execution result is obtained as a task completion result. 7.The method of claim 1, wherein, The task completion information is sent to the preset progress agent, and the execution state of the completion subtask is updated to task completion by the preset progress agent after receiving the information. The operation decision is executed according to the operation decision, and a task execution result is obtained; Task completion information of the completion subtask is generated according to the task execution result; The task completion information is sent to the preset progress agent, and a finite state machine transition rule of the preset progress agent is triggered after receiving the information; The execution state of the completion subtask is updated to task completion according to the finite state machine transition rule.

8. A task processing apparatus based on multi-agent cooperation, characterized by comprising: The device comprises: A dependency graph construction module is configured to obtain a task instruction of a target user, perform high-level semantic analysis and decomposition on the task instruction by using a preset management agent, and construct a subtask dependency graph according to a subtask queue obtained by the decomposition; A subtask parameter completion module is configured to sequentially schedule and execute the subtask queue according to the subtask dependency graph, and use a task execution result obtained by the execution to complete parameter completion of a next subtask in the subtask queue, to obtain a completion subtask; An execution progress tracking module is configured to track execution progress and historical records of the completion subtask by using a preset progress agent, to obtain an execution state and an execution historical trajectory; An operation decision generation module is configured to generate an executable operation decision according to the execution state and the execution historical trajectory by using a preset decision agent; An execution effect judgment module is configured to judge whether a task execution result of the operation decision reaches an expected effect by using a preset reflection agent; An operation decision adjustment module is configured to, if the task execution result of the operation decision does not reach the expected effect, feed back result difference information and a correction suggestion to the preset decision agent, and adjust the operation decision by using the preset decision agent after receiving the feedback, to generate an operation update decision, execute the completion subtask by using the operation update decision, and obtain a task completion result; An execution state updating module is configured to, if the task execution result of the operation decision reaches the expected effect, send task completion information to the preset progress agent, and update the execution state of the completion subtask to task completion by using the preset progress agent after receiving the information.

9. An electronic device, comprising: The electronic device comprises: at least one processor; and a memory connected with the at least one processor in communication; wherein The memory stores a computer program that can be executed by the at least one processor, and the computer program is executed by the at least one processor to enable the at least one processor to execute the task processing method based on multi-agent cooperation according to any one of claims 1 to 7.

10. A computer readable storage medium storing a computer program, characterized in that, The computer program is executed by the processor to implement the task processing method based on multi-agent cooperation according to any one of claims 1 to 7.

Images