Task processing method and device, computer device, and storage medium

Abstract

The present application relates to the technical field of industrial internet and multi-agent cooperation, and discloses a task processing method and device, computer equipment, storage medium and product, the method comprises the following steps: splitting a to-be-executed task to obtain a plurality of to-be-processed subtasks and task attribute information corresponding to each to-be-processed subtask; screening at least one target agent matched with the to-be-processed subtask; determining available agents; generating an agent group corresponding to the to-be-executed task; and executing the to-be-executed task according to the agent group. The present application can dynamically organize available agents according to task requirements, avoids low efficiency caused by forced requisition of available agents, and improves marshalling flexibility and task execution efficiency.

Description

Technical Field

[0001] This invention relates to the fields of industrial internet and multi-agent collaboration technology, specifically to task processing methods, devices, computer equipment, and storage media. Background Technology

[0002] With the rapid development of artificial intelligence technology, especially the rapid iteration and maturation of large-scale language models, intelligent agents powered by these models are gradually penetrating various industries, driving the transformation of these fields towards intelligence. In the industrial sector, the application of multi-agent systems is becoming increasingly widespread. Through the collaborative efforts of multiple intelligent agents with autonomous decision-making capabilities, multi-agent systems can accomplish complex industrial tasks that are difficult for a single intelligent agent to handle.

[0003] Currently, multi-agent cooperative technology mainly adopts a static grouping approach, which predefines the role, capability boundaries, cooperative relationships, and task allocation rules of each agent during the system design phase. Once the grouping is determined, the cooperative relationships between agents remain unchanged regardless of changes in task requirements, and cannot be dynamically adjusted with the task. Summary of the Invention

[0004] In view of this, the present invention provides a task processing method, apparatus, computer device and storage medium to solve the problem that intelligent agents cannot dynamically adjust with the task in the prior art.

[0005] In a first aspect, the present invention provides a task processing method, the method comprising: The task to be executed is broken down to obtain multiple sub-tasks to be processed and the task attribute information corresponding to each sub-task; For each of the sub-tasks to be processed, at least one target agent that matches the sub-task is selected based on the task attribute information corresponding to the sub-task to be processed. For each of the sub-tasks to be processed, a usable agent is determined from each of the target agents; Based on the matching relationship between the available agents and the sub-tasks to be processed, an agent group corresponding to the task to be executed is generated; The task to be performed is executed according to the group of intelligent agents.

[0006] By selecting target agents for each subtask, precise matching of subtasks is achieved, improving the adaptation accuracy between tasks and agents. By identifying available agents from each target agent, it is ensured that the final selected agents have actual execution capabilities. By generating agent groups based on matching relationships, dynamic teaming on demand is achieved, allowing the grouping structure to be flexibly adjusted according to task changes. Finally, through the collaborative execution of tasks by agent groups, this application can dynamically organize available agents according to task requirements, avoiding the inefficiency caused by the forced requisition of available agents, and improving grouping flexibility and task execution efficiency.

[0007] In one optional implementation, the step of filtering target agents matching the sub-task to be processed based on the task attribute information corresponding to the sub-task to be processed includes: Based on the task attribute information corresponding to the subtask to be processed, a task vector corresponding to the subtask to be processed is generated; Generate agent vectors corresponding to each agent based on the ability attribute information of each agent. Calculate the vector similarity between the vectors based on the task vector and each of the agent vectors; Based on vector similarity and a pre-set similarity threshold, at least one candidate similarity is obtained; From the agents corresponding to each candidate similarity, the target agent that matches the subtask to be processed is selected.

[0008] By mapping subtasks and agents to a shared semantic space for quantitative matching, the semantic system differences between agents from different vendors are eliminated, enabling heterogeneous agents to be accurately matched in a unified numerical space. This solves the problem of semantic interoperability barriers between heterogeneous agents in a loosely coupled architecture.

[0009] In one optional implementation, the step of selecting the target agent matching the subtask to be processed from the agents corresponding to each candidate similarity includes: Based on the first constraint, a selection process is performed to obtain candidate agents. The first constraint is used to characterize the conditions that the agent must satisfy for the sub-task to be processed. Based on the second constraint, the first constraint, and the vector similarity, the constraint values ​​corresponding to the candidate agents are calculated. The second constraint is used to characterize the conditions that can improve the trust value by solving the sub-task to be processed based on the target agent. The candidate agent with the highest constraint value is determined as the target agent.

[0010] By combining the veto mechanism of the first constraint with the weighted scoring of the second constraint, further constraint screening is performed based on semantic matching. This ensures that the hard bottom line is not breached while allowing for optimal selection based on soft indicators, so that the agent's matching results simultaneously satisfy semantic relevance and constraint feasibility, thus solving the problem of the one-sidedness of relying solely on semantic matching or rule matching.

[0011] In one optional implementation, the constraint values ​​corresponding to the candidate agents are calculated based on the second constraint and the first constraint. Based on the degree of satisfaction of the second and first constraints, calculate the satisfaction score of the candidate agents; The constraint values ​​for the candidate agents are calculated based on the satisfaction score, vector similarity, and corresponding weights.

[0012] By combining semantic similarity and constraint satisfaction through a weighted fusion approach, a comprehensive score is formed. In practical applications, the weight ratio of the two dimensions is flexibly adjusted according to different scenarios to achieve an optimal balance between semantic relevance and constraint feasibility, thereby further improving the accuracy and suitability of agent selection.

[0013] In one optional implementation, determining available agents from the target agents for each of the subtasks to be processed includes: Send grouping invitations to each of the target intelligent agents; Receive response information returned by each of the target intelligent agents based on their own state information; Based on the response information, the target agent that accepts the grouping invitation is identified as the available agent.

[0014] By delegating some decision-making power to intelligent agents through an invitation mechanism, the agents have the opportunity to autonomously determine whether to accept a task based on their real-time status. The complete information carried in the group invitation provides sufficient basis for the agents' autonomous decision-making, enabling them to make rational judgments based on a full understanding of the task requirements. In an optional implementation, the step of executing the task to be executed according to the group of intelligent agents includes: Based on the matching relationship between the available intelligent agent and the subtask to be processed, a marshalling contract is generated; the marshalling contract is used to constrain the input data format, output data format and quality requirements of the available intelligent agent when executing the subtask; Based on the resource capabilities of each available agent and the requirements of the sub-tasks, the available resources are allocated to each available agent in the agent group. Control each available intelligent agent to execute the corresponding sub-task based on the grouping contract and available resources; The execution process is monitored, and the monitoring results are obtained.

[0015] By monitoring the execution of sub-tasks according to the grouping contract, it is ensured that the intelligent agent group has clear execution specifications, sufficient resource guarantees, and real-time status feedback during task execution. The grouping contract standardizes and regulates the execution behavior of intelligent agents, making the execution results predictable, measurable, and traceable, significantly improving the reliability and controllability of multi-agent collaborative execution.

[0016] In an optional implementation, if the monitoring result indicates execution failure, the method further includes: Obtain the task attribute information and incomplete progress information corresponding to the failed subtask; Based on the task attribute information and the incomplete progress information, alternative agents matching the subtask are re-selected from the target agent set; A grouping invitation is sent to the alternative agent, and the alternative agent that accepts the invitation is added to the agent group to continue executing the subtask.

[0017] Through the failure recovery mechanism, the multi-agent system has automatic fault tolerance and self-recovery capabilities when faced with agent execution failures. It retains completed progress data, avoids repetitive work, significantly improves the completion rate of multi-agent collaborative processing of pending tasks, and effectively supports the high reliability requirements of the system in industrial scenarios.

[0018] In a second aspect, the present invention provides a task processing apparatus, the apparatus comprising: The splitting module is used to split the task to be executed into multiple subtasks to be processed and the task attribute information corresponding to each subtask. The filtering module is used to filter at least one target agent that matches each of the sub-tasks to be processed, based on the task attribute information corresponding to the sub-tasks to be processed. The determining module is used to determine available intelligent agents from the target intelligent agents for each of the sub-tasks to be processed; A module for assembling intelligent agents based on the available intelligent agents corresponding to each of the sub-tasks to be processed; An execution module is used to execute the task to be executed according to the intelligent agent group.

[0019] Thirdly, the present invention provides a computer device, comprising: a memory and a processor, the memory and the processor being communicatively connected to each other, the memory storing computer instructions, and the processor executing the computer instructions to perform the task processing method described in the first aspect or any corresponding embodiment thereof.

[0020] Fourthly, the present invention provides a computer-readable storage medium storing computer instructions for causing a computer to perform the task processing method described in the first aspect or any corresponding embodiment thereof.

[0021] Fifthly, the present invention provides a computer program product, including computer instructions for causing a computer to execute the task processing method described in the first aspect or any corresponding embodiment thereof. Attached Figure Description

[0022] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0023] Figure 1 This is a flowchart illustrating a task processing method according to an embodiment of the present invention; Figure 2 This is a flowchart illustrating another task processing method according to an embodiment of the present invention; Figure 3 This is a flowchart illustrating another task processing method according to an embodiment of the present invention; Figure 4 This is a structural block diagram of a task processing device according to an embodiment of the present invention; Figure 5 This is a schematic diagram of the hardware structure of a computer device according to an embodiment of the present invention. Detailed Implementation

[0024] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0025] The task processing method provided in this invention can be applied to multi-agent collaborative task processing systems in industrial internet scenarios.

[0026] In an Industrial Internet environment, multiple intelligent agents with varying functions are deployed. These agents, developed by different vendors, possess different capabilities and semantic systems, and can join or leave the system at any time based on actual needs. When an enterprise needs to complete a complex industrial task, such as energy consumption optimization of the dehydration unit in a natural gas purification plant, the system, after acquiring the task, breaks it down into multiple independently executable sub-tasks according to the task processing flow, and matches a suitable intelligent agent from the agent cluster for each sub-task. After confirming the availability of each intelligent agent, the system dynamically groups these agents into a temporary intelligent agent group. Each intelligent agent then collaborates according to the logical dependencies between the sub-tasks to jointly complete the complex industrial task.

[0027] Throughout the task execution process, the system monitors the operational status of each agent in real time. When a subtask fails, the system automatically obtains the incomplete progress information of that subtask, selects a replacement agent from the agent cluster to join the agent group, and resumes execution from the breakpoint to ensure that the overall task continues to advance until completion.

[0028] This application scenario is applicable to industrial fields such as petrochemicals, intelligent manufacturing, and smart energy. Through the task processing method of this invention, on-demand dynamic collaboration of intelligent agents can be achieved under a loosely coupled architecture, effectively dealing with complex situations such as diverse task types and frequent sudden tasks in industrial production.

[0029] According to an embodiment of the present invention, a task processing method embodiment is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0030] This embodiment provides a task processing method that can be used in the aforementioned mobile terminals, such as mobile phones and tablet computers. Figure 1 This is a flowchart of a task processing method according to an embodiment of the present invention, such as... Figure 1 As shown, the process includes the following steps: Step S101: The task to be executed is split into multiple sub-tasks to be processed and the task attribute information corresponding to each sub-task to be processed.

[0031] The task to be executed is a complete business objective that requires the collaborative effort of at least two available intelligent agents. The process of breaking down the task to be executed involves decomposing it into several relatively independent subtasks that can be executed in parallel or sequentially, according to its execution logic, functional modules, or business processes. The task attribute information of each subtask includes relevant descriptive data, including but not limited to the task objective, the types of resources required for execution, input / output specifications, quality requirements, and dependencies with other subtasks. Breaking down the task to be executed facilitates matching suitable intelligent agents to each subtask. In this embodiment of the invention, the task to be executed can be an industrial task.

[0032] Step S102: Based on the task attribute information corresponding to the sub-task to be processed, at least one target agent that matches the sub-task to be processed is selected.

[0033] In this embodiment of the invention, intelligent agents that meet certain conditions are selected from an intelligent agent cluster according to certain rules. An intelligent agent is a software entity or a combination of software and hardware that possesses autonomous decision-making capabilities, can perceive the environment, and execute tasks. It can be a large-model-driven AI program, an automated control program, or other computing units with task execution capabilities. An intelligent agent set is a collection of multiple intelligent agents, each of which can dynamically register, join, or leave the Industrial Internet. A target intelligent agent refers to an intelligent agent that, after screening, is initially considered to possess the capabilities and conditions required to execute the sub-task. A sub-task may select multiple target intelligent agents for further determination of the optimal choice.

[0034] Step S103: For each of the sub-tasks to be processed, determine the available intelligent agents from each of the target intelligent agents.

[0035] Available agents are those among the target agents that, after further verification, are capable of participating in the execution of the sub-task. This verification process can consider the agent's own state, unlike preliminary matching based solely on capability information. Although a target agent may meet the capability requirements, its actual availability still needs to be confirmed based on its actual operating state. Typically, the number of available agents is less than the number of target agents, and the number of target agents is less than the group of agents.

[0036] In practical applications, the selection of available agents can usually take into account the constraints of the subtask to be processed, the limitations of the agent itself, and domain restrictions.

[0037] Step S104: Based on the matching relationship between the available agents and the sub-tasks to be processed, generate an agent group corresponding to the task to be executed.

[0038] A matching relationship represents the correspondence between available agents and subtasks to be processed, that is, which available agent is responsible for executing which subtask. An agent group refers to a collection of agents temporarily organized for a single task. The available agents within an agent group work collaboratively according to the logical relationships between the subtasks, which can be sequential, parallel, or dependent execution relationships, to jointly complete the task. Agent groups are dynamically generated; they can be established when a task is created and disbanded when the task is completed.

[0039] Step S105: Execute the task to be executed according to the intelligent agent group.

[0040] Each available agent in the control agent group, according to its own responsibilities and the logical relationship between the sub-tasks to be processed, will work together to execute the corresponding sub-tasks to be processed, and finally complete the task to be executed.

[0041] The task processing method provided in this embodiment achieves precise matching of sub-tasks by screening target intelligent agents for each sub-task, thereby improving the adaptation accuracy between tasks and intelligent agents. By determining available intelligent agents from each target intelligent agent, it ensures that the finally determined intelligent agents have actual execution capabilities. By generating intelligent agent groups based on matching relationships, it achieves dynamic team formation on demand, enabling the grouping structure to be flexibly adjusted according to task changes. Finally, the intelligent agent groups collaboratively execute the task to be executed. This application can dynamically organize available intelligent agents according to task requirements, avoiding the inefficiency caused by the forced requisition of available intelligent agents, and improving the grouping flexibility and task execution efficiency.

[0042] This embodiment provides a task processing method that can be used in the aforementioned mobile terminals, such as mobile phones and tablet computers. Figure 2 This is a flowchart of a task processing method according to an embodiment of the present invention, such as... Figure 2 As shown, the process includes the following steps: Step S201: The task to be executed is split into multiple subtasks to be processed and the task attribute information corresponding to each subtask. For details, please refer to [link to relevant documentation]. Figure 1 Step S101 of the illustrated embodiment will not be described again here.

[0043] Step S202: Based on the task attribute information corresponding to the sub-task to be processed, filter the target intelligent agent that matches the sub-task to be processed.

[0044] Specifically, step S202 includes: Step S2021: Based on the task attribute information corresponding to the subtask to be processed, generate the task vector corresponding to the subtask to be processed.

[0045] A task vector is a high-dimensional numerical sequence obtained by numerically encoding the task attribute information of the subtasks to be processed. Task information described in natural language cannot be directly processed by computers. Encoding transforms the semantic information of the task, such as its goals, requirements, and constraints, into numerical form, enabling computers to perform mathematical operations and comparisons.

[0046] Step S2022: Generate the agent vector corresponding to each agent based on the ability attribute information of each agent.

[0047] Capability attribute information describes the agent's functions, resources, constraints, and states. An agent vector is a high-dimensional numerical sequence obtained by numerically encoding the agent's capability attribute information; it resides in the same vector space as the task vector, facilitating mathematical operations between the two.

[0048] Step S2023: Calculate the vector similarity between the vectors based on the task vector and each of the agent vectors.

[0049] Vector similarity is a numerical metric that measures how close a task vector and an agent vector are in vector space. A higher vector similarity indicates that the agent is semantically more compatible with the requirements of the subtask. Vector similarity can be calculated using mathematical methods such as hyperbolic distance and cosine similarity.

[0050] In one optional implementation, the task semantics of the subtask to be processed are pre-trained using a masked language model. The textual features in the task semantics are modeled using an MLM language model, and the numerical features are normalized and concatenated to output an initial semantic vector of dimension m for the subtask. The agent's semantics are encoded using multimodal fusion, also outputting an initial semantic vector of dimension m for the agent. Then, cross-domain bidirectional attention alignment training is used to obtain a task-aware agent semantic vector and an agent-aware task semantic vector. These two vectors are then dynamically fused using a learnable gating unit, outputting a task-fusion semantic vector based on the subtask and an agent-fusion semantic vector of dimension m, respectively. Finally, the task-fusion semantic vector based on the subtask and the agent-fusion semantic vector of the agent are projected onto a hyperbolic space, outputting a task vector and an agent vector of dimension d, respectively. In this embodiment, m = 768 and d = 128.

[0051] The formula for calculating vector similarity is:

[0052] Where x is a d-dimensional task vector and y is a d-dimensional agent vector. Let the magnitude of the task vector be _____. Let be the magnitude of the agent vector, and arcosh be the inverse hyperbolic cosine function.

[0053] Step S2024: Based on vector similarity and a pre-set similarity threshold, at least one candidate similarity is obtained.

[0054] The similarity threshold is a preset numerical threshold used to determine whether the similarity between the agent and the subtask reaches an acceptable minimum standard. Candidate similarities are the similarity values ​​among all vector similarities that are greater than this threshold. By using the preset similarity threshold, agents with significantly insufficient matching are quickly filtered out. In this embodiment of the invention, the similarity threshold can be set to 0.5, 0.7, or 0.8, etc.

[0055] Step S2025: Select the target agent that matches the subtask to be processed from the agents corresponding to each candidate similarity.

[0056] From the set of candidate agents that have passed the initial screening, further refinement is carried out through other conditions (such as constraint evaluation, comprehensive scoring, etc.) to finally determine the target agent that matches the subtask.

[0057] By mapping subtasks and agents to a shared semantic space for quantitative matching, the semantic system differences between agents from different vendors are eliminated, enabling heterogeneous agents to be accurately matched in a unified numerical space. This solves the problem of semantic interoperability barriers between heterogeneous agents in a loosely coupled architecture.

[0058] In some optional implementations, step S2025 above includes: Step a1: Based on the first constraint, perform screening to obtain candidate agents. The first constraint is used to characterize the conditions of the sub-task to be processed that the agent must meet.

[0059] Step a2: Based on the second constraint, the first constraint, and the vector similarity, calculate the constraint value corresponding to the candidate agent. The second constraint is used to characterize the condition that the trust value can be improved by solving the sub-task to be processed based on the target agent.

[0060] In this embodiment of the invention, both the first and second constraints are attribute information of the subtask to be processed, which can usually be determined during task decomposition. In some optional ways, the first constraint can be a mandatory numerical boundary set for execution parameters, output results, or running status during the subtask execution process. The second constraint can be an expected condition with a flexible space during the subtask execution process. The flexible space represents that the condition has a target value and an acceptable baseline value; the further the deviation from the target value, the lower the satisfaction score. Optionally, the satisfaction of the first constraint is represented by a 0 / 1 indicator function; if not satisfied, the score is 0. The satisfaction of the second constraint is continuously scored within the range of 0 to 1 based on the deviation between the actual value and the target value. Only when the target agent's behavior output continuously satisfies the second constraint over multiple consecutive evaluation periods is the agent deemed trustworthy, thus triggering a positive update of the trust value. Conversely, if the agent's behavior frequently deviates from the expected range of the second constraint, even if the first constraint is fulfilled, its behavior will be judged to have a high risk of disturbance, and the trust value will not be increased or may even be reduced. Trust value, as one of the attribute information of an agent, constitutes the semantic description of the agent.

[0061] Therefore, the essence of the second constraint is to change the focus of trust value enhancement from simply whether the task is completed to whether the completion process is smooth, collaborative, and sustainable, thereby guiding the agent to actively restrain the negative impact on the physical system while pursuing task achievement.

[0062] Based on the degree of satisfaction of the second constraint and the first constraint, calculate the satisfaction score of the candidate agent; The satisfaction score is a comprehensive score obtained by quantifying and scoring the candidate agent's satisfaction with various constraints. For the first constraint, satisfaction is awarded full marks, and non-satisfaction results in direct elimination. For the second constraint, scores are awarded linearly or non-linearly based on the actual degree of satisfaction, with full marks for complete satisfaction, zero marks for complete non-satisfaction, and proportional scores for partial satisfaction.

[0063] In one alternative implementation, the formula for calculating the satisfaction score can be expressed as: , Where k is the sequence number of the first constraint condition corresponding to the subtask during task decomposition, K is the total number of first constraints corresponding to the subtask, l is the sequence number of the second constraint condition corresponding to the subtask during task decomposition, and L is the total number of second constraints corresponding to the subtask. This represents the k-th first constraint condition corresponding to the i-th subtask. This is the l-th second constraint corresponding to the i-th subtask. This indicates the degree to which the first constraint condition is satisfied between the j-th agent and the i-th subtask. This indicates the degree to which the second constraint condition is satisfied for the j-th agent and the i-th subtask. This represents the weight of the k-th first constraint. Let f represent the weight of the l-th second constraint, and f be an indicator function, i.e., 1 if the agent satisfies the condition, and 0 otherwise. =0.5.

[0064] The constraint values ​​for the candidate agents are calculated based on the satisfaction score, vector similarity, and corresponding weights.

[0065] Weights are coefficients assigned to semantic similarity and constraint satisfaction, used to adjust the importance of each dimension in the final score. The constraint value, or comprehensive score, is the weighted sum of the satisfaction score and vector similarity according to their respective weights, yielding the final comprehensive score for each candidate agent. For example, if the semantic similarity weight is 0.7 and the constraint satisfaction weight is 0.3, then the comprehensive score = 0.7 × similarity + 0.3 × satisfaction score.

[0066] In one alternative implementation, the formula for calculating the constraint value can be expressed as: , in, This represents the i-th subtask. Let j represent the j-th agent. Let represent the vector similarity between the i-th subtask and the j-th agent. This represents the satisfaction score between the j-th agent and the i-th subtask. and These represent the proportions of vector similarity distance and constraint satisfaction, respectively. General settings .

[0067] Step a3: The candidate agent with the highest constraint value is determined as the target agent.

[0068] The first constraint is a mandatory condition that must be unconditionally met in the subtask, such as a safety threshold (temperature must not exceed 204℃), physical limits (pressure must be within the range of 2.4-5MPa), and regulatory requirements. The first constraint adopts a veto mechanism; if a candidate agent fails to meet any of the first constraints, it will be eliminated and will not be included in the subsequent selection. Only agents that meet all the first constraints can be retained as candidate agents.

[0069] The second constraint is a quantifiable and evaluable condition that allows for partial non-compliance, such as an energy consumption reduction target (target reduction of 10%, minimum acceptable 8%) or response time requirements. The constraint value is a quantitative score that comprehensively reflects the agent's degree of satisfaction with the soft first constraint. Specifically, the satisfaction of the first constraint is represented by a 0 / 1 indicator function, i.e., 1 for satisfying the first constraint and 0 for not satisfying it. The degree of satisfaction of the second constraint is represented by a weighted scoring method, i.e., full marks for fully satisfying the second constraint, proportional marks for partially satisfying the second constraint, and 0 marks for not satisfying the second constraint at all. The weighted sum of the two results in the constraint value.

[0070] Among all candidate agents, the higher the constraint value, the better the agent can satisfy the second constraint while satisfying the first constraint. Therefore, the agent with the highest constraint value is selected as the final target agent for the subtask to be processed.

[0071] By combining the veto mechanism of the first constraint with the weighted scoring of the second constraint, further constraint screening is performed based on semantic matching. This ensures that the hard bottom line is not breached while allowing for optimal selection based on soft indicators, so that the agent's matching results simultaneously satisfy semantic relevance and constraint feasibility, thus solving the problem of the one-sidedness of relying solely on semantic matching or rule matching.

[0072] Step S203: For each of the sub-tasks to be processed, determine the available intelligent agents from each of the target intelligent agents. See details below. Figure 1 Step S103 of the illustrated embodiment will not be described again here.

[0073] Step S204: Based on the matching relationship between the available agents and the sub-tasks to be processed, generate an agent group corresponding to the task to be executed. See details below. Figure 1 Step S104 of the illustrated embodiment will not be described again here.

[0074] Step S205: Execute the task to be performed according to the intelligent agent group. See details below. Figure 1 Step S105 of the illustrated embodiment will not be described again here.

[0075] The task processing method provided in this embodiment organically combines semantic similarity and constraint satisfaction through weighted fusion to form a comprehensive score. In practical applications, the weight ratio of the two dimensions can be flexibly adjusted according to different scenarios to achieve an optimal balance between semantic relevance and constraint feasibility, further improving the accuracy and adaptability of agent selection.

[0076] This embodiment provides a task processing method that can be used in the aforementioned mobile terminals, such as mobile phones and tablet computers. Figure 3 This is a flowchart of a task processing method according to an embodiment of the present invention, such as... Figure 3 As shown, the process includes the following steps: Step S301: The task to be executed is split into multiple sub-tasks to be processed and the task attribute information corresponding to each sub-task to be processed.

[0077] Step S302: Based on the task attribute information corresponding to the subtask to be processed, at least one target agent matching the subtask to be processed is selected. For details, please refer to [link to relevant documentation]. Figure 2 Step S202 of the illustrated embodiment will not be described again here.

[0078] Step S303: For each of the sub-tasks to be processed, determine the available intelligent agents from each of the target intelligent agents.

[0079] Specifically, step S303 includes: Step S3031: Send a grouping invitation to each of the target intelligent agents.

[0080] A grouping invitation is an invitation sent to selected target agents to inquire whether they are willing to join a group to perform a corresponding subtask. The invitation can include information such as task description, role responsibilities, and expected work time to allow the agent to make a thorough evaluation. In practice, a grouping invitation is valid for 120 seconds; failure to respond within this timeframe indicates automatic rejection.

[0081] Step S3032: Receive response information returned by each target intelligent agent based on its own state information.

[0082] The agent's own state information includes its current load rate (the ratio of occupied resources to total resources), available resources (remaining available computing power, memory, etc.), and health status (whether it is operating normally). The agent's response information is the reply returned by the agent after receiving a grouping invitation, based on its own state, to determine whether it can accept the task. The reply includes acceptance or rejection.

[0083] Step S3033: Based on the response information, the target agent that accepts the grouping invitation is determined as the available agent.

[0084] Based on the returned response information, only agents that accept the invitation are identified as available agents. For agents that reject the invitation, an invitation can be sent to the next target agent in the sorted order until an agent that can accept the task is found.

[0085] Step S304: Based on the matching relationship between the available agents and the sub-tasks to be processed, generate an agent group corresponding to the task to be executed.

[0086] Step S305: Execute the task to be executed according to the intelligent agent group.

[0087] Specifically, step S305 includes: Step S3051: Based on the matching relationship between the available intelligent agent and the sub-task to be processed, a grouping contract is generated; the grouping contract is used to constrain the input data format, output data format and quality requirements of the available intelligent agent in executing the sub-task.

[0088] In this embodiment of the invention, the grouping contract is a binding agreement formed between the available intelligent agent and the subtask to be processed after the agent accepts the invitation. It specifies and confirms the roles and responsibilities within the grouping invitation. The input data format refers to the organizational structure and coding standards that the data received by the agent to execute the subtask should follow, such as CSV format, JSON structure, sampling frequency, etc. The output data format refers to the organizational structure and coding standards that the result data returned by the agent after completing the task should follow. Quality requirements refer to the quantitative indicators that the agent's execution results need to achieve, such as execution accuracy, error range, completion time, etc.

[0089] Step S3052: Based on the resource capabilities of each available agent and the requirements of the subtask, determine the available resources allocated to each available agent in the agent group.

[0090] Resource capacity refers to the computing resources that an agent can access, including the number of CPU cores, GPU memory, RAM capacity, and network bandwidth. Available resources are the actual resource quota allocated to the current subtask from the agent's total resources. This allocation is based on the resource requirements of the subtasks and the resource capabilities of each agent.

[0091] Step S3053: Control each available intelligent agent to execute the corresponding sub-task based on the grouping contract and available resources.

[0092] Within the allocated available resource quota, each agent executes its corresponding sub-task according to the input data format, output data format, and quality requirements agreed upon in the grouping contract, and outputs the execution results as required.

[0093] Step S3054: Monitor the execution process and obtain the monitoring results.

[0094] Monitoring involves continuously collecting operational status information of intelligent agents during task execution. Monitoring results include information such as task execution progress, resource usage, output quality data, and whether any anomalies have occurred. Data is typically collected periodically according to a preset time window. Optionally, the preset time window can be set according to the actual application scenario; it can be set to 1-5 seconds, 10-60 seconds, or 1-5 minutes, depending on the frequency of parameter changes during execution.

[0095] The task processing method provided in this embodiment monitors the execution of sub-tasks according to the grouping contract, ensuring that the intelligent agent group has clear execution specifications, sufficient resource guarantees, and real-time status feedback during task execution. The grouping contract standardizes and regulates the execution behavior of intelligent agents, making the execution results predictable, measurable, and traceable, significantly improving the reliability and controllability of multi-agent collaborative execution.

[0096] Based on the above embodiments, if the monitoring result indicates execution failure, the method further includes: Step S3055: Obtain the task attribute information and incomplete progress information corresponding to the failed subtask.

[0097] In this embodiment of the invention, execution failure indicates that the subtask failed to complete as expected. Possible reasons include timeout, substandard quality, agent failure, resource exhaustion, or being preempted by a higher-priority task. The incomplete progress information records the amount of work already completed by the current subtask, such as the amount of data processed, completed execution phases, and remaining steps to be executed.

[0098] Step S3056: Based on the task attribute information and the incomplete progress information, re-select alternative agents that match the sub-task from the target agent set.

[0099] Alternate agents are used to replace agents that fail and to take over the remaining work. During the re-selection process, the original requirements of the subtasks (task attribute information) and the amount of work yet to be completed (unfinished progress information) are considered together, and agents are re-matched from the agent cluster. Alternate agents can come from within the original agent group, such as other agents that have completed their tasks, or they can come from outside the agent group.

[0100] Step S3057: Send a grouping invitation to the alternative agent and add the alternative agent that accepts the invitation to the agent group to continue executing the sub-task.

[0101] The grouping invitation process is executed again for the newly selected replacement agents. After they accept, they are added to the original agent group. The agents that fail to replace them continue to execute subtasks from the point where the progress is not completed.

[0102] The task processing method provided in this embodiment, through a failure recovery mechanism, enables the multi-agent system to have automatic fault tolerance and self-recovery capabilities when facing agent execution failures. It retains completed progress data, avoids repetitive work, significantly improves the completion rate of multi-agent collaborative processing of pending tasks, and effectively supports the high requirements for system reliability in industrial scenarios.

[0103] In some optional implementations, step S3056 above includes: Step c1: Obtain the incomplete progress information of the subtask.

[0104] Obtain progress data for failed subtasks, including the percentage of work completed, the dataset processed, and the specific details of completed computational phases.

[0105] Step c2: Based on the incomplete progress information, determine the remaining content to be executed for the subtask.

[0106] The remaining tasks are the unfinished portions of the total workload of a subtask, such as unprocessed batches of data, parameter combinations to be calculated, and quality metrics to be verified. The remaining tasks are the original subtask requirements minus the progress already completed.

[0107] Step c3: Select agents from the target agent set that match the remaining content to be executed, and use them as the alternative agents.

[0108] The matching of alternative agents is not based on the requirements of the original complete subtask, but rather on the requirements of the remaining content to be executed. Since the remaining content is usually simpler or smaller in scope than the original task, more candidate agents can be matched, and the matching accuracy is higher.

[0109] Improving the accuracy of substitute matching avoids resource waste and expands the range of available agents, thereby increasing the success rate of substitutes. Compared to matching methods based on the original task, matching based on remaining content can more accurately match actual needs.

[0110] This embodiment also provides a task processing device for implementing the above embodiments and preferred embodiments; details already described will not be repeated. As used below, the term "module" can refer to a combination of software and / or hardware that implements a predetermined function. Although the device described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0111] This embodiment describes the invention in detail using the intelligent control of a triethylene glycol dehydration unit in a natural gas purification plant in the oil and gas field industry as an application scenario.

[0112] The natural gas purification plant has deployed an industry-specific large-scale model as a task scheduling center. The Internet cloud platform has registered eight types of intelligent agents, including control intelligent agents, water dew point prediction intelligent agents, energy consumption optimization intelligent agents, quality intelligent agents, safety intelligent agents, data processing intelligent agents, monitoring intelligent agents, and human-computer interaction intelligent agents.

[0113] The task description obtained by the task scheduling center is: "Optimize the energy consumption of the dehydration unit, and ensure that the natural gas outlet water dew point quality meets the standards and the downstream pipeline natural gas demand is stable." According to the execution logic, the task to be executed is divided into 5 sub-tasks to be processed: real-time data acquisition and processing (v1), current operating condition baseline assessment (v2), process parameter optimization calculation (v3), quality constraint verification (v4), and control command issuance (v5).

[0114] The dependencies between the subtasks to be processed are: v1→v2→v3→v4→v5, v2→v3 parallel branch, with corresponding dependency weights of {0.8, 1, 0.6, 1, 1}.

[0115] The constraints for each subtask to be processed include a first constraint and a second constraint. The first constraint is the boundary condition that must be met: outlet water dew point ≤ -20℃, TEG lean liquor concentration ≥ 99.5%, absorber operating pressure ∈ [2.4, 5] MPa, and reboiler temperature ≤ 204℃. The second constraint is the expected condition that is allowed to be partially unmet: target energy consumption reduction ≥ 10% (minimum acceptable 8%), and process parameter adjustment frequency ≤ 1 time / hour.

[0116] This embodiment uses a hyperbolic space as the shared semantic space and adds specialized terms such as dew point and TEG to the vocabulary. An encoder-only network structure is used, with an initial dimension m of 768 and an output dimension d of 128. Training configuration: batch_size=32, lr=2e-5, epoch=30.

[0117] The process of agent matching is illustrated using the subtask v3 "process parameter optimization calculation" as an example.

[0118] A task vector is generated based on the task attribute information of the subtask v3 to be processed. The task attribute information is then concatenated and input into the hybrid attention mechanism network. t_text = "Multi-objective optimization of TEG circulation rate, reboiler temperature, and absorber pressure combination"; t_sla = "Optimization computation time ≤ 5min, number of Pareto solutions ≥ 10, constraint violation rate = 0"; t_const = "TEG∈[2.5, 4.5], Temp∈[190, 204], Press∈[2.4, 5]"; After encoding and projection, a 128-dimensional task vector is obtained.

[0119] Generate an agent vector based on the capability attribute information of the energy-optimizing agent. Then, concatenate the function dimension, resource dimension, and constraint dimension and input them into a hybrid attention mechanism network. a_func = “TEG energy consumption mechanism modeling + NSGA-III multi-objective optimization + process parameter recommendation”; a_resource = “GPU≥4GB, CPU≥4 cores, sampling frequency 1Hz”; a_qual = "The average error of 50 historical optimizations is 2.8%, and the recommended executability rate is 96%"; After encoding and projection, a 128-dimensional intelligent agent vector is obtained.

[0120] Calculate the vector similarity between the task vector and the agent vector. In this embodiment, the vector similarity between the subtask v3 to be processed and the energy consumption optimization agent a1 is 0.87, indicating a high semantic matching degree; while the vector similarity between the subtask v3 to be processed and the control agent is only 0.34, indicating a low matching degree.

[0121] The calculated similarity of each vector is compared with a preset similarity threshold (set to 0.5 in this embodiment), and candidate similarities greater than the threshold are retained to obtain a set of candidate agents that match each subtask to be processed.

[0122] In the candidate agent set, further screening is performed based on constraints. In this embodiment, both first constraints (steady-state requirements and parameter safety range) of the subtask v3 to be processed are satisfied, and are recorded as 1.0; the second constraint "provide 4 sets of alternative optimization schemes" is not satisfied, and a score of 0.5 is deducted according to the weight. According to the constraint satisfaction score formula, the constraint satisfaction score Sat_constraint(v3, a1) = (1.0 + 1.0) / 2 + 0.5 × 0.5 / 1 = 1.25, which is 0.83 after normalization. The vector similarity and constraint satisfaction score are weighted and summed according to their respective weights (the semantic similarity weight is 0.7 and the constraint satisfaction weight is 0.3 in this embodiment) to obtain the comprehensive matching score: Score(v3, a1) = 0.7 × 0.87 + 0.3 × 0.83 = 0.858.

[0123] The candidate agent with the highest overall matching score is selected as the target agent for subtask v3 to be processed.

[0124] Once an available agent is identified, a grouping invitation is sent to the energy optimization agent. The content of the grouping invitation is as follows: G_id: TEG-OPT-20260526-1430-0001-0017 (Encoding meaning: TEG device-optimization task-date-time-task priority-sequence number).

[0125] G_goal: The list of subtasks to be completed is [v1, v2, v3, v4, v5]; the DAG of subtasks is v1→v2→v3→v4→v5, with parallel branches of v1→v3; the overall constraints include: total time ≤ 30 minutes, energy consumption optimization margin ≥ 10% (minimum acceptable 8%), outlet water dew point ≤ -20℃, downstream gas supply ≥ 80×10 4 m³ / d.

[0126] Role_spec: In the task allocation matrix R, R[v3]=1 (primarily responsible), R[v2]=0.3 (participates in the calculation, weight 0.3); the collaboration relationship is to receive standardized data from the data processing agent, push optimized parameter combinations to the prediction agent, and submit optimization schemes to the human-computer interaction agent for review (asynchronous, response timeout of 5 minutes); the permission boundaries are read-only access to SCADA historical data, parameter recommendations are written to the pending approval queue without direct issuance permission, and the GPU resource quota limit is 4GB.

[0127] Exit_rule: Upon normal completion, outputs 3 sets of Pareto optimal parameter combinations and a quality assessment report, then releases all resources; for negotiated exit, request 10 minutes in advance and exit after completing the current calculation batch; for default exit, the task completion quality is below 70%, resulting in forced exit. The invitation validity period (TTL) is 120 seconds; failure to respond within the time limit will automatically reject the request.

[0128] Upon receiving a grouping invitation, the energy optimization agent autonomously evaluates its own status information (including current load rate, task list, etc.). If it confirms acceptance of the sub-task, it returns an acceptance response and generates a grouping contract. Based on the grouping invitation, the grouping contract defines the agent's role and responsibilities as process parameter optimization calculations. The input data format is standardized CSV data, and the output data format consists of three Pareto solution sets and a quality assessment report. The quality requirement is an energy consumption optimization margin of ≥8%.

[0129] Based on the resource capabilities of the energy optimization agent and the requirements of the sub-tasks, the available resources are determined and allocated. The energy optimization agent is controlled to execute the corresponding sub-tasks based on the marshalling contract and available resources, and the execution process is monitored.

[0130] During execution, an increase in upstream raw gas flow rate was detected, from 350,000 cubic meters / day to 520,000 cubic meters / day, an increase of approximately 20%, indicating a change in upstream operating conditions. The quality deviations of each agent are as follows: the water dew point prediction agent has a quality deviation of Δq = 4.5℃, the energy consumption optimization agent has a quality deviation of Δq = 43kW, and the quality of other agents is normal. The water dew point quality deviation exceeds the set threshold of 3℃, triggering a grouping adjustment.

[0131] The capability attribute information of the dew point prediction agent was updated, and the matching degree was recalculated. After the update, the matching degree dropped to 0.72. The cost-benefit ratio of the adjustment methods was further evaluated: the cost-benefit ratio of micro-adjustment (adding resources) was 1.25, and the cost-benefit ratio of large adjustment (agent entering or leaving the group) was 0.87. Adding resources was selected, the capability profile of the dew point prediction agent was updated, and semantic matching was re-performed. The matching degree recovered to above 0.82.

[0132] Once all subtasks have met their objectives, an optimization report is generated, triggering the decompilation conditions. Resources within the group are released in an orderly manner according to priority, and the real-time status information of each agent is updated. Each agent then continues to execute the next task based on its current status and priority in the task pool.

[0133] This embodiment provides a task processing device, such as Figure 4 As shown, it includes: The splitting module 401 is used to split the task to be executed to obtain multiple sub-tasks to be processed and task attribute information corresponding to each sub-task to be processed. The filtering module 402 is used to filter at least one target agent that matches the subtask to be processed based on the task attribute information corresponding to the subtask to be processed. The determining module 403 is used to determine available intelligent agents from the target intelligent agents for each of the sub-tasks to be processed; The generation module 404 is used to generate a group of agents corresponding to the task to be executed based on the matching relationship between the available agents and the sub-task to be processed; The execution module 405 is used to execute the task to be executed according to the intelligent agent group.

[0134] In some alternative implementations, the filtering module 402 includes: The first generation unit is used to generate a task vector corresponding to the subtask to be processed based on the task attribute information corresponding to the subtask to be processed. The second generation unit is used to generate the vector of each agent based on the ability attribute information of each agent. A computing unit is configured to calculate the vector similarity between the vectors based on the task vector and each of the agent vectors; A comparison unit is used to obtain at least one candidate similarity based on vector similarity and a pre-set similarity threshold; The filtering unit is used to filter out the target agent that matches the subtask to be processed from the agents corresponding to each candidate similarity.

[0135] In some optional implementations, the filtering unit includes: The sub-units are obtained and used to filter candidates based on the first constraint, whereby the first constraint is used to characterize the conditions of the sub-tasks to be processed that the agent must satisfy. The computational subunit is used to calculate the constraint value corresponding to the candidate agent based on the second constraint, the first constraint, and the vector similarity. The second constraint is used to characterize the condition that the trust value can be improved by solving the sub-task to be processed based on the target agent. The first determining subunit is used to determine the candidate agent with the highest constraint value as the target agent.

[0136] In some optional implementations, the computational subunit is further specifically used to calculate the satisfaction score of the candidate agent based on the degree of satisfaction of the second constraint and the first constraint; and to calculate the constraint value corresponding to the candidate agent according to the satisfaction score, vector similarity and corresponding weight.

[0137] In some alternative implementations, the determining module 403 includes: The sending unit is used to send grouping invitations to each of the target intelligent agents; A receiving unit is used to receive response information returned by each of the target intelligent agents based on their own state information; The grouping unit is used to determine the target agent that accepts the grouping invitation as the available agent based on the response information.

[0138] In some alternative implementations, execution module 405 includes: The third generation unit is used to generate a grouping contract based on the matching relationship between the available intelligent agent and the sub-task to be processed; the grouping contract is used to constrain the input data format, output data format and quality requirements of the available intelligent agent in executing the sub-task; The determining unit is configured to determine the available resources allocated to each available agent in the agent group based on the resource capabilities of each available agent and the requirements of the sub-tasks. A control unit is used to control each available intelligent agent to execute the corresponding sub-task based on the grouping contract and available resources; The monitoring unit is used to monitor the execution process and obtain monitoring results.

[0139] In some optional implementations, the execution module 405 further includes: The acquisition unit is used to acquire the task attribute information and incomplete progress information corresponding to the failed subtask when the monitoring result is execution failure; The re-screening unit is used to re-screen alternative agents that match the subtask from the target agent set based on the task attribute information and the incomplete progress information. The execution unit is configured to send a grouping invitation to the alternative agent and add the alternative agent that accepts the invitation to the agent group to continue executing the subtask.

[0140] In some alternative implementations, the rescreening unit includes: The acquisition subunit is used to acquire the incomplete progress information of the subtask; The second determining subunit is used to determine the remaining content to be executed in the subtask based on the incomplete progress information; As a sub-unit, it is used to select agents from the target agent set that match the remaining content to be executed, and use them as the alternative agents.

[0141] Further functional descriptions of the above modules and units are the same as those in the corresponding embodiments described above, and will not be repeated here.

[0142] In this embodiment, the task processing device is presented in the form of a functional unit. Here, a unit refers to an ASIC (Application Specific Integrated Circuit) circuit, a processor and memory that execute one or more software or fixed programs, and / or other devices that can provide the above functions.

[0143] This invention also provides a computer device having the above-described features. Figure 4 The task processing device shown.

[0144] Please see Figure 5 , Figure 5 This is a schematic diagram of the structure of a computer device provided in an optional embodiment of the present invention, such as... Figure 5As shown, the computer device includes one or more processors 10, memory 20, and interfaces for connecting the components, including high-speed interfaces and low-speed interfaces. The components communicate with each other via different buses and can be mounted on a common motherboard or otherwise installed as needed. The processors can process instructions executed within the computer device, including instructions stored in or on memory to display graphical information of a GUI on external input / output devices (such as display devices coupled to the interfaces). In some alternative implementations, multiple processors and / or multiple buses can be used with multiple memories and multiple memory modules, if desired. Similarly, multiple computer devices can be connected, each providing some of the necessary operations (e.g., as a server array, a group of blade servers, or a multiprocessor system).

[0145] Processor 10 may be a central processing unit, a network processor, or a combination thereof. Processor 10 may further include a hardware chip. The hardware chip may be an application-specific integrated circuit (ASIC), a programmable logic device (PLD), or a combination thereof. The programmable logic device may be a complex programmable logic device (CAMP), a field-programmable gate array (FPGA), a general-purpose array logic (GDA), or any combination thereof.

[0146] The memory 20 stores instructions executable by at least one processor 10 to cause the at least one processor 10 to perform the method shown in the above embodiments.

[0147] The memory 20 may include a program storage area and a data storage area. The program storage area may store the operating system and applications required for at least one function; the data storage area may store data created based on the use of the computer device. Furthermore, the memory 20 may include high-speed random access memory and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some alternative embodiments, the memory 20 may optionally include memory remotely located relative to the processor 10, which can be connected to the computer device via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0148] The memory 20 may include volatile memory, such as random access memory; the memory may also include non-volatile memory, such as flash memory, hard disk or solid-state drive; the memory 20 may also include a combination of the above types of memory.

[0149] The computer device also includes an input device 30 and an output device 40. The processor 10, memory 20, input device 30, and output device 40 can be connected via a bus or other means. Figure 5Taking the example of a connection between China and Israel via a bus.

[0150] Input device 30 can receive input numerical or character information, and generate key signal inputs related to user settings and function control of the computer device, such as a touchscreen, keypad, mouse, trackpad, touchpad, joystick, one or more mouse buttons, trackball, joystick, etc. Output device 40 may include display devices, auxiliary lighting devices (e.g., LEDs), and haptic feedback devices (e.g., vibration motors). The aforementioned display devices include, but are not limited to, liquid crystal displays, light-emitting diodes, displays, and plasma displays. In some alternative embodiments, the display device may be a touchscreen.

[0151] This invention also provides a computer-readable storage medium. The methods described above according to embodiments of the invention can be implemented in hardware or firmware, or implemented as computer code that can be recorded on a storage medium, or implemented as computer code downloaded via a network and originally stored on a remote storage medium or a non-transitory machine-readable storage medium and then stored on a local storage medium. Thus, the methods described herein can be processed by software stored on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. The storage medium can be a magnetic disk, optical disk, read-only memory, random access memory, flash memory, hard disk, or solid-state drive, etc.; further, the storage medium can also include combinations of the above types of memory. It is understood that computers, processors, microprocessor controllers, or programmable hardware include storage components capable of storing or receiving software or computer code, which, when accessed and executed by the computer, processor, or hardware, implements the methods shown in the above embodiments.

[0152] A portion of this invention can be applied as a computer program product, such as computer program instructions, which, when executed by a computer, can invoke or provide the methods and / or technical solutions according to the invention through the operation of the computer. Those skilled in the art will understand that the forms in which computer program instructions exist in a computer-readable medium include, but are not limited to, source files, executable files, installation package files, etc. Correspondingly, the ways in which computer program instructions are executed by a computer include, but are not limited to: the computer directly executing the instructions, or the computer compiling the instructions and then executing the corresponding compiled program, or the computer reading and executing the instructions, or the computer reading and installing the instructions and then executing the corresponding installed program. Here, the computer-readable medium can be any available computer-readable storage medium or communication medium accessible to a computer.

[0153] Although embodiments of the invention have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of the invention, and all such modifications and variations fall within the scope defined by the appended claims.

Claims

1. A task processing method, characterized in that, The method includes: The task to be executed is broken down to obtain multiple sub-tasks to be processed and the task attribute information corresponding to each sub-task; Based on the task attribute information corresponding to the subtask to be processed, at least one target agent that matches the subtask to be processed is selected; For each of the sub-tasks to be processed, a usable agent is determined from each of the target agents; Based on the matching relationship between the available agents and the sub-tasks to be processed, an agent group corresponding to the task to be executed is generated; The task to be performed is executed according to the group of intelligent agents.

2. The method according to claim 1, characterized in that, The step of filtering target agents matching the subtask to be processed based on the task attribute information corresponding to the subtask to be processed includes: Based on the task attribute information corresponding to the subtask to be processed, a task vector corresponding to the subtask to be processed is generated; Generate agent vectors for each agent based on their capability attribute information. Calculate the vector similarity between the vectors based on the task vector and each of the agent vectors; Based on vector similarity and a pre-set similarity threshold, at least one candidate similarity is obtained; From the agents corresponding to each candidate similarity, the target agent that matches the subtask to be processed is selected.

3. The method according to claim 2, characterized in that, The step of selecting the target agent that matches the subtask to be processed from the agents corresponding to each candidate similarity includes: Based on the first constraint, a selection process is performed to obtain candidate agents. The first constraint is used to characterize the conditions of the sub-task to be processed that the agent must satisfy. Based on the second constraint, the first constraint, and the vector similarity, the constraint values ​​corresponding to the candidate agents are calculated. The second constraint is used to characterize the conditions that can improve the trust value by solving the sub-task to be processed based on the target agent. The candidate agent with the highest constraint value is determined as the target agent.

4. The method according to claim 3, characterized in that, The step of calculating the constraint values ​​corresponding to the candidate agents based on the second constraint, the first constraint, and the vector similarity includes: Based on the degree of satisfaction of the second constraint and the first constraint, calculate the satisfaction score of the candidate agent; The constraint values ​​for the candidate agents are calculated based on the satisfaction score, vector similarity, and corresponding weights.

5. The method according to claim 1, characterized in that, For each of the subtasks to be processed, determining the available intelligent agents from each of the target intelligent agents includes: Send grouping invitations to each of the target intelligent agents; Receive response information returned by each of the target intelligent agents based on their own state information; Based on the response information, the target agent that accepts the grouping invitation is identified as the available agent.

6. The method according to claim 1, characterized in that, The step of executing the task to be executed according to the intelligent agent group includes: Based on the matching relationship between the available intelligent agent and the subtask to be processed, a marshalling contract is generated; the marshalling contract is used to constrain the input data format, output data format and quality requirements of the available intelligent agent when executing the subtask; Based on the resource capabilities of each available agent and the requirements of the sub-tasks, the available resources are allocated to each available agent in the agent group. Control each available intelligent agent to execute the corresponding sub-task based on the grouping contract and available resources; The execution process is monitored, and the monitoring results are obtained.

7. The method according to claim 6, characterized in that, If the monitoring result indicates execution failure, the method further includes: Obtain the task attribute information and incomplete progress information corresponding to the failed subtask; Based on the task attribute information and the incomplete progress information, alternative agents matching the subtask are re-selected from the target agent set; A grouping invitation is sent to the alternative agent, and the alternative agent that accepts the invitation is added to the agent group to continue executing the subtask.

8. A task processing device, characterized in that, The device includes: The splitting module is used to split the task to be executed into multiple subtasks to be processed and the task attribute information corresponding to each subtask. The filtering module is used to filter at least one target agent that matches each of the sub-tasks to be processed, based on the task attribute information corresponding to the sub-tasks to be processed. The determining module is used to determine available intelligent agents from the target intelligent agents for each of the sub-tasks to be processed; A module for assembling intelligent agents based on the available intelligent agents corresponding to each of the sub-tasks to be processed; An execution module is used to execute the task to be executed according to the intelligent agent group.

9. A computer device, characterized in that, include: A memory and a processor are communicatively connected, the memory stores computer instructions, and the processor executes the computer instructions to perform the task processing method of any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions for causing the computer to perform the task processing method according to any one of claims 1 to 7.

Images