Search

CN-121981194-A - Multi-agent dynamic task planning and cooperation method and system for assisting teaching process

CN121981194ACN 121981194 ACN121981194 ACN 121981194ACN-121981194-A

Abstract

The application discloses a multi-agent dynamic task planning and collaboration method and system for assisting a teaching process. The multi-Agent dynamic task planning and collaboration method for the auxiliary teaching flow comprises the steps of obtaining task demand information, generating a structured task planning diagram and an Agent collaboration path diagram according to the task demand information through multi-Agent task planning and path design, and generating an executable final task planning scheme according to the structured task planning diagram and the Agent collaboration path diagram. According to the application, by introducing a multi-agent dynamic collaboration mechanism, the system can dynamically disassemble and distribute complex comprehensive learning tasks to sub-agents with different expertise for parallel processing according to autonomous planning of a large language model. The expert cooperation mode effectively overcomes the limitation of a single agent in the aspects of professional skills and parallel processing, so that the system can flexibly cope with the multi-element learning requirements of high scene coupling degree and complex flow, and outputs deeper, complete and strict structure results.

Inventors

  • ZHAO JIAN

Assignees

  • 北京竞业达数码科技股份有限公司

Dates

Publication Date
20260505
Application Date
20260107

Claims (10)

  1. 1. The multi-agent dynamic task planning and cooperation method for the auxiliary teaching process is characterized by comprising the following steps of: acquiring task demand information; through multi-Agent task planning and path design, a structured task planning diagram and an Agent cooperation path diagram are generated according to task demand information; and generating an executable final task planning scheme according to the structured task planning diagram and the Agent cooperation path diagram.
  2. 2. The multi-agent dynamic task planning and collaboration method of claim 1, wherein the obtaining task demand information comprises: acquiring natural language task information input by a user; converting the natural language task information into structured task data; Generating task feature vectors according to the structured task data; Generating a collaborative mode decision result according to the task feature vector; And generating task demand information according to the structured task data, the task feature vector and the collaborative mode decision result.
  3. 3. The multi-agent dynamic task planning and collaboration method of assisted instruction process of claim 2 wherein converting natural language task information into structured task data comprises: Acquiring a preset knowledge base in the teaching field; Preprocessing an input natural language task text, so as to obtain a preprocessed task text code; Performing similarity retrieval on the task text codes and a preset knowledge base in the teaching field, so as to obtain a related retrieval result set; obtaining a trained BERT-NER algorithm; text splicing is carried out on the natural language task text and the related search result set, so that text information to be input is obtained; And inputting the text information to be input into a trained BERT-NER algorithm, so as to obtain the structured task data.
  4. 4. A multi-agent dynamic task planning and collaboration method as claimed in claim 3 wherein said generating task feature vectors from structured task data comprises: Acquiring a preset teaching task complexity evaluation index system; extracting a plurality of index features from the structured task data; normalizing each index feature to obtain normalized index features; obtaining a trained lightweight multi-layer perceptron model; inputting the normalized index features into a trained lightweight multi-layer perceptron model, thereby obtaining task complexity scores; Acquiring a preset teaching task constraint classification rule base; Carrying out constraint type classification and constraint intensity scoring on the structured task data through a preset teaching task constraint classification rule base, so as to obtain constraint intensity vectors and constraint classification information; and generating a task feature vector according to the structured task data, the constraint intensity vector, the constraint classification information and the task complexity score.
  5. 5. The method for multi-agent dynamic task planning and collaboration of assisted teaching process according to claim 4, wherein generating collaborative mode decision results from task feature vectors comprises: Acquiring a cooperative mode; generating a high-order characteristic interaction map according to the task characteristic vector and the cooperative mode; generating a dynamic potential energy weight matrix according to the high-order characteristic interaction map; generating evolution robustness adaptation scores of the collaborative mode according to the high-order characteristic interaction atlas and the dynamic potential energy weight matrix; and generating a collaborative mode candidate result according to the evolution robustness adaptation score of the collaborative mode, the high-order characteristic interaction spectrum and the dynamic potential energy weight matrix.
  6. 6. The method for multi-agent dynamic task planning and collaboration of assisted instruction process of claim 5, wherein generating a high-order feature interaction map from task feature vectors comprises: each feature vector in the task feature vectors is used as a feature node, and the initial weight of the node is set as a corresponding feature value; For any K-order feature combination, calculating an interaction gain value G based on a knowledge base K, only preserving beneficial interaction of G more than or equal to 0.1, and constructing interaction edges among feature nodes, wherein the edge weight is the gain value G; And encoding 3-5 order high-order interaction information through an attention aggregation strategy, generating an interaction enhancement characteristic value of each characteristic node, and generating a high-order characteristic interaction map, wherein the high-order characteristic interaction map comprises a plurality of characteristic nodes and a plurality of interaction edges, and labeling the interaction enhancement characteristic value and the edge weight of each node.
  7. 7. The method for multi-agent dynamic task planning and collaboration of assisted instruction process of claim 6, wherein generating a dynamic potential energy weight matrix from the high-order feature interaction atlas comprises: acquiring an initial dynamic potential energy weight matrix W according to the high-order characteristic interaction map; and generating a dynamic potential energy weight matrix according to the initial dynamic potential energy weight matrix W.
  8. 8. The method for multi-agent dynamic task planning and collaboration of assisted teaching process according to claim 7, wherein generating an evolutionary robustness adaptation score for a collaborative model from a high-order feature interaction graph and a dynamic potential energy weight matrix comprises: acquiring a trained LSTM prediction model and a trained generated countermeasure network; inputting the task feature vector into a trained LSTM prediction model, thereby obtaining scene feature vector change probability distribution; Generating a virtual interference scene by generating an countermeasure network; Generating an adaptation score of the virtual interference scene corresponding to each cooperative mode; Calculating the resource coupling degree of each cooperative mode; and generating evolution robustness adaptation scores of all the collaborative modes according to the scene feature vector change probability distribution, the adaptation scores of the virtual interference scenes and the resource coupling degree.
  9. 9. The method for multi-agent dynamic task planning and collaboration of assisted teaching process according to claim 8, wherein the generating a collaborative pattern candidate result according to the evolving robustness adaptation score, the higher-order feature interaction map and the dynamic potential energy weight matrix of the collaborative pattern comprises: generating a feature-mode adaptation bipartite graph and a mode-scene community graph according to the high-order feature interaction graph and the cooperative mode; and fusing the feature-mode adaptation bipartite graph and the mode-scene community graph, so as to obtain a cooperative mode meeting preset conditions as a cooperative mode candidate result.
  10. 10. The multi-agent dynamic task planning and collaboration system for the auxiliary teaching process is characterized by comprising: the task demand information acquisition module is used for acquiring task demand information; the diagram generation module is used for generating a structured task planning diagram and an Agent cooperation path diagram according to task demand information through multi-Agent task planning and path design; And the final task planning scheme generation module is used for generating an executable final task planning scheme according to the structured task planning diagram and the Agent cooperation path diagram.

Description

Multi-agent dynamic task planning and cooperation method and system for assisting teaching process Technical Field The application relates to the technical field of multi-agent cooperation, in particular to a multi-agent dynamic task planning and cooperation method and system for assisting a teaching process. Background The Agent (AI Agent) brings the largest imagination space for people in its ability to "do work autonomously". In the past, AI has been more seen as a "production tool" to assist people in accomplishing various tasks, and today, AI is gradually evolving from the production tool to the "productivity" itself as AI agents evolve. In essence, AI agents are intelligent systems composed of autonomy (Autonomy) together with mobility (Action), and can be generalized to a collaborative structure of "brain + hand". The brain should not only be able to think autonomously, but also interact with the environment and dynamically adjust its own behavior strategies according to the environmental changes, while the hand should be able to complete the work directly according to the instructions of the brain (e.g. DEEP RESEARCH), but also use external tools (e.g. Tool calling). The behavior is not static response any more, but comprises complete cycle of planning, executing and adjusting, thereby realizing task closed loop in the true sense. Under the AI Agent mode, the capability of a single Agent is increasingly limited in the face of task scenes of auxiliary teaching flows, and the problems such as high scene coupling degree, difficult parallel processing and the like are difficult to effectively solve. Disclosure of Invention The invention aims to provide a multi-agent dynamic task planning and cooperation method for assisting a teaching process, which at least solves one technical problem. The invention provides a multi-agent dynamic task planning and cooperation method for an auxiliary teaching process, which comprises the following steps: acquiring task demand information; through multi-Agent task planning and path design, a structured task planning diagram and an Agent cooperation path diagram are generated according to task demand information; and generating an executable final task planning scheme according to the structured task planning diagram and the Agent cooperation path diagram. Optionally, the acquiring task demand information includes: acquiring natural language task information input by a user; converting the natural language task information into structured task data; Generating task feature vectors according to the structured task data; Generating a collaborative mode decision result according to the task feature vector; And generating task demand information according to the structured task data, the task feature vector and the collaborative mode decision result. Optionally, the converting the natural language task information into the structured task data includes: Acquiring a preset knowledge base in the teaching field; Preprocessing an input natural language task text, so as to obtain a preprocessed task text code; Performing similarity retrieval on the task text codes and a preset knowledge base in the teaching field, so as to obtain a related retrieval result set; obtaining a trained BERT-NER algorithm; text splicing is carried out on the natural language task text and the related search result set, so that text information to be input is obtained; And inputting the text information to be input into a trained BERT-NER algorithm, so as to obtain the structured task data. Optionally, the generating task feature vector from the structured task data includes: Acquiring a preset teaching task complexity evaluation index system; extracting a plurality of index features from the structured task data; normalizing each index feature to obtain normalized index features; obtaining a trained lightweight multi-layer perceptron model; inputting the normalized index features into a trained lightweight multi-layer perceptron model, thereby obtaining task complexity scores; Acquiring a preset teaching task constraint classification rule base; Carrying out constraint type classification and constraint intensity scoring on the structured task data through a preset teaching task constraint classification rule base, so as to obtain constraint intensity vectors and constraint classification information; and generating a task feature vector according to the structured task data, the constraint intensity vector, the constraint classification information and the task complexity score. Optionally, the generating the collaborative mode decision result according to the task feature vector includes: Acquiring a cooperative mode; generating a high-order characteristic interaction map according to the task characteristic vector and the cooperative mode; generating a dynamic potential energy weight matrix according to the high-order characteristic interaction map; generating evolution robustness adaptation scores