Search

CN-121985365-A - Network performance layering evaluation method based on multi-agent system

CN121985365ACN 121985365 ACN121985365 ACN 121985365ACN-121985365-A

Abstract

The invention discloses a network performance hierarchical assessment method based on a multi-agent system, which comprises the steps of carrying out hierarchical division on a wireless communication network to be assessed to construct a hierarchical assessment system, collecting performance data of corresponding wireless network objects, constructing a hierarchical assessment graph, carrying out input embedding processing on each assessment node in the hierarchical assessment graph, generating a cross-hierarchical semantic distance coding result and a conflict perception bias result, inputting a node representation set, an edge attribute coding result, the cross-hierarchical semantic distance coding result and the conflict perception bias result into an improved Graphormer model, executing multi-head attention calculation, carrying out gradual convergence on the assessment node update representation set by utilizing a hierarchical virtual agent node chain, and correcting the network global performance assessment result to generate a resource regulation strategy or an operation parameter optimization strategy. The invention adopts multi-agent layered evaluation to realize accurate optimization of wireless network performance.

Inventors

  • ZHOU XIANGJUN

Assignees

  • 睿石网云(杭州)科技有限公司

Dates

Publication Date
20260505
Application Date
20260323

Claims (9)

  1. 1. The network performance layering evaluation method based on the multi-agent system is characterized by comprising the following steps of: performing hierarchical division on a wireless communication network to be evaluated, constructing a hierarchical evaluation system, and deploying a plurality of evaluation agents in corresponding hierarchies to obtain an evaluation agent set; Collecting performance data of a corresponding wireless network object by each evaluation agent in the evaluation agent set to generate a performance data set and a local evaluation result set; Constructing a layered evaluation graph according to the layered evaluation system, the evaluation agent set, the performance data set and the local evaluation result set; performing input embedding processing on each evaluation node in the hierarchical evaluation graph to obtain a node representation set and an edge attribute coding result; Based on any two evaluation nodes in the hierarchical evaluation graph, generating a cross-level semantic distance coding result, and obtaining a conflict perception bias result according to the evaluation nodes of different levels; Inputting the node representation set, the edge attribute coding result, the cross-level semantic distance coding result and the conflict perception bias result into an improved Graphormer model, and executing multi-head attention calculation to obtain an evaluation node update representation set; Using a hierarchical virtual agent node chain to gradually aggregate the update representation set of the evaluation node to obtain a network global performance evaluation result; Detecting a network global performance evaluation result and an evaluation conflict condition, correcting the network global performance evaluation result, and generating a resource regulation strategy or an operation parameter optimization strategy.
  2. 2. The multi-agent system-based network performance hierarchical assessment method of claim 1, wherein the hierarchical assessment system comprises a radio link layer, a radio access node layer, a regional radio network layer, and a global radio network layer.
  3. 3. The method for hierarchical network performance evaluation based on a multi-agent system according to claim 1, wherein the step of collecting, by each evaluation agent in the set of evaluation agents, performance data of a corresponding wireless network object, and generating the set of performance data and the set of local evaluation results specifically comprises: Each evaluation agent in the total set of evaluation agents performs performance data acquisition on the corresponding hierarchical network object according to the corresponding relation matrix; Constructing a performance data vector according to the hierarchy of the hierarchical network object and the service type; performing time alignment and effectiveness screening on each index component in the performance data vector to obtain a standardized acquisition sequence; Constructing an index normalization result aiming at each index component in the standardized acquisition sequence, and constructing a corresponding reverse normalization result aiming at the negative performance index; generating a local evaluation result according to the index weight corresponding to each performance index component; And carrying out aggregation on the performance data vectors, the standardized acquisition sequences and the local evaluation results output by each evaluation agent in the total set of the evaluation agents according to the hierarchical set and the hierarchical attribution relation to generate a performance data set and a local evaluation result set.
  4. 4. The method for hierarchical network performance evaluation based on a multi-agent system according to claim 1, wherein constructing the hierarchical evaluation graph according to the hierarchical evaluation system, the evaluation agent set, the performance data set and the local evaluation result set specifically comprises: establishing corresponding evaluation nodes for each evaluation agent in the total set of evaluation agents to obtain an evaluation node set; establishing corresponding virtual proxy nodes for each level in the level set to obtain a virtual proxy node set; generating a node set of the hierarchical evaluation graph according to the evaluation node set and the virtual agent node set, and distributing a hierarchical index, a node type identifier and an object attribution identifier for each node; obtaining an intra-layer association side set and an inter-layer convergence side set according to the intra-layer association relation set and the inter-layer convergence relation set; Establishing an interlayer feedback relation according to the result transmission requirement and the correction transmission requirement between adjacent layers, and establishing an interlayer feedback edge set based on the interlayer feedback relation; According to the performance data set and the local evaluation result set, calculating conflict judgment values among all evaluation node pairs to obtain a conflict judgment result; According to the conflict judgment result of each evaluation node pair, a conflict node pair set is determined, a conflict check relation is established among each conflict node pair, and a conflict check side set is established; Generating an edge set of the hierarchical evaluation graph according to the intra-layer association edge set, the inter-layer convergence edge set, the proxy convergence edge set, the inter-layer feedback edge set, the proxy feedback edge set and the conflict checking edge set; And constructing the hierarchical evaluation graph according to the node set of the hierarchical evaluation graph and the edge set of the hierarchical evaluation graph.
  5. 5. The method for hierarchical evaluation of network performance based on a multi-agent system according to claim 1, wherein the step of performing input embedding processing on each evaluation node in the hierarchical evaluation graph to obtain a node representation set and an edge attribute coding result specifically comprises: acquiring a hierarchical evaluation graph, and performing input embedding processing on each evaluation node in the hierarchical evaluation graph to generate a performance evaluation input vector, an agent role vector, a hierarchical identification vector, a time state vector, a wireless network topological relation vector and a centrality vector; constructing a spliced input vector according to the performance evaluation input vector, the intelligent agent role vector, the hierarchical identification vector, the time state vector, the wireless network topology relation vector and the centrality vector; linearly mapping the spliced input vectors of all the evaluation nodes to obtain node representations of the corresponding evaluation nodes, and generating a node representation set; For each side in the hierarchical evaluation graph, constructing an edge attribute vector according to the relationship type of the side, the initial node level, the target node level, the side direction, the side connection strength and the conflict judgment value; And carrying out linear mapping on the edge attribute vectors of each edge to obtain edge attribute codes of the corresponding edges, and generating edge attribute coding results.
  6. 6. The network performance hierarchical assessment method based on the multi-agent system according to claim 1, wherein the generating a cross-level semantic distance coding result based on any two assessment nodes in the hierarchical assessment graph, and obtaining a conflict perception bias result according to different levels of assessment nodes specifically comprises: Aiming at any two evaluation nodes in the hierarchical evaluation graph, a node pair set is constructed according to a node representation set, an edge attribute coding result, a performance data set and a local evaluation result set; Generating topology association quantity, hierarchical cross-measurement quantity, index correlation quantity, time synchronization deviation quantity and abnormal propagation risk quantity according to the hierarchical evaluation graph aiming at any node pair in the node pair set; generating a cross-level semantic distance value according to the topological association quantity, the level cross-level quantity, the index correlation quantity, the time synchronization deviation quantity and the abnormal propagation risk quantity; generating cross-level semantic distance codes according to the cross-level semantic distance values, and collecting the cross-level semantic distance codes to generate a cross-level semantic distance code result; Aiming at any node pair in the node pair set, generating a local performance score difference according to local performance scores of two corresponding evaluation nodes in the local evaluation result set at the current moment; Generating a trend correlation quantity according to a local performance score sequence of the node pair corresponding to the two evaluation nodes in the local evaluation result set in a preset time window, and generating a time deviation quantity according to a time state vector of the node pair corresponding to the two evaluation nodes; Generating conflict degree information according to the local performance score difference quantity, the trend correlation quantity and the time deviation quantity, and generating conflict perception bias according to the conflict degree information and the conflict judgment value; and collecting conflict degree information and conflict perception bias corresponding to each node pair in the node pair set, and generating a conflict perception bias result.
  7. 7. The multi-agent system-based network performance hierarchical assessment method according to claim 1, wherein the inputting the node representation set, the edge attribute coding result, the cross-level semantic distance coding result and the conflict perception bias result into the improved Graphormer model, performing multi-head attention calculation, and obtaining the assessment node update representation set specifically comprises: inputting the node representation set, the edge attribute coding result, the cross-level semantic distance coding result and the conflict perception bias result into an improved Graphormer model to construct an input node feature matrix; constructing a relationship enhancement attention unit, and generating a relationship enhancement bias item according to the edge attribute coding result, the cross-level semantic distance coding result and the conflict perception bias result aiming at any node pair; constructing an intra-layer local consistency feature extraction path and an inter-layer convergence feature extraction path in the improved Graphormer model, and generating an intra-layer local consistency mask and an inter-layer convergence mask; in each coding layer, respectively executing query mapping, key mapping and value mapping on the input node feature matrix of the current coding layer to obtain a query matrix, a key matrix and a value matrix; in the intra-layer local consistency feature extraction path, calculating intra-layer attention weights, and carrying out weighted summation to obtain intra-layer local consistency features; in the interlayer convergence feature extraction path, calculating interlayer attention weight, and carrying out weighted summation to obtain interlayer convergence features; splicing intra-layer local consistency characteristics output by each attention head to obtain intra-layer local consistency representation; Splicing the interlayer convergence characteristics output by each attention head to obtain an interlayer convergence representation; In the dual-path fusion updating unit, the intra-layer local consistency representation and the inter-layer convergence representation are spliced and then input into a feature fusion mapping matrix to obtain a fusion representation; and performing residual updating and feedforward updating based on the fusion representation, obtaining output representations of all the evaluation nodes in the current coding layer, and generating an evaluation node updating representation set.
  8. 8. The method for hierarchical evaluation of network performance based on a multi-agent system according to claim 1, wherein the step-by-step aggregation of the update representation sets of the evaluation nodes by using the hierarchical virtual agent node chain to obtain the network global performance evaluation result specifically comprises: Acquiring an evaluation node update representation set, extracting a hierarchy index corresponding to each evaluation node, and obtaining a node update representation subset; In the hierarchical virtual agent node chain, aiming at the virtual agent node corresponding to the wireless link layer, receiving a node update representation subset corresponding to the wireless link layer, and converging step by step to obtain an agent aggregation result of each layer; For virtual proxy nodes corresponding to any hierarchy, calculating proxy aggregation weight, and obtaining a proxy aggregation result corresponding to the current hierarchy; obtaining a global performance representation of the wireless communication network according to agent aggregation results corresponding to each level in the hierarchical virtual agent node chain; and inputting the wireless communication network global performance representation into a global evaluation output layer to obtain a network global performance evaluation result.
  9. 9. The method for hierarchical network performance evaluation based on multi-agent system according to claim 1, wherein the detecting the network global performance evaluation result and the evaluation conflict situation, correcting the network global performance evaluation result, and generating a resource regulation strategy or an operation parameter optimization strategy specifically comprises: acquiring a network global performance evaluation result and agent aggregation results corresponding to virtual agent nodes of each level, and constructing an evaluation conflict detection input set; Based on the network global performance evaluation result, calculating an evaluation conflict detection value corresponding to each level; determining whether an evaluation conflict exists according to the evaluation conflict detection values corresponding to the layers, and triggering a conflict checking process when the existence of the evaluation conflict is determined; According to the hierarchical virtual agent nodes generating the evaluation conflict, combining a conflict perception bias result and a conflict checking edge, positioning corresponding conflict node pairs to obtain a conflict node pair set; updating the attention weight, the side relation and the local evaluation result of the corresponding conflict node pair aiming at each conflict node pair in the conflict node pair set; Regenerating a corresponding edge attribute coding result, a cross-level semantic distance coding result and a conflict perception bias result according to the updated attention weight, the updated edge relation and the updated local evaluation result, and reconstructing a hierarchical evaluation graph; re-inputting the updated hierarchical evaluation graph into the improved Graphormer model, and regenerating a corrected network global performance evaluation result; and generating a resource regulation strategy or an operation parameter optimization strategy of the wireless communication network based on the corrected network global performance evaluation result.

Description

Network performance layering evaluation method based on multi-agent system Technical Field The invention relates to the technical field of wireless communication network performance evaluation and optimization, in particular to a network performance layering evaluation method based on a multi-agent system. Background As wireless communication networks continue to scale up, network structures increasingly exhibit multi-level, heterogeneous, and dynamically changing features, there is a correlation between link state, access node loading, regional network operation, and global network performance. In the prior art, aiming at wireless communication network performance analysis, a single-layer index statistics, local node monitoring or a comprehensive evaluation mode based on fixed rules is generally adopted to monitor indexes such as time delay, throughput, packet loss rate, jitter, resource utilization rate, channel quality, switching success rate and the like, and a network running state result is output according to the indexes. In the prior art, when the association relationship, the convergence relationship and the feedback relationship among the cross-level wireless network objects are processed, a unified hierarchical evaluation mechanism is generally lacked, and the coupling relationship between the local evaluation result and the global evaluation result is not expressed enough. For performance conflict, time deviation and abnormal propagation risk among nodes of different levels, the conventional scheme is difficult to perform joint modeling, so that the overall performance evaluation result is limited in the capability of describing the state of the complex network, and the pertinence of subsequent resource regulation and control and operation parameter optimization is further influenced. Therefore, how to provide a network performance hierarchical assessment method based on a multi-agent system is a problem that needs to be solved by those skilled in the art. Disclosure of Invention The invention aims to provide a network performance layering evaluation method based on a multi-agent system, which comprehensively utilizes multi-agent cooperative sensing, layering evaluation graph construction, cross-level semantic distance coding, conflict sensing bias modeling and improved Graphormer model fusion analysis technology to perform unified modeling, layering evaluation and global fusion on a multi-level performance state of a wireless communication network, performs check correction under the condition of evaluation conflict, and has the advantages of strong layering evaluation capability, high global expression accuracy, strong conflict recognition capability and high resource regulation pertinence. According to the embodiment of the invention, the network performance layering evaluation method based on the multi-agent system comprises the following steps of: performing hierarchical division on a wireless communication network to be evaluated, constructing a hierarchical evaluation system, and deploying a plurality of evaluation agents in corresponding hierarchies to obtain an evaluation agent set; Collecting performance data of a corresponding wireless network object by each evaluation agent in the evaluation agent set to generate a performance data set and a local evaluation result set; Constructing a layered evaluation graph according to the layered evaluation system, the evaluation agent set, the performance data set and the local evaluation result set; performing input embedding processing on each evaluation node in the hierarchical evaluation graph to obtain a node representation set and an edge attribute coding result; Based on any two evaluation nodes in the hierarchical evaluation graph, generating a cross-level semantic distance coding result, and obtaining a conflict perception bias result according to the evaluation nodes of different levels; Inputting the node representation set, the edge attribute coding result, the cross-level semantic distance coding result and the conflict perception bias result into an improved Graphormer model, and executing multi-head attention calculation to obtain an evaluation node update representation set; Using a hierarchical virtual agent node chain to gradually aggregate the update representation set of the evaluation node to obtain a network global performance evaluation result; Detecting a network global performance evaluation result and an evaluation conflict condition, correcting the network global performance evaluation result, and generating a resource regulation strategy or an operation parameter optimization strategy. Optionally, the hierarchical assessment system includes a radio link layer, a radio access node layer, a regional radio network layer, and a global radio network layer. Optionally, the step of collecting, by each evaluation agent in the evaluation agent set, performance data of a corresponding wireless network object, and generating the