Self-adaptive multi-Agent cooperation method, equipment and medium

Abstract

The invention discloses a self-adaptive multi-Agent cooperation method and device and a medium, and the method comprises the steps that a meta Agent receives a task input by a user, and analyzes the task to obtain a task decomposition tree; the task management module obtains a task decomposition tree to which the task management module belongs, determines a dependency relationship corresponding to the task according to the task decomposition tree, determines a task execution plan according to the dependency relationship, and allocates the task according to the task execution plan so as to allocate a plurality of sub-tasks to an expert Agent pool; and executing the distributed sub-tasks by a plurality of expert Agents in the expert Agent pool, and collecting and integrating execution results of the sub-tasks to determine a total task result. According to the method, various complex tasks can be efficiently processed, the continuous optimization and self-evolution capabilities are achieved, and the complex task processing capability and flexibility of an artificial intelligence system are remarkably improved.

Description

Technical Field

[0001] The present application relates to the field of natural language processing technology, and in particular to an adaptive multi-agent collaboration method, device and medium. Background Art

[0002] With the rapid development of artificial intelligence technology, large language models (LLMs) are gradually showing their unparalleled strength in natural language processing, deep knowledge understanding, and diversified task execution. These models are trained with massive amounts of data to generate coherent and logical texts, and even simulate human conversations and thinking to some extent. However, when faced with comprehensive tasks involving complex logic, extensive knowledge coverage, and multi-domain intersections, a single large language model often seems to be unable to cope with the situation and is unable to fully and accurately meet various needs. Summary of the invention

[0003] In order to solve the above problems, the present application proposes an adaptive multi-agent collaboration method, which is applied in an adaptive multi-agent collaboration system, wherein the system includes a meta-agent, an expert agent pool, and a task management module; the method includes: the meta-agent receives a task input by a user, and parses the task to obtain a task decomposition tree; the task management module obtains the corresponding task decomposition tree, determines the dependency relationship corresponding to the task according to the task decomposition tree, and determines a task execution plan according to the dependency relationship, so as to allocate the task according to the task execution plan, thereby allocating multiple subtasks to the expert agent pool; multiple expert agents in the expert agent pool execute the allocated multiple subtasks, collect and integrate the execution results of the multiple subtasks, so as to determine the total task result.

[0004] In one example, the task is parsed, and the method further includes: preprocessing the task to obtain a plurality of words; the preprocessing process includes word segmentation, stop word removal, and stem extraction; converting the plurality of words to obtain a plurality of word vectors, and determining key information, context information, and semantic relationships of the plurality of word vectors; determining a plurality of subtasks based on the key information, and determining dependency relationships between the plurality of subtasks based on the context information and the semantic relationships, so as to determine the task decomposition tree based on the dependency relationships.

[0005] In one example, determining a task execution plan based on the dependency relationship specifically includes: determining an execution order of the multiple subtasks based on the dependency relationship to determine a directed acyclic graph of task execution based on the execution order; determining an execution path of the multiple subtasks based on the directed acyclic graph to determine the task execution plan based on the execution path.

[0006] In one example, allocating the task according to the task execution plan specifically includes: determining agent types corresponding to a plurality of pre-set expert agents, and allocating the plurality of subtasks to expert agents of corresponding agent types according to the task execution plan, wherein the agent types include: domain expert agents, functional agents, and coordination agents.

[0007] In one example, the method further includes: determining the expert agent description information pre-placed in the meta-agent, and determining a knowledge base according to the expert agent description information; determining a dynamic matrix of the expert agent pool according to the knowledge base, so as to record the functions of the expert agent pool according to the dynamic matrix, wherein the functions include expertise, processing efficiency and historical performance; monitoring the running status of the expert agent pool through the meta-agent, and evaluating the expert agent pool according to the dynamic matrix, so as to reallocate the multiple subtasks according to the evaluation results.

[0008] In one example, the system also includes a knowledge sharing center; collecting and integrating the execution results of the multiple subtasks, specifically including: obtaining multiple execution results of the expert agent pool through the task management module, formatting and cleaning the multiple execution results to obtain an execution result package, and sending the execution result package to the knowledge sharing center.

[0009] In one example, the method further includes: determining a knowledge base and an experience pool of the knowledge sharing center, and integrating the multiple execution results through the knowledge base and the experience pool to determine the overall task result.

[0010] In one example, the method further includes: determining priorities corresponding to the plurality of subtasks and determining a load condition of the expert agent pool through the task management module, and determining allocation of the plurality of subtasks according to the priorities and the load condition.

[0011] On the other hand, the present application also proposes an adaptive multi-agent collaboration device, which is applied in an adaptive multi-agent collaboration system, wherein the system includes a meta-agent, an expert agent pool, and a task management module; the device includes: at least one processor; and a memory connected to the at least one processor in communication; wherein the memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor so that the adaptive multi-agent collaboration device can execute: the meta-agent receives a task input by a user, parses the task to obtain a task decomposition tree; the task management module obtains the task decomposition tree, determines the dependency relationship corresponding to the task according to the task decomposition tree, and determines the task execution plan according to the dependency relationship, so as to allocate the task according to the task execution plan, thereby allocating multiple subtasks to the expert agent pool; multiple expert agents in the expert agent pool execute the multiple subtasks after allocation, and collect and integrate the execution results of the multiple subtasks to determine the total task result.

[0012] On the other hand, the present application also proposes a non-volatile computer storage medium storing computer executable instructions, which is applied in an adaptive multi-agent collaboration system, wherein the system includes a meta-agent, an expert agent pool, and a task management module; the computer executable instructions are configured as follows: the meta-agent receives a task input by a user, and parses the task to obtain a task decomposition tree; the task management module obtains the corresponding task decomposition tree, determines the dependency relationship corresponding to the task according to the task decomposition tree, and determines a task execution plan according to the dependency relationship, so as to allocate the task according to the task execution plan, thereby allocating multiple subtasks to the expert agent pool; multiple expert agents in the expert agent pool execute the allocated multiple subtasks, collect and integrate the execution results of the multiple subtasks, and determine the total task result.

[0013] This application, through the introduction of meta-agents, realizes efficient parsing and decomposition of complex tasks, as well as accurate evaluation and dynamic scheduling of agent capabilities, improves the accuracy and efficiency of task processing, and ensures the reasonable allocation of resources and smooth operation of the system. The use of expert agent pools reflects the flexibility and scalability of the system. Different types of agents can work together to cope with various complex task requirements. At the same time, each agent has an independent reasoning engine and local knowledge base, which improves the response speed and independent processing ability of the system. The task management module ensures the efficient and orderly execution of tasks through core functions such as task dependency analysis and resource allocation. The intelligent resource allocation strategy effectively balances the system load and improves the overall processing efficiency. The knowledge sharing center, as the knowledge base and learning platform of the entire framework, promotes knowledge transfer and capability improvement among agents. Through a centralized knowledge management and learning system, the system can quickly access and utilize various complex knowledge, providing strong support for task solving. From task input to result output, the entire process of this application has been carefully designed and arranged, which can efficiently handle various complex tasks and has the ability of continuous optimization and self-evolution. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0015] Figure 1 A schematic diagram of a process of an adaptive multi-agent collaboration method in an embodiment of the present application;

[0016] Figure 2 This is a schematic diagram of an adaptive multi-agent collaboration device in an embodiment of the present application. DETAILED DESCRIPTION

[0017] In order to make the purpose, technical solution and advantages of the present application clearer, the technical solution of the present application will be clearly and completely described below in combination with the specific embodiments of the present application and the corresponding drawings. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present application.

[0018] The technical solutions provided by various embodiments of the present application are described in detail below in conjunction with the accompanying drawings.

[0019] An agent system is a computer system that runs in a specific environment and has characteristics such as autonomy, sociability, responsiveness, and initiative. At present, traditional multi-agent systems usually rely on predefined rules or simple heuristic methods to decompose tasks, which makes it difficult to adapt to complex and changing task requirements. Most systems adopt a static agent organizational structure and lack the ability to dynamically adjust the collaboration mode according to task characteristics, resulting in low resource utilization efficiency. The knowledge transfer and capability complementarity mechanism between agents is imperfect, and the overall learning and evolutionary ability of the system is limited. Existing systems are often designed for specific fields, making it difficult to flexibly respond to task requirements in different fields and lacking versatility. Most systems cannot self-adjust and optimize based on feedback from task execution, and long-term performance is difficult to guarantee.

[0020] like Figure 1 As shown, in order to solve the above problems, an adaptive multi-agent collaboration method provided in an embodiment of the present application is applied in an adaptive multi-agent collaboration system, wherein the system includes a meta-agent, an expert agent pool, a task management module, and a knowledge sharing center; the method includes:

[0021] S101: The meta-agent receives a task input by a user, and parses the task to obtain a task decomposition tree.

[0022] The meta-agent is the core control unit of the entire framework. It is built on a large language model and has multiple responsibilities. In terms of task parsing and decomposition, the meta-agent uses its powerful semantic understanding ability to deeply analyze complex tasks input by users. Through carefully constructed prompts and complex semantic analysis and logical reasoning, it can generate a detailed task decomposition tree and accurately divide complex tasks into multiple specific and executable subtasks. In this process, the meta-agent not only fully considers the logical relationship of the task, but also carefully evaluates the priority and resource requirements of each subtask to ensure the rationality and executability of the task decomposition.

[0023] In terms of agent capability assessment, the meta-agent uses the pre-stored expert agent description information as a knowledge base to conduct a comprehensive and dynamic assessment of each agent in the expert agent pool. The meta-agent builds and continuously updates a dynamic agent capability matrix, which records in detail the multi-dimensional indicators such as each agent's expertise, processing efficiency, and historical performance. This continuous capability assessment mechanism enables the system to always maintain an accurate understanding of each agent's capabilities, providing a solid and reliable basis for subsequent task allocation.

[0024] In terms of global monitoring and scheduling, the Meta-Agent monitors the operating status of the entire system in real time and pays close attention to the work progress and performance of each Agent. Based on these real-time data, the Meta-Agent can keenly discover potential problems or bottlenecks and quickly take corresponding adjustment measures. If necessary, it will re-decompose tasks or reallocate resources to ensure the smooth operation of the entire system and the efficient completion of tasks.

[0025] In one embodiment, during the task input and initialization phase, the user submits a task to the system. After receiving the task, the meta-agent immediately uses a large language model to perform in-depth analysis, and the meta-agent can generate a preliminary task decomposition tree. This process not only covers the logical decomposition of the task, but also includes a preliminary assessment of the task difficulty, priority, and resource requirements.

[0026] Next, the system enters the task planning and allocation phase. In this phase, the task management module will first analyze the dependencies between tasks in detail according to the task decomposition tree, and then formulate a detailed execution plan. At the same time, the meta-agent will play its role again, comprehensively evaluate the currently available agent capabilities, and closely combine the actual needs of the task to make the best task allocation.

[0027] S102, the task management module obtains the task decomposition tree, determines the dependency relationship corresponding to the task according to the task decomposition tree, and determines the task execution plan according to the dependency relationship, so as to allocate the task according to the task execution plan, thereby allocating multiple subtasks to the expert agent pool.

[0028] In one embodiment, the meta-agent first receives the task description input by the user, which contains multiple semantic elements and logical relationships. Using the semantic understanding ability of the large language model, the meta-agent performs in-depth semantic analysis on the user input to identify the key information, concepts, relationships, and possible constraints in the task. Based on the results of the semantic analysis, the meta-agent builds prompts related to the task, which are intended to guide the large language model to perform further logical reasoning and task decomposition.

[0029] Specifically, the meta-Agent splits the input text into individual words or phrases through a large language model, which serves as the basis for subsequent processing. Common stop words that contribute little to the meaning of the text, such as "de" (的) and "le" (了), are removed. For languages like English, words are reduced to their basic forms to decrease lexical diversity. After preprocessing, it enters the semantic understanding and analysis stage. Each word or phrase is converted into a high-dimensional vector, i.e., a lexical vector, which can capture the semantic features of the word. This step relies on pre-trained word embedding models, such as the embedding layers of Word2Vec, GloVe, or BERT. Attention mechanisms or self-attention mechanisms, such as multi-head attention in the Transformer architecture, are used to capture the context information in the input text, enabling the model to understand the different meanings of words in different contexts and adjust its interpretations accordingly. Syntactic structures are implicitly captured through statistical regularities in the training data. At the same time, the model can also understand semantic relationships between words, such as synonyms, antonyms, hyponymy, etc. The model also identifies key information, concepts, relationships, and possible constraints in the input text. This includes entity recognition (such as person names, place names, organization names, etc.), event extraction (such as actions, times, places, etc.), and relationship extraction (such as causal relationships, parallel relationships, etc.). After completing semantic understanding and analysis, it enters the reasoning and generation stage. Logical reasoning is carried out using its knowledge base and reasoning rules, including deductive reasoning (from general to specific), inductive reasoning (from specific to general), and analogical reasoning, etc. Based on the semantic content of the input text and the reasoning results, the model generates corresponding outputs, which may be an answer, an explanation, a suggestion, or a generated new text. During the generation process, various factors are considered, such as the fluency, coherence, accuracy of the output, and whether it meets the user's expectations.

[0030] In one embodiment, the task management module is the central system of the entire framework, responsible for coordinating and orchestrating the task execution process of the entire system to ensure that tasks can be completed efficiently and orderly. Its core responsibilities cover task dependency analysis, resource allocation, progress tracking and exception handling, and result integration.

[0031] In terms of task dependency analysis, the task management module deeply analyzes the complex dependency relationships between subtasks based on the task decomposition tree generated by the meta-Agent. By constructing a directed acyclic graph (DAG) of task execution, this module can clearly reveal the logical connections and execution order between tasks, and then determine the optimal execution path. This dependency analysis not only improves the efficiency of task execution but also effectively avoids the risks of resource conflicts and deadlocks.

[0032] Resource allocation is another important responsibility of the task management module. According to the priority of the task and the current load of the agent, the computing resources are scientifically and reasonably allocated. By implementing parallel processing of tasks, the task management module can maximize the use of system resources. It helps to balance the system load and further improve the overall processing efficiency.

[0033] After the task is completed, the task management module is also responsible for summarizing the execution results of each subtask, performing necessary format conversion and data cleaning, and passing the sorted results to the knowledge sharing center.

[0034] S103: The multiple expert agents in the expert agent pool execute the multiple assigned subtasks, and collect and integrate the execution results of the multiple subtasks to determine the overall task result.

[0035] The expert agent pool is a collection of agents with flexible and scalable design, which contains multiple agents with expertise and processing capabilities in specific fields. The expert agent pool can adapt to various complex task requirements, demonstrating the flexibility and scalability of the system.

[0036] The agents in the expert agent pool are divided into three types: domain expert agents, functional agents, and coordination agents. Domain expert agents focus on specific knowledge areas, such as natural language processing, data analysis, or image processing. With deep domain knowledge and professional skills, they can efficiently cope with complex tasks in related fields. Functional agents mainly undertake general functional tasks, such as information retrieval, code generation, or logical reasoning, which greatly improves the overall functionality and flexibility of the system. Coordination agents play a key intermediate role, responsible for information transmission between subtasks and integration of intermediate results, to ensure the smoothness and coherence of the entire task execution process.

[0037] Each agent is equipped with an independent reasoning engine, which can be a professional model in a specific field or a lightweight large language model that has been carefully fine-tuned. This design ensures that each agent can perform at its best in its own field of expertise. At the same time, each agent also has its own local knowledge base for storing domain-specific knowledge and processing experience. This localized knowledge storage strategy not only improves the agent's response speed, but also enhances its ability to handle tasks independently.

[0038] In one embodiment, during the parallel task execution phase, each expert agent starts working in parallel according to the assigned tasks. In this process, the coordination agent plays a vital role. They are responsible for the transmission and integration of intermediate results, ensuring effective collaboration and information flow between subtasks. This parallel execution mechanism significantly improves the processing efficiency of the system, enabling it to quickly respond to complex task requirements.

[0039] In one embodiment, the task management module is responsible for dealing with various abnormal situations that may occur during the entire execution process to ensure that the task can proceed smoothly. This dynamic adjustment mechanism gives the system the ability to flexibly deal with various emergencies, thereby ensuring the efficient completion of the task.

[0040] In one embodiment, when all subtasks are successfully completed, the system enters the critical stage of result integration and output. In this stage, the task management module is responsible for comprehensively collecting and carefully integrating the results of all subtasks. The meta-agent is responsible for strictly evaluating and optimizing the final results. This process covers in-depth analysis and optimization of the results, ensuring that the final output results are both high-quality and highly practical.

[0041] In one embodiment, the knowledge sharing center is the knowledge foundation and learning platform of the entire framework, which aims to promote the knowledge circulation and capability growth among agents. By building a centralized knowledge management and learning system, the knowledge sharing center significantly improves the intelligence level and adaptability of the entire framework.

[0042] The core component of the knowledge sharing center is its central knowledge base, which brings together common knowledge, task processing models and best practices across fields. The knowledge base adopts an efficient knowledge graph architecture to support fast retrieval and deep association analysis. With this structured knowledge storage method, the system can quickly acquire and apply various complex knowledge, providing solid support for task solving.

[0043] In parallel with the central knowledge base is the experience pool, which records in detail the execution process, results and feedback of historical tasks. With the help of advanced data mining technology, the system can mine valuable patterns and rules from these valuable historical experiences. These extracted experiences and insights can not only provide guidance for future task execution, but also continuously optimize the performance of the entire system.

[0044] Finally, the information in these knowledge bases and experience pools, together with the execution results of the expert agent, are summarized and integrated in the processing unit of the knowledge sharing center. With the help of the prompt function of the large language model, the system can complete the integration work and output the final answer to the user.

[0045] like Figure 2As shown, the embodiment of the present application also provides an adaptive multi-agent collaboration device, which is applied in an adaptive multi-agent collaboration system. The system includes a meta-agent, an expert agent pool, and a task management module; the device includes:

[0046] at least one processor; and,

[0047] a memory communicatively connected to at least one processor; wherein,

[0048] The memory stores instructions that can be executed by at least one processor, and the instructions are executed by at least one processor to enable an adaptive multi-agent collaboration device to perform:

[0049] The meta-agent receives the task input by the user and parses the task to obtain a task decomposition tree;

[0050] The task management module obtains a decomposition tree of the task, determines the dependency relationship corresponding to the task according to the task decomposition tree, and determines a task execution plan according to the dependency relationship, so as to allocate the task according to the task execution plan, thereby allocating multiple subtasks to the expert agent pool;

[0051] The multiple expert agents in the expert agent pool execute the multiple assigned subtasks, and collect and integrate the execution results of the multiple subtasks to determine the overall task result.

[0052] The embodiment of the present application also provides a non-volatile computer storage medium storing computer executable instructions, which is applied in an adaptive multi-agent collaboration system, wherein the system includes a meta-agent, an expert agent pool, and a task management module; the computer executable instructions are set as:

[0053] The meta-agent receives the task input by the user and parses the task to obtain a task decomposition tree;

[0054] The task management module obtains a decomposition tree of the task, determines the dependency relationship corresponding to the task according to the task decomposition tree, and determines a task execution plan according to the dependency relationship, so as to allocate the task according to the task execution plan, thereby allocating multiple subtasks to the expert agent pool;

[0055] The multiple expert agents in the expert agent pool execute the multiple assigned subtasks, and collect and integrate the execution results of the multiple subtasks to determine the overall task result.

[0056] In the 1990s, improvements to a technology could be clearly distinguished as hardware improvements (for example, improvements to the circuit structure of diodes, transistors, switches, etc.) or software improvements (improvements to the method flow). However, with the development of technology, many improvements to the method flow today can be regarded as direct improvements to the hardware circuit structure. Designers almost always obtain the corresponding hardware circuit structure by programming the improved method flow into the hardware circuit. Therefore, it cannot be said that an improvement in a method flow cannot be implemented using a hardware entity module. For example, a programmable logic device (PLD) (such as a field programmable gate array (FPGA)) is such an integrated circuit whose logical function is determined by the user's programming of the device. Designers can "integrate" a digital system on a PLD by programming it themselves, without having to ask a chip manufacturer to design and produce a dedicated integrated circuit chip. Moreover, nowadays, instead of manually making integrated circuit chips, this kind of programming is mostly implemented by "logic compiler" software, which is similar to the software compiler used when developing and writing programs, and the original code before compilation must also be written in a specific programming language, which is called hardware description language (HDL). There is not only one HDL, but many kinds, such as ABEL (Advanced Boolean Expression Language), AHDL (Altera Hardware Description Language), Confluence, CUPL (Cornell University Programming Language), HDCal, JHDL (Java Hardware Description Language), Lava, Lola, MyHDL, PALASM, RHDL (Ruby Hardware Description Language), etc. The most commonly used ones are VHDL (Very-High-Speed ​​Integrated Circuit Hardware Description Language) and Verilog. Those skilled in the art should also know that it is only necessary to program the method flow slightly in the above-mentioned hardware description languages ​​and program it into the integrated circuit, and then it is easy to obtain the hardware circuit that implements the logic method flow.

[0057] The controller can be implemented in any appropriate manner, for example, the controller can take the form of a microprocessor or processor and a computer-readable medium storing a computer-readable program code (such as software or firmware) that can be executed by the (micro)processor, a logic gate, a switch, an application-specific integrated circuit (ASIC), a programmable logic controller, and an embedded microcontroller. Examples of controllers include, but are not limited to, the following microcontrollers: ARC 625D, Atmel AT91SAM, Microchip PIC18F26K20, and Silicone Labs C8051F320. The memory controller can also be implemented as part of the control logic of the memory. Those skilled in the art also know that in addition to implementing the controller in a purely computer-readable program code manner, the controller can be implemented in the form of a logic gate, a switch, an application-specific integrated circuit, a programmable logic controller, and an embedded microcontroller by logically programming the method steps. Therefore, this controller can be considered as a hardware component, and the devices included therein for implementing various functions can also be regarded as structures within the hardware component. Or even, the devices for implementing various functions can be regarded as both software modules for implementing the method and structures within the hardware component.

[0058] The systems, devices, modules or units described in the above embodiments may be implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer. Specifically, the computer may be, for example, a personal computer, a laptop computer, a cellular phone, a camera phone, a smart phone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or a combination of any of these devices.

[0059] For the convenience of description, the above device is described in various units according to their functions. Of course, when implementing this specification, the functions of each unit can be implemented in the same or multiple software and / or hardware.

[0060] Each embodiment in this application is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the device and medium embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiments.

[0061] The devices and media provided in the embodiments of the present application correspond one-to-one to the methods. Therefore, the devices and media also have similar beneficial technical effects as the corresponding methods. Since the beneficial technical effects of the methods have been described in detail above, the beneficial technical effects of the devices and media will not be repeated here.

[0062] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that include computer-usable program code.

[0063] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0064] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0065] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0066] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0067] The memory may include non-permanent storage in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. The memory is an example of a computer-readable medium.

[0068] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.

[0069] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0070] The above are only embodiments of the present application and are not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included within the scope of the claims of the present application.

Claims

1. An adaptive multi-agent collaboration method, characterized in that: The method is applied in an adaptive multi-agent collaboration system, the system comprising a meta-agent, an expert agent pool, and a task management module; the method comprises: The meta-agent receives the task input by the user and parses the task to obtain a task decomposition tree; The task management module obtains a decomposition tree of the task, determines the dependency relationship corresponding to the task according to the task decomposition tree, and determines a task execution plan according to the dependency relationship, so as to allocate the task according to the task execution plan, thereby allocating multiple subtasks to the expert agent pool; The multiple expert agents in the expert agent pool execute the multiple assigned subtasks, and collect and integrate the execution results of the multiple subtasks to determine the overall task result.

2. The method according to claim 1, characterized in that Parsing the task, the method further includes: Preprocessing the task to obtain multiple words; the preprocessing process includes word segmentation, stop word removal, and stem extraction; Converting the plurality of words to obtain a plurality of word vectors, and determining key information, context information, and semantic relationships of the plurality of word vectors; A plurality of subtasks are determined according to the key information, and dependency relationships between the plurality of subtasks are determined according to the context information and the semantic relationship, so as to determine the task decomposition tree according to the dependency relationships.

3. The method according to claim 1, characterized in that Determine the task execution plan based on the dependency relationship, specifically including: Determining the execution order of the plurality of subtasks according to the dependency relationship, so as to determine a directed acyclic graph of task execution according to the execution order; The execution paths of the plurality of subtasks are determined according to the directed acyclic graph, so as to determine the task execution plan according to the execution paths.

4. The method according to claim 1, characterized in that: The tasks are allocated according to the task execution plan, specifically including: The agent types corresponding to the preset multiple expert agents are determined, and the multiple subtasks are allocated to the expert agents of the corresponding agent types according to the task execution plan. The agent types include: domain expert agent, functional agent and coordination agent.

5. The method according to claim 1, characterized in that The method further comprises: Determine the expert agent description information pre-placed in the meta-agent, and determine the knowledge base according to the expert agent description information; Determining a dynamic matrix of the expert agent pool according to the knowledge base, so as to record functions of the expert agent pool according to the dynamic matrix, wherein the functions include areas of expertise, processing efficiency, and historical performance; The running state of the expert agent pool is monitored by the meta-agent, and the expert agent pool is evaluated according to the dynamic matrix, so as to reallocate the plurality of subtasks according to the evaluation result.

6. The method according to claim 1, characterized in that The system also includes a knowledge sharing center; Collecting and integrating the execution results of the multiple subtasks, specifically including: The task management module is used to obtain multiple execution results of the expert agent pool, and the multiple execution results are format converted and data cleaned to obtain an execution result package, which is then sent to the knowledge sharing center.

7. The method according to claim 6, characterized in that The method further comprises: A knowledge base and an experience pool of the knowledge sharing center are determined, and the plurality of execution results are integrated through the knowledge base and the experience pool to determine the overall task result.

8. The method according to claim 1, characterized in that The method further comprises: The priorities corresponding to the plurality of subtasks are determined by the task management module, and the load condition of the expert agent pool is determined, and the allocation of the plurality of subtasks is determined according to the priorities and the load condition.

9. An adaptive multi-agent collaboration device, characterized in that: The invention is applied in an adaptive multi-agent collaboration system, wherein the system includes a meta-agent, an expert agent pool, and a task management module; and the equipment includes: at least one processor; and, a memory communicatively connected to the at least one processor; wherein, The memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the adaptive multi-agent collaboration device to execute: The meta-agent receives the task input by the user and parses the task to obtain a task decomposition tree; The task management module obtains a decomposition tree of the task, determines the dependency relationship corresponding to the task according to the task decomposition tree, and determines a task execution plan according to the dependency relationship, so as to allocate the task according to the task execution plan, thereby allocating multiple subtasks to the expert agent pool; The multiple expert agents in the expert agent pool execute the multiple assigned subtasks, and collect and integrate the execution results of the multiple subtasks to determine the overall task result.

10. A non-volatile computer storage medium storing computer executable instructions, characterized in that: The invention is applied in an adaptive multi-agent collaboration system, wherein the system includes a meta-agent, an expert agent pool, and a task management module; the computer executable instructions are configured as follows: The meta-agent receives the task input by the user and parses the task to obtain a task decomposition tree; The task management module obtains a decomposition tree of the task, determines the dependency relationship corresponding to the task according to the task decomposition tree, and determines a task execution plan according to the dependency relationship, so as to allocate the task according to the task execution plan, thereby allocating multiple subtasks to the expert agent pool; The multiple expert agents in the expert agent pool execute the multiple assigned subtasks, and collect and integrate the execution results of the multiple subtasks to determine the overall task result.

Images