Virtual workplace training task allocation method and system based on multi-agent collaboration

Abstract

The invention provides a virtual workplace training task allocation method and system based on multi-agent collaboration, and relates to the technical field of intelligent training, and the method comprises the steps: decomposing a training task into subtasks through constructing a knowledge dependence graph; calculating intellectual body knowledge complementarity, cooperation proficiency and task bearing capacity scores, and fusing the scores into a collaborative score; establishing a knowledge topology network based on the difficulty coefficient and the collaborative score, and selecting an optimal agent combination; and recording a training data optimization distribution mechanism. According to the invention, the training efficiency is improved, the optimal configuration of agent resources is realized, and the pertinence and adaptability of training are enhanced.

Description

Technical Field

[0001] The present invention relates to intelligent training technology, and in particular to a virtual workplace training task allocation method and system based on multi-agent collaboration. Background Art

[0002] Virtual workplace training systems complete training tasks through agent collaboration and have become an important means to improve enterprise training efficiency. Existing virtual workplace training methods mainly rely on a single agent for knowledge transfer, which is difficult to meet the multi-dimensional requirements of complex training tasks and lacks a systematic evaluation of the knowledge coverage and collaboration ability of agents.

[0003] As the complexity of training tasks continues to increase, single agents face bottlenecks in both knowledge breadth and processing power. Although multi-agent collaborative training methods are gradually emerging, existing solutions generally have problems such as unreasonable task decomposition, blind agent selection, and low collaboration efficiency, making it difficult to fully utilize the advantages of agent groups.

[0004] Therefore, there is an urgent need for a multi-agent collaborative training task allocation method based on a knowledge topology network. By analyzing the knowledge dependency relationship of training tasks for task decomposition, evaluating the collaborative effect between agents to establish a knowledge transfer network, dynamic optimization allocation of training tasks is achieved, and the virtual workplace training effect is improved. Summary of the Invention

[0005] Embodiments of the present invention provide a virtual workplace training task allocation method and system based on multi-agent collaboration, which can solve the problems in the prior art.

[0006] In the first aspect of the embodiments of the present invention, a virtual workplace training task allocation method based on multi-agent collaboration is provided, including: Collect training task requirements and the running status of virtual workplace training agents, construct a knowledge dependency graph based on the training task requirements, identify the hierarchical relationship and association degree between training knowledge nodes, and divide the training task into multiple training subtasks; establish an execution priority sequence for the training subtasks, evaluate the difficulty coefficient of each training subtask according to the knowledge node coverage, and generate a training task decomposition plan including the execution order and difficulty rating; Calculate the knowledge complementarity score, cooperation proficiency score, and task carrying capacity score between agents based on the running status of virtual workplace training agents, and perform weighted fusion on the scores to generate an agent collaboration score; Build a knowledge topology network based on the difficulty coefficient in the training task decomposition scheme and the collaborative scoring of agents, extract the transfer path scores of agents in the knowledge topology network, select the agent with the highest transfer path score and whose computing resources are not saturated as the leading training agent, and determine the collaborative training agents according to the knowledge distribution of the transfer path to generate an agent combination configuration scheme for training subtasks; Execute the training subtasks according to the agent collaborative training scheme, record the training process data, update the agent collaborative scoring based on the training process data, and optimize the dynamic task allocation mechanism.

[0007] In an alternative embodiment, Construct a knowledge dependency graph based on the training task requirements, identify the hierarchical relationship and association degree between training knowledge nodes, and divide the training task into multiple training subtasks including: Extract the knowledge point content from the training task requirements, convert the knowledge point content into vector form, calculate the distance between vectors of any two knowledge points to obtain the dependency strength value, and construct a knowledge dependency relationship matrix according to the dependency strength value; Perform singular value decomposition on the knowledge dependency relationship matrix to obtain the eigenvectors of the knowledge points, use the eigenvectors to construct the distribution structure of the knowledge points in the knowledge space, and generate a knowledge dependency graph; In the knowledge dependency graph, calculate the out-degree value and in-degree value of each knowledge point, and determine the hierarchical position coefficient of the current knowledge point according to the ratio of the out-degree value and the in-degree value; Layer the knowledge points in the knowledge dependency graph according to the hierarchical position coefficient, and divide the knowledge points with the same hierarchical position coefficient and whose dependency strength value is greater than the preset strength threshold into a group to obtain multiple training subtasks.

[0008] In an alternative embodiment, Establish an execution priority sequence for training subtasks, evaluate the difficulty coefficient of each training subtask according to the knowledge node coverage, and generate a training task decomposition scheme including the execution order and difficulty rating, including: Obtain the knowledge dependency relationship of the training subtasks, calculate the number of direct incoming edges and direct outgoing edges of the knowledge nodes to obtain the initial dependency value, calculate the number of indirect incoming edges and indirect outgoing edges of the knowledge nodes to obtain the transfer dependency value, and calculate the dependency coefficient of the training subtasks based on the initial dependency value and the transfer dependency value; Mark the reachable nodes on the knowledge dependency graph, calculate the node transfer factor according to the hierarchical distribution of the reachable nodes, and use the product of the node transfer factor and the total number of reachable nodes as the importance value of the training subtask; Construct a state transition matrix using the dependency coefficient and importance value, calculate the state distribution vector through eigenvalue decomposition, use the state distribution vector as the execution priority sequence of the training subtasks, and at the same time calculate the degree centrality value and betweenness centrality value of the knowledge nodes, and use the weighted sum of the degree centrality value and betweenness centrality value as the node coverage, and calculate the node weight in combination with the reference chain length of the knowledge nodes; Accumulate the weights of all knowledge nodes within the training subtasks to obtain the difficulty coefficient, classify the difficulty coefficients of all training subtasks according to the execution priority sequence, and generate a training task decomposition plan including the execution order and difficulty rating.

[0009] In an alternative embodiment, Mark reachable nodes on the knowledge dependency graph, and calculate the node transfer factor according to the hierarchical distribution of the reachable nodes, including: Select a starting node in the knowledge dependency graph, generate a node access mark sequence, traverse the knowledge dependency graph using depth-first search, record the access depth and access path of the nodes, mark the target nodes reachable from the starting node as reachable nodes, and form the access record of the reachable nodes; Use the access depth of the nodes in the access record to divide the node levels, calculate the dependency weights of adjacent nodes, and assign level identifiers to the reachable nodes based on the dependency weights and access depth to obtain the hierarchical distribution structure of the nodes; Based on the hierarchical distribution structure, count the number of predecessor nodes and successor nodes of the nodes, calculate the change rate of the number of nodes in adjacent levels, and calculate the inter-level connection strength of the nodes according to the change rate and the number of predecessor and successor nodes; Generate a node diffusion factor based on the inter-level connection strength, construct a transfer probability table of the nodes in combination with the hierarchical distribution structure, calculate the hierarchical influence strength of the nodes through the transfer probability table, and perform a weighted combination of the node diffusion factor and the hierarchical influence strength to generate the knowledge transfer factor of the nodes.

[0010] In an alternative embodiment, Calculate the knowledge complementarity score, cooperation proficiency score, and task carrying capacity score among the virtual workplace training agents based on the running state of the virtual workplace training agents, and perform a weighted fusion of the scores to generate the agent cooperation score, including: Obtain the running state data of the virtual workplace training agents, including the knowledge state data, interaction state data, and resource state data of the agents; Extract the knowledge vectors and skill vectors of the agents based on the knowledge state data, calculate the difference distribution of the knowledge structures among the agents, analyze the coverage range of the knowledge transfer paths among the agents, and calculate the knowledge complementarity score among the agents according to the difference distribution and the coverage range; Extract the collaboration records of the agents based on the interaction status data, including interaction latency, interaction duration, and task achievement. Introduce a time decay function to the collaboration records, and calculate the cooperation proficiency scores among the agents according to the collaboration records processed by the time decay function; Analyze the computing resource occupancy, storage resource occupancy, and communication resource occupancy of the agents based on the resource status data, evaluate the resource scheduling efficiency and task switching loss of the agents, and calculate the task carrying capacity scores of the agents according to the resource scheduling efficiency and task switching loss; Extract the task scenario features, calculate the weight coefficients of each score according to the task scenario features, and perform non-linear weighted fusion on the knowledge complementarity scores, cooperation proficiency scores, and task carrying capacity scores with the weight coefficients to generate the agent collaboration score.

[0011] In an alternative embodiment, Establish a knowledge topology network based on the difficulty coefficient in the training task decomposition scheme and the agent collaboration score, extract the transfer path scores of the agents in the knowledge topology network, and select the agent with the highest transfer path score and whose computing resources are not saturated as the leading training agent, including: Extract the difficulty coefficients of the training subtasks from the training task decomposition scheme, convert the difficulty coefficients into node basic weights, and at the same time convert the agent collaboration score into node dynamic weights; Construct an initial knowledge topology network using the node basic weights and node dynamic weights, analyze the knowledge level relationship between adjacent nodes in the initial knowledge topology network, calculate the knowledge overlap degree between nodes, determine the knowledge flow direction according to the knowledge level relationship and knowledge overlap degree, and construct a directed transfer path; Calculate the node complexity on the directed transfer path, substitute the node complexity and transfer distance into the exponential decay function to obtain the path knowledge decay value, and multiply the knowledge overlap degree between nodes by the path knowledge decay value to obtain the transfer path score; Collect the task queue length, memory occupancy rate, and response delay time of the agent, calculate the processing load index, compare the processing load index with the preset multi-dimensional saturation threshold to obtain the resource status determination result of the agent, and perform a combined evaluation on the transfer path score and the resource status determination result, and select the agent with the highest transfer path score and not reaching the resource saturation state as the leading training agent.

[0012] In an alternative embodiment, Determine the collaborative training agents according to the knowledge distribution of the transfer path, and generate the agent combination configuration plan for the training subtasks, including: Extract the transfer path of the leading training agent in the knowledge topology network, calculate the proportion of the knowledge weight of each node on the path, construct a knowledge transfer chain based on the knowledge association strength between nodes, analyze the knowledge coverage based on the knowledge transfer chain, and generate knowledge distribution characteristics; Calculate the matching degree between the knowledge distribution characteristics and the requirements of the training subtask, identify the uncovered knowledge content, search for agents with the uncovered knowledge content in the knowledge topology network, calculate the knowledge complementarity degree between the candidate agents and the leading training agent according to the knowledge association strength, and select the agent with the highest knowledge complementarity degree as the collaborative training agent; Determine the knowledge transfer order between the leading training agent and the collaborative training agent based on the knowledge association strength, allocate computing resources to the agents according to the knowledge distribution characteristics and the knowledge complementarity degree, and combine the knowledge transfer order and the computing resource allocation result to generate an agent combination configuration plan for the training subtask.

[0013] In a second aspect of the embodiments of the present invention, there is provided a virtual workplace training task allocation system based on multi-agent collaboration, including: A first unit for collecting training task requirements and the running status of virtual workplace training agents, constructing a knowledge dependency graph based on the training task requirements, identifying the hierarchical relationship and association degree between training knowledge nodes, dividing the training task into multiple training subtasks; establishing an execution priority sequence for the training subtasks, evaluating the difficulty coefficient of each training subtask according to the knowledge node coverage, and generating a training task decomposition plan including the execution order and difficulty rating; A second unit for calculating the knowledge complementarity score, cooperation proficiency score, and task bearing capacity score between agents based on the running status of virtual workplace training agents, and performing weighted fusion on the scores to generate an agent collaboration score; A third unit for establishing a knowledge topology network based on the difficulty coefficient in the training task decomposition plan and the agent collaboration score, extracting the transfer path score of the agent in the knowledge topology network, selecting the agent with the highest transfer path score and whose computing resources are not saturated as the leading training agent, and determining the collaborative training agent according to the knowledge distribution of the transfer path, and generating an agent combination configuration plan for the training subtask; A fourth unit for executing the training subtasks according to the agent collaborative training plan, recording the training process data, updating the agent collaboration score according to the training process data, and optimizing the dynamic task allocation mechanism.

[0014] In a third aspect of the embodiments of the present invention, there is provided an electronic device, including: A processor; A memory for storing instructions executable by the processor; Wherein, the processor is configured to call the instructions stored in the memory to execute the method described above.

[0015] In a fourth aspect of the embodiments of the present invention, there is provided a computer-readable storage medium, on which computer program instructions are stored, and when the computer program instructions are executed by a processor, the method described above is implemented.

[0016] In this embodiment, through the virtual workplace training task allocation method based on multi-agent collaboration, accurate decomposition and intelligent allocation of training tasks are achieved, improving the efficiency and quality of virtual workplace training. By constructing a knowledge dependency graph and a knowledge topology network, it is possible to scientifically evaluate the difficulty coefficient of training tasks and reasonably allocate agent resources, ensuring the systematicness and coherence of training content, and at the same time making full use of the knowledge complementarity and collaborative advantages of different agents. The present invention also establishes a dynamic feedback optimization mechanism, continuously updates the agent collaboration score by recording training process data, enables the task allocation process to be adaptively adjusted, and thus realizes the continuous improvement of training effects and the efficient utilization of system resources. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 It is a schematic flowchart of the virtual workplace training task allocation method based on multi-agent collaboration according to an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0018] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0019] The technical solutions of the present invention will be described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments.

[0020] Figure 1 It is a schematic flowchart of the virtual workplace training task allocation method based on multi-agent collaboration according to an embodiment of the present invention, as Figure 1 shown, the method includes: Collect the training task requirements and the running status of the virtual workplace training agent. Based on the training task requirements, construct a knowledge dependency graph, identify the hierarchical relationship and correlation degree among training knowledge nodes, and divide the training task into multiple training subtasks; establish an execution priority sequence for the training subtasks, evaluate the difficulty coefficient of each training subtask according to the knowledge node coverage, and generate a training task decomposition plan including the execution order and difficulty rating; Calculate the knowledge complementarity score, cooperation proficiency score, and task carrying capacity score among agents based on the running status of the virtual workplace training agent, and perform weighted fusion on each score to generate an agent collaboration score; Based on the difficulty coefficient in the training task decomposition plan and the agent collaboration score, establish a knowledge topology network, extract the transfer path score of the agent in the knowledge topology network, select the agent with the highest transfer path score and whose computing resources have not reached saturation as the leading training agent, and determine the collaborative training agent according to the knowledge distribution of the transfer path to generate an agent combination configuration plan for the training subtasks; Execute the training subtasks according to the agent collaborative training plan, record the training process data, update the agent collaboration score according to the training process data, and optimize the dynamic task allocation mechanism.

[0021] In an optional implementation manner, constructing a knowledge dependency graph based on the training task requirements, identifying the hierarchical relationship and correlation degree among training knowledge nodes, and dividing the training task into multiple training subtasks includes: Extract the knowledge point content from the training task requirements, convert the knowledge point content into a vector form, calculate the distance between vectors of any two knowledge points to obtain a dependency strength value, and construct a knowledge dependency relationship matrix according to the dependency strength value; Perform singular value decomposition on the knowledge dependency relationship matrix to obtain the eigenvectors of the knowledge points, use the eigenvectors to construct the distribution structure of the knowledge points in the knowledge space, and generate a knowledge dependency graph; In the knowledge dependency graph, calculate the out-degree value and in-degree value of each knowledge point, and determine the hierarchical position coefficient of the current knowledge point according to the ratio of the out-degree value and in-degree value; Layer the knowledge points in the knowledge dependency graph according to the hierarchical position coefficient, and divide the knowledge points with the same hierarchical position coefficient and a dependency strength value greater than a preset strength threshold into a group to obtain multiple training subtasks.

[0022] This implementation manner provides a method for constructing a knowledge dependency graph based on training task requirements. This method can identify the hierarchical relationship and correlation degree among training knowledge nodes, divide the training task into multiple training subtasks, thereby improving the training efficiency and learning effect.

[0023] In this embodiment, extracting knowledge point content from training task requirements is the basic step in constructing a knowledge dependency graph. For example, for the task requirements of a "Machine Learning Basics Training", the system extracts relevant knowledge points through text analysis techniques, such as "Linear Regression", "Decision Tree", "Neural Network", "Data Preprocessing", "Model Evaluation", etc. After extraction, the system uses natural language processing techniques to convert this knowledge point content into vector form, and each knowledge point is represented as a high-dimensional vector. In a specific implementation, word embedding techniques or pre-trained language models can be used to convert each knowledge point description into a 300-dimensional vector representation.

[0024] After converting to vector form, the system calculates the distance between the vectors of any two knowledge points to obtain the dependency strength value. For example, by calculating the cosine similarity, the system obtains a dependency strength value of 0.72 between "Linear Regression" and "Decision Tree", while the dependency strength value between "Linear Regression" and "Data Preprocessing" is 0.85, indicating a higher degree of association between "Linear Regression" and "Data Preprocessing". Based on the calculated dependency strength values for all pairs of knowledge points, the system constructs an n×n knowledge dependency matrix M, where n is the total number of knowledge points, and each element M[i][j] in the matrix represents the dependency strength of knowledge point i on knowledge point j.

[0025] Performing singular value decomposition on the constructed knowledge dependency matrix, the system obtains the eigenvectors of the knowledge points. For example, for the dependency matrix of the above "Machine Learning Basics Training", the system obtains 5 main eigenvectors through singular value decomposition, and each eigenvector corresponds to the position coordinates of a knowledge point in the knowledge space. Using these eigenvectors, the system constructs the distribution structure of knowledge points in the knowledge space and generates a visual knowledge dependency graph. In this graph, each node represents a knowledge point, the connection between nodes represents the dependency relationship, and the thickness or color depth of the connection represents the magnitude of the dependency strength.

[0026] In the generated knowledge dependency graph, the system calculates the out-degree value and in-degree value of each knowledge point. The out-degree value represents the degree of influence of this knowledge point on other knowledge points, and the in-degree value represents the degree of influence of other knowledge points on this knowledge point. For example, the out-degree value of "Data Preprocessing" is 4.2 and the in-degree value is 1.3, indicating that "Data Preprocessing" is the basis for multiple other knowledge points. The system determines the hierarchical position coefficient of the current knowledge point according to the ratio of the out-degree value to the in-degree value. Specifically, the hierarchical position coefficient is equal to the out-degree value divided by the in-degree value. If this value is greater than 1.5, then this knowledge point belongs to the basic level; if this value is between 0.7 and 1.5, then this knowledge point belongs to the intermediate level; if this value is less than 0.7, then this knowledge point belongs to the application level.

[0027] According to the calculated hierarchical position coefficient, the system stratifies the knowledge points in the knowledge dependency graph. In this embodiment, the hierarchical position coefficient of "data preprocessing" is 3.23 and it is classified into the basic level; the hierarchical position coefficient of "linear regression" is 1.2 and it is classified into the intermediate level; the hierarchical position coefficient of "model evaluation" is 0.5 and it is classified into the application level. The system divides the knowledge points with the same hierarchical position coefficient and a dependency strength value greater than the preset strength threshold into a group, thereby obtaining multiple training subtasks. For example, when the preset strength threshold is set to 0.7, the system divides "data preprocessing" and "feature engineering" into one training subtask, and divides "linear regression" and "decision tree" into another training subtask.

[0028] In practical applications, for the task requirements of "programming language training", the system extracts knowledge points such as "variable declaration", "data type", "control structure", "function definition", "exception handling", etc. After converting to vector representation, the system calculates that the dependency strength value between "variable declaration" and "data type" is 0.92, and the dependency strength value between "function definition" and "exception handling" is 0.65. After constructing the dependency relationship matrix, the eigenvectors of each knowledge point are obtained through singular value decomposition, and a knowledge dependency graph is generated. The out-degree value of "variable declaration" is calculated to be 5.1, the in-degree value is 0.9, and the hierarchical position coefficient is 5.67, belonging to the basic level; the out-degree value of "exception handling" is 1.8, the in-degree value is 3.2, and the hierarchical position coefficient is 0.56, belonging to the application level. Finally, the system divides "variable declaration" and "data type" into the first training subtask, divides "control structure" and "function definition" into the second training subtask, and divides "exception handling" and "debugging skills" into the third training subtask.

[0029] In this embodiment, by vectorizing the knowledge point content and calculating the dependency strength value, the association relationship between knowledge points can be accurately identified; by extracting the eigenvectors of knowledge points through singular value decomposition, the distribution structure in the knowledge space is constructed to form a knowledge dependency graph with a hierarchical relationship; further, the hierarchical position coefficient is calculated according to the ratio of in-degree to out-degree to achieve the hierarchical positioning of knowledge points, and based on this, the knowledge points are hierarchically clustered, and the knowledge points with strong relevance and similar levels are divided into the same training subtask, thereby ensuring that the knowledge organization within each subtask is tight and the progression is reasonable, which helps to improve the coherence and learning efficiency of the training content.

[0030] In an alternative embodiment, an execution priority sequence of training subtasks is established, and the difficulty coefficient of each training subtask is evaluated according to the knowledge node coverage, and the training task decomposition plan including the execution order and difficulty rating is generated, including: Obtain the knowledge dependencies of the training subtasks, calculate the number of direct incoming edges and direct outgoing edges of the knowledge nodes to obtain the initial dependency values, calculate the number of indirect incoming edges and indirect outgoing edges of the knowledge nodes to obtain the transfer dependency values, and calculate the dependency coefficient of the training subtasks based on the initial dependency values and transfer dependency values; Mark reachable nodes on the knowledge dependency graph, calculate the node transfer factor according to the hierarchical distribution of the reachable nodes, and take the product of the node transfer factor and the total number of reachable nodes as the importance value of the training subtask; Use the dependency coefficient and importance value to construct a state transition matrix, calculate the state distribution vector through eigenvalue decomposition, take the state distribution vector as the execution priority sequence of the training subtasks, and at the same time calculate the degree centrality value and betweenness centrality value of the knowledge nodes, and take the weighted sum of the degree centrality value and betweenness centrality value as the node coverage, and calculate the node weight in combination with the reference chain length of the knowledge nodes; Accumulate the weights of all knowledge nodes within the training subtask to obtain the difficulty coefficient, classify the difficulty coefficients of all training subtasks according to the execution priority sequence, and generate a training task decomposition plan including the execution order and difficulty rating.

[0031] In the specific implementation, first obtain the knowledge dependencies of the training subtasks. Taking a programming training course consisting of 50 knowledge nodes as an example, these nodes include basic concepts such as "variable declaration", "conditional statement", "loop structure", etc. Analyze the dependencies between knowledge nodes and calculate the initial dependency values of each node. For the "variable declaration" node, the number of direct incoming edges is 0 (not dependent on other nodes), and the number of direct outgoing edges is 7 (directly dependent on 7 other nodes). Accordingly, the initial dependency value of "variable declaration" is calculated as -7 (incoming edges minus outgoing edges). At the same time, the indirect incoming edges of "variable declaration" are calculated as 0, and the indirect outgoing edges are 22 (indirectly dependent on 22 nodes through intermediate nodes), obtaining a transfer dependency value of -22. Combining the initial dependency value and the transfer dependency value, according to the weight ratio of 8:2, the dependency coefficient of the "variable declaration" node is calculated as -9.8, indicating that this node is the basis for a large number of other knowledge.

[0032] Mark reachable nodes on the knowledge dependency graph. Starting from the "variable declaration" node, 29 reachable nodes are marked, distributed in 4 different layers. There are 7 directly reachable nodes in the first layer, 12 nodes in the second layer, 8 nodes in the third layer, and 2 nodes in the fourth layer. Calculate the node transfer factor as 0.82, which is obtained through the decreasing ratio of the number of nodes in each layer. Multiply the transfer factor 0.82 by the total number of reachable nodes 29 to obtain the importance value of the "variable declaration" node as 23.78.

[0033] Using the calculated dependency coefficients and importance values above, construct a state transition matrix. For 50 knowledge nodes, the system creates a 50×50 matrix, where each element in the matrix represents the probability of transitioning from one node to another. For example, the transition probability from "Variable Declaration" to "Conditional Statement" is 0.23, which is calculated based on the strength of the dependency between the two nodes. The system processes this matrix through the eigenvalue decomposition algorithm to obtain the steady-state distribution vector. The element values of this vector represent the relative importance of each knowledge node, and the larger the value, the higher the priority. At the same time, calculate the coverage of the knowledge nodes. For the "Variable Declaration" node, its degree centrality value is 7 (directly connected to 7 other nodes), and its betweenness centrality value is 285 (the node is on 285 shortest paths). The node coverage is calculated as 122.2 according to the weight ratio of 6:4. Combining the reference chain length of this node, which is 4 (the hierarchical depth of being cited in the document), the weight of the "Variable Declaration" node is calculated as 488.8.

[0034] For training subtask A that includes three knowledge nodes: "Variable Declaration", "Data Type", and "Operator", accumulate the weight values of these three nodes, which are 488.8, 432.5, and 405.2, to obtain the difficulty coefficient of subtask A as 1326.5. Similarly, calculate the difficulty coefficients of other training subtasks. According to the execution priority sequence, all training subtasks are divided into five levels according to the difficulty coefficient: elementary (difficulty coefficient < 1000), basic (1000 - 2000), intermediate (2000 - 3000), advanced (3000 - 4000), and expert level (> 4000). For example, training subtask A is classified as the basic level.

[0035] Finally, generate a training task decomposition plan that includes the execution order and difficulty ratings. In this plan, the training subtasks are sorted according to the execution priority and the corresponding difficulty ratings are marked. For example, the system recommends completing training subtask A at the basic level first (priority 0.082, difficulty coefficient 1326.5), then proceeding to training subtask C at the intermediate level (priority 0.075, difficulty coefficient 2245.8), and finally executing training subtask B at the advanced level (priority 0.063, difficulty coefficient 3562.1).

[0036] Through this method, the system can generate a scientific and reasonable training task execution plan based on the dependency relationships and importance among knowledge nodes, helping learners master relevant knowledge more effectively. At the same time, the assessment of the difficulty coefficient can help teachers or training institutions better allocate teaching resources and provide targeted guidance for learners at different levels. This method is particularly suitable for learning planning of complex knowledge systems, such as programming, engineering design, medical training and other fields. By calculating the initial dependency value and the transmitted dependency value of knowledge nodes, it comprehensively reflects the direct and indirect relationships between knowledge points, and then accurately calculates the dependency coefficient of subtasks; determines the importance value by combining the hierarchical distribution of reachable nodes, making the task sorting take into account both the structural level and the propagation breadth; constructs the subtask priority sequence through the state transition matrix and eigenvalue decomposition to ensure the scientific and reasonable training path; at the same time, introduces centrality analysis and citation chain length to evaluate the weight of knowledge nodes, finally quantifies the task difficulty, realizes the difficulty grading of all subtasks, and thus generates a training task decomposition plan with clear structure, controllable execution and appropriate difficulty.

[0037] In an alternative implementation, marking reachable nodes on the knowledge dependency graph and calculating the node transfer factor based on the hierarchical distribution of the reachable nodes includes: Select a starting node in the knowledge dependency graph, generate a node access mark sequence, traverse the knowledge dependency graph using depth-first search, record the access depth and access path of the nodes, mark the target nodes reachable from the starting node as reachable nodes, and form the access record of the reachable nodes; Use the access depth of the nodes in the access record to divide the node levels, calculate the dependency weights of adjacent nodes, and assign level identifiers to the reachable nodes based on the dependency weights and access depth to obtain the hierarchical distribution structure of the nodes; Based on the hierarchical distribution structure, count the number of predecessor nodes and successor nodes of the nodes, calculate the change rate of the number of nodes in adjacent levels, and calculate the inter-level connection strength of the nodes according to the change rate and the number of predecessor and successor nodes; Generate a node diffusion factor based on the inter-level connection strength, construct a transfer probability table of the nodes in combination with the hierarchical distribution structure, calculate the hierarchical influence strength of the nodes through the transfer probability table, and perform weighted combination of the node diffusion factor and the hierarchical influence strength to generate the knowledge transfer factor of the nodes.

[0038] In one embodiment, the nodes of the knowledge dependency graph represent knowledge points, and the edges represent the dependency relationships between knowledge points. First, select a starting node from the knowledge dependency graph, such as the "data structure" node, and then generate a node access marker sequence to record whether a node has been visited. Use the depth-first search algorithm to traverse the knowledge dependency graph. For each visited node, record its access depth and the access path from the starting node to this node. For example, when visiting the "data structure" node, its access depth is 0, and the access path only contains itself; when continuing to visit the "array" node, its access depth is 1, and the access path is "data structure → array". During the traversal process, mark the target nodes reachable from the starting node as reachable nodes, forming an access record of reachable nodes. In a specific example, starting from the "data structure" node, the reachable nodes include "array", "linked list", "sorting algorithm", etc., and the access depth and access path of each node are recorded.

[0039] Based on the access depth of the nodes in the access record, divide the node levels. For example, nodes with an access depth of 0 belong to level 0, nodes with an access depth of 1 belong to level 1, and so on. For calculating the dependency weight between adjacent nodes, consider the edge weight value between nodes and the importance of nodes. If the edge weight value from node A to node B is 0.8, it means that node B has a relatively high dependency on node A. Based on the dependency weight and access depth, assign a level identifier to the reachable nodes to obtain the hierarchical distribution structure of the nodes. In practical applications, the "data structure" node may be identified as level 0, the "array", "linked list" and other nodes may be identified as level 1, and the "sorting algorithm" node may be identified as level 2.

[0040] Based on the hierarchical distribution structure, count the number of predecessor nodes and successor nodes of each node. Predecessor nodes refer to the nodes that point to the current node, and successor nodes refer to the nodes that the current node points to. Taking the "sorting algorithm" node as an example, its predecessor nodes may include "array" and "linked list", and the number of predecessor nodes is 2; its successor nodes may include "quick sort" and "merge sort", and the number of successor nodes is 2. Calculate the change rate of the number of nodes between adjacent levels, such as the change rate of the number of nodes from level 1 to level 2. If there are 5 nodes in level 1 and 8 nodes in level 2, then the change rate is (8 - 5) / 5 = 0.6. According to the change rate of the number of nodes and the number of predecessor and successor nodes, calculate the inter-level connection strength of the nodes. For example, for the "sorting algorithm" node, if the number of its predecessor nodes is 2, the number of its successor nodes is 2, and the change rate of the number of nodes between adjacent levels is 0.6, then its inter-level connection strength may be 2×2×0.6 = 2.4.

[0041] Generate a node diffusion factor based on the inter-layer connection strength. The node diffusion factor represents the ability of knowledge to spread from one node to its adjacent nodes. For example, the diffusion factor of the "sorting algorithm" node may be 0.8, indicating a strong knowledge dissemination ability. Combine the hierarchical distribution structure to construct a transfer probability table for nodes, recording the knowledge transfer probabilities from one node to other nodes. For example, the transfer probability from the "data structure" node to the "array" node is 0.7, and the transfer probability to the "linked list" node is 0.6. Through the transfer probability table, calculate the hierarchical influence strength of the node, representing the influence degree of a node on the nodes at its own level and other levels. For example, the influence strength of the "data structure" node on the nodes at the first level is 0.65, and the influence strength on the nodes at the second level is 0.45. Finally, perform a weighted combination of the node diffusion factor and the hierarchical influence strength to generate the knowledge transfer factor of the node. If the diffusion factor of the "data structure" node is 0.85 and the influence strength on the first level is 0.65, then its knowledge transfer factor may be 0.85 × 0.65 = 0.5525.

[0042] In practical applications, it is possible to process knowledge dependency graphs containing thousands of knowledge points. For example, a knowledge dependency graph in the field of computer science may include multiple major categories such as "programming languages", "algorithms", "operating systems", etc., and each major category contains numerous sub-knowledge points. Calculate the knowledge transfer factor of each knowledge point through the above method to provide data support for personalized learning path planning. For example, before a learner studies "algorithms", it is recommended to first study "data structure" with a higher transfer factor, then study "sorting algorithm", and finally study "quick sort" and "merge sort".

[0043] This method for calculating the node transfer factor based on the knowledge dependency graph can effectively capture the dependency relationships and transfer characteristics between knowledge points, provide a reliable decision-making basis for adaptive learning systems, knowledge recommendation systems, and learning path planning, and improve learning efficiency and the quality of knowledge acquisition.

[0044] In this embodiment, it is possible to accurately model the hierarchical positioning and transfer ability of knowledge nodes in the graph. Generate access records through depth-first traversal to clarify the reachability relationships and path depths between nodes, and construct a clear hierarchical distribution structure; use the access depth and dependency weights to assign hierarchical identifiers to refine the relative positions between nodes; further quantify the inter-layer connection strength by statistically calculating the change rate of the number of nodes between layers and the number of predecessor and successor nodes, reflecting the propagation potential of information between layers; combine the diffusion factor and the hierarchical influence strength to form a complete knowledge transfer factor evaluation system, thereby providing efficient support for optimizing the dissemination path of training content, evaluating the importance of nodes, and formulating task execution strategies.

[0045] In an alternative embodiment, based on the running states of virtual workplace training agents, the knowledge complementarity score, cooperation proficiency score, and task carrying capacity score among the agents are calculated, and the weighted fusion of each score is performed to generate the agent collaboration score, including: Obtain the running state data of virtual workplace training agents, including the knowledge state data, interaction state data, and resource state data of the agents; Based on the knowledge state data, extract the knowledge vectors and skill vectors of the agents, calculate the difference distribution of the knowledge structures among the agents, analyze the coverage range of the knowledge transfer paths among the agents, and calculate the knowledge complementarity score among the agents according to the difference distribution and coverage range; Based on the interaction state data, extract the collaboration records of the agents, including interaction latency, interaction duration, and task achievement, introduce a time decay function to the collaboration records, and calculate the cooperation proficiency score among the agents according to the collaboration records processed by the time decay function; Based on the resource state data, analyze the computing resource occupancy, storage resource occupancy, and communication resource occupancy of the agents, evaluate the resource scheduling efficiency and task switching loss of the agents, and calculate the task carrying capacity score of the agents according to the resource scheduling efficiency and task switching loss; Extract the task scenario features, calculate the weight coefficients of each score according to the task scenario features, and perform non-linear weighted fusion on the knowledge complementarity score, cooperation proficiency score, and task carrying capacity score with the weight coefficients to generate the agent collaboration score.

[0046] This embodiment provides a method for calculating the agent collaboration score based on the running states of virtual workplace training agents. The method first obtains the running state data of virtual workplace training agents, including knowledge state data, interaction state data, and resource state data, then calculates the knowledge complementarity score, cooperation proficiency score, and task carrying capacity score respectively, and finally performs weighted fusion according to the task scenario features to generate the agent collaboration score.

[0047] Obtaining the running state data of virtual workplace training agents is the basis of the scoring process. The real-time data of the agents in the virtual workplace environment is collected through the API interface. The knowledge state data includes the concept set mastered by the agents, the skill mastery level, and the knowledge update timestamp; the interaction state data includes the message passing records, cooperation task completion status, and feedback evaluation among the agents; the resource state data includes indicators such as CPU occupancy rate, memory usage, and network bandwidth consumption. For example, for a sales training agent, its knowledge state data may include "Product knowledge: 0.85, Negotiation skills: 0.73, Customer management: 0.91", etc.

[0048] When calculating the knowledge complementarity score, first extract the knowledge vectors and skill vectors of the agents from the knowledge state data. The knowledge vector is represented as a point in an n-dimensional space, where each dimension corresponds to the degree of mastery of a knowledge domain. For example, the knowledge vector of agent A may be [0.9, 0.5, 0.3, 0.8], indicating the knowledge levels in four domains. The skill vector reflects the practical ability of the agent. For example, [0.7, 0.6, 0.9] represents the proficiency levels of three core skills. Calculate the difference distribution of the knowledge structures between agents using vector difference measurement algorithms, such as the complement of the cosine similarity or the normalized value of the Euclidean distance. For example, the Euclidean distance between the knowledge vectors of two agents is 1.2, and the normalized difference degree is 0.6 after normalization. Also analyze the coverage range of the knowledge transfer paths between agents, construct a knowledge graph, and evaluate the knowledge circulation efficiency through path analysis algorithms. For example, all knowledge can be transferred between two agents through three intermediate nodes, with a coverage rate of 95%. Finally, calculate the knowledge complementarity score by integrating the difference distribution and the coverage range, with a value range of 0 - 100 points.

[0049] When calculating the cooperation proficiency score, extract the cooperation records of the agents from the interaction state data. The interaction latency refers to the time interval from when one agent sends a request to when the other agent responds. For example, the average response time is 2.3 seconds; the interaction duration represents the total time to complete a cooperation task. For example, the average cooperation duration is 15 minutes; the task achievement rate measures the completion quality of the task goal. For example, the average achievement rate is 85%. The system introduces a time decay function to the cooperation records, making the recent cooperation records have a higher weight. For example, the weight of a cooperation record one month ago may only be 0.7 times that of the record in the most recent week. Calculate the cooperation proficiency score between agents based on the weighted cooperation records, taking into account the response speed, cooperation efficiency, and task quality. In specific implementation, the basic score can be calculated according to the method of "reciprocal of latency × 0.3 + reciprocal of duration × 0.3 + achievement rate × 0.4", and then multiplied by the time weight to obtain the final cooperation proficiency score, with the value range also being 0 - 100 points.

[0050] When calculating the computing task carrying capacity score, the resource occupancy of the intelligent agent is analyzed based on the resource status data. The computing resource occupancy includes the peak and average CPU usage rates, such as a peak of 85% and an average of 45%; the storage resource occupancy includes the memory usage and storage space requirements, such as a memory occupancy of 600MB; the communication resource occupancy includes the network bandwidth usage rate and data transfer volume, such as an average bandwidth occupancy of 20Mbps. Next, the resource scheduling efficiency of the intelligent agent is evaluated, including the rationality of resource allocation and the parallel processing ability, such as a resource utilization rate of 75%. The task switching loss reflects the time and resource overhead required for the intelligent agent to switch from one task to another, such as an average switching time of 1.2 seconds and a resource loss rate of 8%. The task carrying capacity score of the intelligent agent is calculated based on the resource scheduling efficiency and task switching loss, and the value range is 0-100 points.

[0051] When generating the final collaboration score, first extract the task scenario features, including task complexity, time urgency, and resource limitation degree. For example, a scenario with high complexity, time urgency, and resource limitation may be represented as [0.9, 0.8, 0.7]. Calculate the weight coefficients of each score according to the task scenario features. For example, in knowledge-intensive tasks, the weight of the knowledge complementarity score may be 0.5, the weight of the cooperation proficiency score may be 0.3, and the weight of the task carrying capacity score may be 0.2. The system uses a non-linear weighted fusion method to combine the three scores with the weight coefficients to generate the final intelligent agent collaboration score. The non-linear fusion takes into account the interaction effects between the scores, rather than a simple linear combination. For example, when the knowledge complementarity score is 80 points, the cooperation proficiency score is 70 points, and the task carrying capacity score is 90 points, with respective weights of 0.5, 0.3, and 0.2, the final collaboration score may be 78 points. This score can be used to guide the combination optimization and task allocation decisions of intelligent agents in the virtual workplace.

[0052] In this embodiment, by introducing the operation state data of the virtual workplace training agent, including the knowledge state, interaction state, and resource state, a more comprehensive and dynamic collaborative ability evaluation mechanism is proposed on the basis of only evaluating the agent's ability according to the task completion rate or static features in the prior art. The prior art often ignores key factors such as the knowledge structure differences between agents, the timeliness of the interaction process, and the resource scheduling efficiency, resulting in a lagging and lack of pertinence in the collaborative evaluation results. Starting from the knowledge complementarity, this application extracts the knowledge and skill vectors of the agents, analyzes their structural differences and the coverage of the transfer paths, and improves the perception ability of the collaborative evaluation for the adaptability of the knowledge structure; introduces a time decay function to dynamically adjust the interaction records, and strengthens the description of the evolution characteristics of the cooperation proficiency over time; evaluates the task switching loss and scheduling efficiency through the resource state data, effectively filling the blind spot in the existing evaluation system for the task carrying capacity. Finally, dynamically adjusts the weights of various indicators according to the task scenario characteristics, realizes the non-linear fusion of multi-dimensional scores, and generates a more adaptable and real-time collaborative scoring result. This improvement aims to enhance the rationality of the collaborative task allocation and the collaborative efficiency of the training agent, effectively enhancing the accuracy, flexibility, and practical value of the evaluation system.

[0053] In an alternative embodiment, a knowledge topology network is established based on the difficulty coefficient in the training task decomposition scheme and the collaborative score of the agents, the transfer path score of the agents in the knowledge topology network is extracted, and the agent with the highest transfer path score and whose computing resources are not saturated is selected as the leading training agent, including: Extract the difficulty coefficient of the training subtasks from the training task decomposition scheme, convert the difficulty coefficient into the node basic weight, and at the same time convert the collaborative score of the agents into the node dynamic weight; Construct an initial knowledge topology network using the node basic weight and the node dynamic weight, analyze the knowledge hierarchy relationship between adjacent nodes in the initial knowledge topology network, calculate the knowledge overlap degree between nodes, determine the knowledge flow direction according to the knowledge hierarchy relationship and the knowledge overlap degree, and construct a directed transfer path; Calculate the node complexity on the basis of the directed transfer path, substitute the node complexity and the transfer distance into the exponential decay function to obtain the path knowledge decay value, and multiply the knowledge overlap degree between nodes by the path knowledge decay value to obtain the transfer path score; Collect the task queue length, memory occupancy rate, and response delay time of the agents, calculate the processing load index, compare the processing load index with the preset multi-dimensional saturation threshold to obtain the resource state determination result of the agents, and conduct a combined evaluation of the transfer path score and the resource state determination result, and select the agent with the highest transfer path score and not reaching the resource saturation state as the leading training agent.

[0054] This embodiment provides a method for establishing a knowledge topology network based on the difficulty coefficient and the collaborative scoring of agents in a training task decomposition scheme, extracting the transfer path score of the agent in the knowledge topology network, and thus selecting the optimal leading training agent. This method determines the candidate object most suitable as the leading training agent by constructing a knowledge topology network, calculating the transfer path score, and combining the resource status of the agents.

[0055] In the specific implementation process, first, the difficulty coefficients of the training subtasks are extracted from the training task decomposition scheme. For example, for an image recognition training task containing multiple subtasks, the difficulty coefficient of subtask A is 0.8, the difficulty coefficient of subtask B is 0.6, and the difficulty coefficient of subtask C is 0.4. These difficulty coefficients are converted into node base weights through linear normalization processing so that the sum of the base weights of all nodes is 1. At the same time, the system obtains the collaborative scores between agents. For example, the collaborative score between agent X and agent Y is 0.75, and the collaborative score between agent X and agent Z is 0.6. These collaborative scores are converted into node dynamic weights through the softmax function. The node dynamic weights reflect the performance quality of the agents in historical collaborations.

[0056] Using the node base weights and node dynamic weights, an initial knowledge topology network is constructed. In this network, each node represents an agent, and the size of the node is determined by the weighted sum of the base weight and the dynamic weight. For example, a combination of 0.7 base weight and 0.3 dynamic weight is adopted. The connection between nodes represents the possibility of knowledge transfer between agents, and the initial connection strength is determined by the collaborative score of the two agents.

[0057] Analyze the knowledge hierarchy relationship between adjacent nodes in the initial topology network. This step is achieved by comparing the complexity of the knowledge bases of the nodes. Suppose the knowledge complexity of agent X is 8.5 and the complexity of agent Y is 7.2, then X is higher than Y in the knowledge hierarchy. The system calculates the knowledge overlap degree between nodes. For example, through the method of cosine similarity of word vectors, the knowledge overlap degree between agent X and Y is 0.65. Determine the knowledge flow direction according to the knowledge hierarchy relationship and the overlap degree, flowing from the high-level node to the low-level node, and construct a directed transfer path. In the example, the weight of the directed path from X to Y is 0.65 multiplied by their hierarchy difference (8.5 - 7.2), that is, 0.845.

[0058] Based on the directed transfer path, calculate the complexity of the nodes on the path. For example, in the path X→Y→Z, the complexity of X is 8.5, Y is 7.2, and Z is 6.8. Substitute the node complexity and the transfer distance into the exponential decay function to calculate the path knowledge decay value. Suppose the decay function used is e (-0.1d), where d is the transmission distance. The attenuation value from X to Y is 0.905, and the attenuation value from Y to Z is 0.905. The total attenuation value of the path X→Y→Z is 0.905×0.905 = 0.819. Multiply the knowledge overlap degree between nodes by the path knowledge attenuation value to obtain the transmission path score. Suppose the overlap degree from X to Y is 0.65 and the overlap degree from Y to Z is 0.58. Then the transmission path score of the path X→Y→Z is 0.65×0.58×0.819 = 0.309.

[0059] Collect the task queue length, memory occupancy rate, and response latency of the intelligent agent in real time. For example, the task queue length of intelligent agent X is 15, the memory occupancy rate is 68%, and the response latency is 120 milliseconds. Calculate the processing load metric through weighted average for these metrics, and obtain the processing load metric of X as 0.72. The system presets multi-dimensional saturation thresholds, such as the task queue length threshold is 20, the memory occupancy rate threshold is 85%, and the response latency threshold is 200 milliseconds. Compare the processing load metric 0.72 of X with the normalized value 0.85 corresponding to the preset threshold, and it is obtained that X has not reached the resource saturation state.

[0060] Perform similar calculations for all intelligent agents to obtain their transmission path scores and resource status determination results. For example, the score of the path X→Y→Z is 0.309, and X has not reached the resource saturation; the score of the path Y→X→W is 0.278, and Y has not reached the resource saturation; the score of the path Z→Y→X is 0.245, and Z has reached the resource saturation. By comprehensively evaluating the transmission path score and the resource status, select the intelligent agent with the highest transmission path score and not reaching the resource saturation state as the leading training intelligent agent. In this example, intelligent agent X is selected as the leading training intelligent agent.

[0061] Through the above steps, the system has successfully selected the most suitable leading training intelligent agent from multiple intelligent agents. This intelligent agent has the highest transmission path score in the knowledge topology network, and its computing resources have not reached the saturation state, and it can effectively assume the leading role of the training task, improving the efficiency and quality of the entire training process.

[0062] In this embodiment, by integrating the training task difficulty and the collaborative ability of agents, a knowledge topology network including static cognitive ability and dynamic operating state is established, significantly improving the scientificity and adaptability of training task allocation. Compared with the existing technology that only performs static allocation based on preset rules or task history records, which cannot fully reflect the real-time differences between agents and the degree of knowledge structure matching, this application proposes mapping the difficulty coefficient and collaborative score to the basic and dynamic weights respectively, and constructing a knowledge flow path. By analyzing the knowledge level and overlap degree to determine the transfer direction, combined with the path complexity and knowledge attenuation model, the effectiveness of each path is accurately quantified. At the same time, real-time indicators such as task queue, resource occupancy, and response latency are introduced to conduct a multi-dimensional evaluation of the operating load status of agents, realizing the linkage judgment of resource status and knowledge transfer efficiency. Finally, based on the comprehensive evaluation results, the optimal agent is dynamically selected as the task leader, optimizing the task allocation logic from the source and improving the knowledge transfer efficiency and the overall operating performance of the training system. This technology aims to achieve efficient training scheduling, taking into account task adaptability and resource carrying capacity, and solves the problems of single allocation result and slow response in the existing system.

[0063] In an alternative implementation manner, determining the collaborative training agents according to the knowledge distribution of the transfer path, and the agent combination configuration scheme for generating the training subtasks includes: Extract the transfer path of the leading training agent in the knowledge topology network, calculate the proportion of the knowledge weight of each node on the path, construct a knowledge transfer chain according to the knowledge association strength between nodes, analyze the knowledge coverage based on the knowledge transfer chain, and generate a knowledge distribution feature; Calculate the matching degree between the knowledge distribution feature and the requirements of the training subtasks, identify the uncovered knowledge content, search for agents with the uncovered knowledge content in the knowledge topology network, calculate the knowledge complementarity degree between the candidate agents and the leading training agent according to the knowledge association strength, and select the agent with the highest knowledge complementarity degree as the collaborative training agent; Determine the knowledge transfer order between the leading training agent and the collaborative training agent based on the knowledge association strength, allocate computing resources to the agents according to the knowledge distribution feature and the knowledge complementarity degree, and combine the knowledge transfer order and the computing resource allocation result to generate an agent combination configuration scheme for the training subtasks.

[0064] Exemplarily, it is first necessary to establish a knowledge topology network. This network consists of multiple agent nodes and their connection relationships, and each agent node has a knowledge distribution in different fields. The weight of the connection between nodes represents the knowledge association strength, and the value range is 0 - 1. The higher the value, the closer the association. For example, in the autonomous driving training task, the knowledge association strength between the perception and recognition agent and the decision-making and control agent may be 0.75, while the association strength with the environment simulation agent may be 0.45.

[0065] When extracting the transfer path of the leading training agent in the knowledge topology network, the depth-first search algorithm is used to traverse all the nodes directly connected to the agent, and the path sequence formed during the traversal is recorded. Suppose the leading training agent is A, and the agents directly connected to it are B, C, and D, with knowledge association strengths of 0.8, 0.6, and 0.4 respectively. Then, the A-B path with a priority search strength of 0.8 is searched first, and then the A-C and A-D paths are searched in turn.

[0066] Calculating the proportion of the knowledge weight of each node on the path involves two factors: the knowledge quantity of the node itself and the position of the node in the path. The knowledge quantity of the node itself can be represented by a vector. For example, the knowledge quantity of agent A is [0.9, 0.7, 0.5, 0.2], corresponding to four knowledge dimensions respectively. The influence of the position of the node in the path decays as it moves away from the leading node, and the decay coefficient can be set to 0.8. Taking the A-B-E path as an example, the effective knowledge weights of the three nodes are the weight of A, the weight of B multiplied by 0.8, and the weight of E multiplied by 0.64 respectively.

[0067] When constructing the knowledge transfer chain, the paths are screened according to the knowledge association strength between nodes. The association strength threshold is set to 0.5, and only the paths where the knowledge association strength of all connections on the path is greater than the threshold are retained. For example, if the A-B association strength is 0.8 and the B-E association strength is 0.4, then the A-B-E path is pruned, and only the A-B part is retained. The finally formed knowledge transfer chain may contain multiple paths, such as {A-B, A-C, B-F}.

[0068] Based on the knowledge transfer chain, analyze the knowledge coverage, and calculate the cumulative result of the knowledge vectors of all nodes on the chain. Suppose the dimension of the knowledge vector of the agent is 5, corresponding to five knowledge domains respectively. The knowledge vectors of three agents A, B, and C are [0.9, 0.7, 0.5, 0.2, 0.1], [0.2, 0.8, 0.7, 0.6, 0.3], and [0.4, 0.3, 0.9, 0.7, 0.8] respectively. Considering the position decay, the cumulative knowledge coverage vector of the knowledge transfer chain may be [1.12, 1.27, 1.42, 0.92, 0.73]. This constitutes the knowledge distribution characteristics.

[0069] When calculating the matching degree between the knowledge distribution characteristics and the requirements of the training subtasks, it is necessary to first clarify the knowledge demand vector of the training subtasks. Suppose the knowledge demand vector of a certain training subtask is [0.8, 0.9, 0.7, 0.8, 0.9]. By comparing the elements with the knowledge distribution characteristic vector, the knowledge gap vector can be obtained as [-0.32, -0.37, -0.72, -0.12, 0.17]. Negative values indicate that there is a surplus of knowledge in that dimension, and positive values indicate a knowledge gap. In this example, there is a gap of 0.17 in the fifth knowledge dimension, which is identified as uncovered knowledge content.

[0070] When searching for agents with uncovered knowledge content in the knowledge topology network, agents with high values in the fifth knowledge dimension are selected as candidates. Suppose there are agents G and H in the network, and the knowledge amounts in the fifth dimension are 0.9 and 0.7 respectively, then they become the candidate agents.

[0071] When calculating the knowledge complementarity between the candidate agents and the leading training agent, two aspects are considered: the degree of supplement of the uncovered knowledge and the similarity of the overall knowledge structure. The degree of supplement of the uncovered knowledge is directly compared between the gap value and the knowledge amount of the corresponding dimension of the candidate agent. For example, G can supplement 0.9, exceeding the gap of 0.17, and the supplement rate is 100%; H can supplement 0.7, and the supplement rate is 100%. The similarity of the overall knowledge structure is calculated using the cosine similarity. Suppose the similarity between G and the leading training agent A is 0.65, and the similarity between H and A is 0.82.

[0072] Taking all factors into consideration, the calculation of the knowledge complementarity degree is: the weighted sum of the supplement rate and the similarity. The weights can be set as the supplement rate weight of 0.7 and the similarity weight of 0.3. The knowledge complementarity degree of G is 0.7×1 + 0.3×0.65 = 0.895, and the knowledge complementarity degree of H is 0.7×1 + 0.3×0.82 = 0.946. The knowledge complementarity degree of H is higher, so H is selected as the collaborative training agent.

[0073] When determining the knowledge transfer order, a directed graph is constructed based on the knowledge association strength. If the knowledge association strength between the leading training agent A and the collaborative training agent H is 0.6, and A has the prerequisite knowledge, the transfer order is A→H; conversely, if H has the prerequisite knowledge, it is H→A. In complex cases, a transfer chain like A→C→H may be formed.

[0074] When allocating computing resources according to the knowledge distribution characteristics and the knowledge complementarity degree, the resource allocation ratio is proportional to the importance of the agent in the knowledge transfer. Suppose the total computing resources are 100 units. Since the leading training agent A controls the overall training direction, it obtains 60 units of basic resources; based on its knowledge complementarity degree of 0.946, the collaborative training agent H obtains 60×0.946 = 56.76 units of resources, which is rounded to 57 units.

[0075] The finally generated intelligent agent combination configuration plan for training subtasks includes: the leading training agent A and the collaborative training agent H. The knowledge transfer order is A→H, and the computing resource allocation is 60 units for A and 57 units for H, which are used together to complete the knowledge and skill transfer of specific training subtasks.

[0076] In this embodiment, by introducing the comprehensive analysis of knowledge distribution characteristics and the knowledge complementary relationship between agents, the configuration rationality and collaborative efficiency of training subtasks in a multi-agent system are improved. Existing technologies mostly adopt fixed or empirically set agent combination methods, which cannot dynamically adapt to changes in knowledge structure and individual ability differences, resulting in insufficient coverage of some training tasks or uneven resource allocation. This solution realizes a high-precision match between the task knowledge requirements and the agent's knowledge reserve by extracting the knowledge transfer path of the leading training agent and constructing a knowledge distribution feature model containing knowledge weights and association strengths, thereby identifying knowledge blind spots and intelligently selecting collaborative agents. At the same time, based on knowledge complementarity and transfer order, the collaborative process is optimized, and computing resources are reasonably allocated in combination with the knowledge distribution structure to achieve a double balance between task bearing and knowledge collaboration. This technology effectively solves the problems of random collaboration relationship, resource waste, and weak coverage ability in traditional methods, and enhances the adaptability of training tasks and the collaborative ability of agents.

[0077] In the second aspect of the embodiment of the present invention, a virtual workplace training task allocation system based on multi-agent collaboration is provided. The system includes: The first unit is used to collect training task requirements and the running status of virtual workplace training agents, construct a knowledge dependency graph based on the training task requirements, identify the hierarchical relationship and association degree between training knowledge nodes, and divide the training task into multiple training subtasks; establish an execution priority sequence for the training subtasks, evaluate the difficulty coefficient of each training subtask according to the knowledge node coverage, and generate a training task decomposition plan including the execution order and difficulty rating. The second unit is used to calculate the knowledge complementarity score, cooperation proficiency score, and task bearing ability score between agents based on the running status of virtual workplace training agents, and perform weighted fusion on the scores to generate an agent collaboration score. The third unit is used to establish a knowledge topology network based on the difficulty coefficient in the training task decomposition plan and the agent collaboration score, extract the transfer path score of the agent in the knowledge topology network, select the agent with the highest transfer path score and whose computing resources have not reached the saturation state as the leading training agent, and determine the collaborative training agent according to the knowledge distribution of the transfer path, and generate an intelligent agent combination configuration plan for the training subtask. The fourth unit is used to execute the training subtasks according to the intelligent agent collaborative training plan, record the training process data, update the agent collaboration score according to the training process data, and optimize the dynamic task allocation mechanism.

[0078] In a third aspect of the embodiments of the present invention, there is provided an electronic device, comprising: a processor; a memory for storing instructions executable by the processor; wherein the processor is configured to call the instructions stored in the memory to execute the method described above.

[0079] In a fourth aspect of the embodiments of the present invention, there is provided a computer-readable storage medium, on which computer program instructions are stored, and when the computer program instructions are executed by a processor, the method described above is implemented.

[0080] The present invention may be a method, an apparatus, a system, and / or a computer program product. The computer program product may include a computer-readable storage medium, on which computer-readable program instructions for executing various aspects of the present invention are loaded.

[0081] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some or all of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A virtual workplace training task allocation method based on multi-agent collaboration, characterized in that Including: Collect the training task requirements and the running status of the virtual workplace training agent, construct a knowledge dependency graph based on the training task requirements, identify the hierarchical relationship and association degree among training knowledge nodes, and divide the training task into multiple training subtasks; Establish the execution priority sequence of the training subtasks, evaluate the difficulty coefficient of each training subtask according to the knowledge node coverage, and generate a training task decomposition plan including the execution order and difficulty rating; Calculate the knowledge complementarity score, cooperation proficiency score, and task carrying capacity score among agents based on the running status of the virtual workplace training agent, and perform weighted fusion on each score to generate an agent collaboration score; Establish a knowledge topology network based on the difficulty coefficient in the training task decomposition plan and the agent collaboration score, extract the transfer path score of the agent in the knowledge topology network, select the agent with the highest transfer path score and whose computing resources have not reached saturation as the leading training agent, and determine the collaborative training agents according to the knowledge distribution of the transfer path, and generate an agent combination configuration plan for the training subtasks; Execute the training subtasks according to the agent collaborative training plan, record the training process data, update the agent collaboration score according to the training process data, and optimize the dynamic task allocation mechanism.

2. The method according to claim 1, characterized in that Construct a knowledge dependency graph based on the training task requirements, identify the hierarchical relationship and association degree among training knowledge nodes, and dividing the training task into multiple training subtasks includes: Extract the knowledge point content from the training task requirements, convert the knowledge point content into a vector form, calculate the distance between vectors of any two knowledge points to obtain a dependency intensity value, and construct a knowledge dependency relationship matrix according to the dependency intensity value; Perform singular value decomposition on the knowledge dependency relationship matrix to obtain the eigenvectors of the knowledge points, and use the eigenvectors to construct the distribution structure of the knowledge points in the knowledge space to generate a knowledge dependency graph; In the knowledge dependency graph, calculate the out-degree value and in-degree value of each knowledge point, and determine the hierarchical position coefficient of the current knowledge point according to the ratio of the out-degree value and in-degree value; Layer the knowledge points in the knowledge dependency graph according to the hierarchical position coefficient, and divide the knowledge points with the same hierarchical position coefficient and a dependency intensity value greater than a preset intensity threshold into a group to obtain multiple training subtasks.

3. The method according to claim 1, characterized in that, Establish the execution priority sequence of the training subtasks, evaluate the difficulty coefficient of each training subtask according to the knowledge node coverage, and generate a training task decomposition plan including the execution order and difficulty rating includes: Obtain the knowledge dependency relationship of the training subtask, calculate the number of direct incoming edges and direct outgoing edges of the knowledge node to obtain the initial dependency value, calculate the number of indirect incoming edges and indirect outgoing edges of the knowledge node to obtain the transfer dependency value, and calculate the dependency coefficient of the training subtask based on the initial dependency value and the transfer dependency value; Mark the reachable nodes on the knowledge dependency graph, calculate the node transfer factor according to the hierarchical distribution of the reachable nodes, and use the product of the node transfer factor and the total number of reachable nodes as the importance value of the training subtask; Construct a state transition matrix using the dependency coefficient and importance value, calculate the state distribution vector through eigenvalue decomposition, use the state distribution vector as the execution priority sequence of the training subtasks, and at the same time calculate the degree centrality value and betweenness centrality value of the knowledge nodes. Take the weighted sum of the degree centrality value and betweenness centrality value as the node coverage, and calculate the node weight in combination with the reference chain length of the knowledge nodes; Accumulate the weights of all knowledge nodes within the training subtasks to obtain the difficulty coefficient. Grade the difficulty coefficients of all training subtasks according to the execution priority sequence to generate a training task decomposition plan including the execution order and difficulty rating.

4. The method according to claim 3, wherein Mark reachable nodes on the knowledge dependency graph. Calculate the node transfer factor according to the hierarchical distribution of the reachable nodes, including: Select a starting node in the knowledge dependency graph, generate a node access mark sequence, traverse the knowledge dependency graph using depth-first search, record the access depth and access path of the nodes, and mark the target nodes reachable from the starting node as reachable nodes to form the access record of the reachable nodes; Use the access depth of the nodes in the access record to divide the node levels, calculate the dependency weights of adjacent nodes, and assign level identifiers to the reachable nodes based on the dependency weights and access depth to obtain the hierarchical distribution structure of the nodes; Based on the hierarchical distribution structure, count the number of predecessor nodes and successor nodes of the nodes, calculate the change rate of the number of nodes in adjacent levels, and calculate the inter-layer connection strength of the nodes according to the change rate and the number of predecessor and successor nodes; Generate a node diffusion factor based on the inter-layer connection strength, construct a transfer probability table of the nodes in combination with the hierarchical distribution structure, calculate the hierarchical influence strength of the nodes through the transfer probability table, and perform a weighted combination of the node diffusion factor and the hierarchical influence strength to generate the knowledge transfer factor of the nodes.

5. The method according to claim 1, wherein Calculate the knowledge complementarity score, cooperation proficiency score, and task carrying capacity score among the agents based on the running state of the virtual workplace training agents. Perform a weighted fusion of the scores to generate the agent cooperation score, including: Obtain the running state data of the virtual workplace training agents, including the knowledge state data, interaction state data, and resource state data of the agents; Extract the knowledge vectors and skill vectors of the agents based on the knowledge state data, calculate the difference distribution of the knowledge structures among the agents, analyze the coverage range of the knowledge transfer paths among the agents, and calculate the knowledge complementarity score among the agents according to the difference distribution and coverage range; Extract the cooperation records of the agents based on the interaction state data, including interaction delay, interaction duration, and task achievement. Introduce a time decay function to the cooperation records, and calculate the cooperation proficiency score among the agents according to the cooperation records processed by the time decay function; Analyze the computing resource occupancy, storage resource occupancy, and communication resource occupancy of the agents based on the resource state data, evaluate the resource scheduling efficiency and task switching loss of the agents, and calculate the task carrying capacity score of the agents according to the resource scheduling efficiency and task switching loss; Extract the task scenario features, calculate the weight coefficients of each score according to the task scenario features, and perform non-linear weighted fusion on the knowledge complementarity score, cooperation proficiency score, and task carrying capacity score with the weight coefficients to generate the intelligent agent collaboration score.

6. The method according to claim 1, wherein Based on the difficulty coefficient in the training task decomposition scheme and the intelligent agent collaboration score, establish a knowledge topology network, extract the transfer path score of the intelligent agent in the knowledge topology network, and select the intelligent agent with the highest transfer path score and the computing resources not reaching the saturation state as the leading training intelligent agent, including: Extract the difficulty coefficient of the training subtask from the training task decomposition scheme, convert the difficulty coefficient into the node basic weight, and at the same time convert the intelligent agent collaboration score into the node dynamic weight; Use the node basic weight and node dynamic weight to construct an initial knowledge topology network, analyze the knowledge hierarchy relationship between adjacent nodes in the initial knowledge topology network, calculate the knowledge overlap degree between nodes, determine the knowledge flow direction according to the knowledge hierarchy relationship and knowledge overlap degree, and construct a directed transfer path; Based on the directed transfer path, calculate the node complexity on the path, substitute the node complexity and the transfer distance into the exponential decay function to obtain the path knowledge decay value, and multiply the knowledge overlap degree between nodes by the path knowledge decay value to obtain the transfer path score; Collect the task queue length, memory occupancy rate, and response delay time of the intelligent agent, calculate the processing load index, compare the processing load index with the preset multi-dimensional saturation threshold to obtain the resource status determination result of the intelligent agent, and perform a combined evaluation on the transfer path score and the resource status determination result, and select the intelligent agent with the highest transfer path score and not reaching the resource saturation state as the leading training intelligent agent.

7. The method according to claim 1, characterized in that, Determine the collaborative training intelligent agent according to the knowledge distribution of the transfer path, and generate the intelligent agent combination configuration plan for the training subtask, including: Extract the transfer path of the leading training intelligent agent in the knowledge topology network, calculate the knowledge weight ratio of each node on the path, construct a knowledge transfer chain according to the knowledge association strength between nodes, and analyze the knowledge coverage based on the knowledge transfer chain to generate the knowledge distribution characteristics; Calculate the matching degree between the knowledge distribution characteristics and the requirements of the training subtask, identify the uncovered knowledge content, search for the intelligent agent with the uncovered knowledge content in the knowledge topology network, calculate the knowledge complementarity degree between the candidate intelligent agent and the leading training intelligent agent according to the knowledge association strength, and select the intelligent agent with the highest knowledge complementarity degree as the collaborative training intelligent agent; Determine the knowledge transfer order between the leading training intelligent agent and the collaborative training intelligent agent based on the knowledge association strength, allocate computing resources to the intelligent agent according to the knowledge distribution characteristics and knowledge complementarity degree, and combine the knowledge transfer order and the computing resource allocation result to generate the intelligent agent combination configuration plan for the training subtask.

8. A virtual workplace training task allocation system based on multi-agent collaboration, which is used to implement the method described in any one of the foregoing claims 1-7, characterized in that, Including: The first unit is used to collect the training task requirements and the running status of the virtual workplace training intelligent agent, construct a knowledge dependency graph based on the training task requirements, identify the hierarchical relationship and association degree between the training knowledge nodes, and divide the training task into multiple training subtasks; Establish an execution priority sequence for training subtasks, evaluate the difficulty coefficient of each training subtask according to the knowledge node coverage, and generate a training task decomposition plan including the execution order and difficulty rating; The second unit is used to calculate the knowledge complementarity score, cooperation proficiency score, and task carrying capacity score among agents based on the running state of the virtual workplace training agent, and perform weighted fusion on each score to generate an agent collaboration score; The third unit is used to establish a knowledge topology network based on the difficulty coefficient in the training task decomposition plan and the agent collaboration score, extract the transfer path score of the agent in the knowledge topology network, select the agent with the highest transfer path score and whose computing resources have not reached the saturation state as the leading training agent, and determine the collaborative training agent according to the knowledge distribution of the transfer path, and generate an agent combination configuration plan for training subtasks; The fourth unit is used to execute the training subtasks according to the agent collaborative training plan, record the training process data, update the agent collaborative score according to the training process data, and optimize the dynamic task allocation mechanism.

9. An electronic device, characterized in that, Including: A processor; A memory for storing instructions executable by the processor; Wherein, the processor is configured to call the instructions stored in the memory to execute the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having computer program instructions stored thereon, characterized in that, When the computer program instructions are executed by the processor, the method according to any one of claims 1 to 7 is implemented.

Images