A real-time network quality evaluation and adjustment method

Abstract

The application relates to the technical field of information science, and discloses a real-time network quality evaluation and adjustment method, which comprises the following steps: collecting real-time network quality multi-modal data, performing data fusion and standardization processing on the collected multi-modal data, generating a network quality fusion feature data set, guaranteeing the comprehensiveness and consistency of evaluation basic data, simultaneously using a data cleaning and feature fusion algorithm to process noise and abnormalities in the original data in real time, effectively overcoming the quality evaluation conclusion distortion problem caused by data source deviation and feature missing, guaranteeing the accuracy of real-time network quality evaluation, reducing evaluation errors, and being capable of correcting in real time when evaluation deviation occurs, guaranteeing the dynamic alignment of evaluation criteria and optimization targets, so that the accuracy of network quality perception and the timeliness of decision-making are ensured.

Description

Technical Field

[0001] This invention relates to the field of information science and technology, specifically to a method for real-time network quality assessment and adjustment. Background Technology

[0002] Network quality refers to the network transmission performance reflected by indicators such as bandwidth, latency, and packet loss rate. It includes overall evaluation elements of hardware, software, and environmental systems. Its measurement covers five dimensions: downlink bandwidth, latency, jitter, packet loss rate, and reliability. Evaluation methods include the analytic hierarchy process (AHP) and fuzzy comprehensive evaluation. The technical standard adopts the DiffServ model to implement traffic classification and congestion management.

[0003] Currently, due to the highly dynamic nature of network environments and service loads, real-time network quality assessments rely on data collection sources from fixed probes and periodic active probing. This makes it impossible to detect in real time whether the original data on which the network quality assessment is based is biased or biased. When the underlying data has missing features or noise interference, the conclusions of the quality assessment will be significantly distorted, and the accuracy of the assessment results cannot be guaranteed.

[0004] Therefore, a real-time network quality assessment and adjustment method is proposed to address the above problems. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a real-time network quality assessment and adjustment method, which solves the problem mentioned in the background art of being unable to perceive in real time whether the original data on which the network quality assessment is based is biased or biased.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a real-time network quality assessment and adjustment method, the method comprising the following steps: S1. Collect real-time network quality multimodal data; S2. Perform data fusion and standardization processing on the collected multimodal data to generate a network quality fusion feature dataset; S3. Based on the network quality fusion feature dataset, use a pre-trained real-time quality assessment model to calculate and generate multi-dimensional real-time quality index data. S4. Based on the multidimensional real-time quality index data and combined with the preset real-time quality scoring rules, generate a dynamic quality score. S5. Obtain historical network quality data and network topology data, and construct a twin simulation environment for network quality assessment; S6. Based on the dynamic quality score, the network quality fusion feature dataset, and the twin simulation environment, the optimal network quality adjustment scheme is generated by using the dynamic optimization strategy generation module. S7. Based on the optimal network quality adjustment scheme, send real-time network configuration adjustment instructions to the target network nodes and links; S8. Monitor and collect network quality feedback data in real time after executing the network configuration adjustment command; S9. Based on the network quality feedback data, adaptive parameter optimization is performed on the real-time quality assessment model, the real-time quality scoring rules, and the dynamic optimization strategy generation module.

[0007] Preferably, the acquisition of real-time network quality multimodal data in step S1 includes the following steps: S11. By deploying probes on network nodes and user terminals, network traffic data packets are collected in real time. The network traffic data packets include transport layer protocol header information, application layer payload characteristic information, and data packet arrival time sequence information. S12. Using an active detection device, send probe data packets to the target service address at a preset period and collect network performance parameters, including end-to-end latency, jitter, packet loss rate and bandwidth availability data. S13. Real-time collection of network device operating status parameters through the network infrastructure management interface, including port utilization, CPU load, memory utilization, cache queue depth, and error packet count.

[0008] Preferably, the generation of the network quality fusion feature dataset in S2 includes the following steps: S21. Perform timestamp alignment and data cleaning on the network traffic data packets, the network performance parameters, and the network device operating status parameters respectively to remove outliers and noisy data. S22. Map the cleaned network traffic data packets, network performance parameters, and network device operating status parameters to a unified spatiotemporal data coordinate system, wherein the spatiotemporal data coordinate system uses absolute timestamps as the primary index and network node and link identifiers as the spatial index. S23. A feature fusion algorithm is used to fuse heterogeneous data mapped to the spatiotemporal data coordinate system. The feature fusion algorithm includes feature weighted fusion based on attention weights and feature relationship extraction based on graph neural networks to generate the network quality fusion feature dataset.

[0009] Preferably, the generation of multidimensional real-time quality index data in step S3 includes the following steps: S31. Input the network quality fusion feature dataset into the pre-trained real-time quality assessment model, wherein the real-time quality assessment model is a hybrid neural network model based on the combination of temporal convolutional networks and long short-term memory networks. S32. The real-time quality assessment model outputs quality assessment results in multiple dimensions, including real-time throughput assessment value, real-time latency satisfaction, real-time service availability index, and real-time user experience score, which together constitute the multi-dimensional real-time quality indicator data.

[0010] Preferably, generating a dynamic quality score in step S4 includes the following steps: S41. Obtain the preset real-time quality scoring rules, which define the weight coefficient, scoring threshold, and non-linear scoring function for each dimension in the multi-dimensional real-time quality index data. S42. Substitute the real-time throughput evaluation value, the real-time latency satisfaction, the real-time service availability index, and the real-time user experience score into the nonlinear scoring function to calculate the initial scores for each dimension. S43. The initial scores of each dimension are weighted and summed according to the weight coefficients, and normalization is applied to generate the dynamic quality score.

[0011] Preferably, the construction of the twin simulation environment for network quality assessment in S5 includes the following steps: S51. Extract the historical network quality data from the historical network quality database. The historical network quality data includes historical quality indicators and corresponding network status data under the same and similar network load conditions. S52. Obtain the network topology data of the current network, the network topology data including node connection relationships, link attributes and deployment information of network service function chains; S53. Based on the historical network quality data and the network topology data, construct the twin simulation environment on the digital twin platform generator. The twin simulation environment can simulate the changes of the multidimensional real-time quality index data and the dynamic quality score under different network configuration parameters.

[0012] Preferably, generating the optimal network quality adjustment scheme in S6 includes the following steps: S61. With the goal of improving the dynamic quality score, an adjustable set of network configuration parameters is set in the twin simulation environment. The set of network configuration parameters includes routing weight, traffic scheduling strategy, service quality level identifier, congestion avoidance algorithm parameters, and bandwidth reservation ratio. S62. The dynamic quality score, the network quality fusion feature dataset, and the current network configuration status are taken as inputs, and the dynamic optimization strategy generation module performs iterative simulation and optimization in the twin simulation environment. The dynamic optimization strategy generation module adopts a decision algorithm based on deep reinforcement learning. S63. The dynamic optimization strategy generation module outputs a combination of adjustment parameters for the set of network configuration parameters, which serves as the optimal network quality adjustment scheme.

[0013] Preferably, sending the real-time network configuration adjustment command in S7 includes the following steps: S71. Analyze the optimal network quality adjustment scheme and decompose it into a sequence of executable configuration commands for specific network devices; S72. The configuration command sequence is encapsulated into network configuration adjustment instructions through a secure encrypted channel and sent to the corresponding target network node controller and link controller respectively. S73. Receive and verify the instruction execution confirmation signal returned by the target network node and link.

[0014] Preferably, the real-time monitoring and collection of network quality feedback data in step S8 includes the following steps: S81. Within one or more preset monitoring cycles after sending the network configuration adjustment instruction and receiving the instruction execution confirmation signal, S1 and S2 are re-executed to collect and generate the adjusted network quality fusion feature dataset. S82. Input the adjusted network quality fusion feature dataset into the real-time quality assessment model to generate adjusted multi-dimensional real-time quality index data. S83. Based on the adjusted multidimensional real-time quality index data and the real-time quality scoring rules, the adjusted dynamic quality score is calculated and used as a core component of the network quality feedback data.

[0015] Preferably, the adaptive parameter optimization in S9 includes the following steps: S91. The network quality feedback data, the network quality fusion feature dataset before adjustment, and the optimal network quality adjustment scheme are used together as training samples to form an experience replay dataset. S92. Using the experience replay dataset, periodically fine-tune the network weight parameters of the real-time quality assessment model so that its assessment results are closer to the actual network state changes. S93. Based on the long-term optimization effect reflected in the network quality feedback data, adaptively calibrate the weight coefficients and scoring thresholds in the real-time quality scoring rules. S94. Input the experience replay dataset into the deep reinforcement learning model of the dynamic optimization policy generation module for offline training, and update its policy network parameters to improve the accuracy and efficiency of future generation adjustment schemes.

[0016] Compared with existing technologies, the present invention provides a real-time network quality assessment and adjustment method, which has the following beneficial effects: 1. In this invention, when performing real-time network quality assessment, a unified network quality fusion feature dataset is generated by integrating multimodal data sources and performing standardized fusion processing. This ensures the comprehensiveness and consistency of the assessment basis data. At the same time, data cleaning and feature fusion algorithms are used to process noise and anomalies in the original data in real time. This effectively overcomes the problem of distorted quality assessment conclusions caused by data source bias and feature loss, ensuring the accuracy of real-time network quality assessment and reducing assessment errors.

[0017] 2. In this invention, when performing real-time network quality assessment and adjustment, a closed-loop optimization mechanism based on network quality feedback data is established to perform parameter calibration and adaptive updates on the real-time quality assessment model, real-time quality scoring rules, and dynamic optimization strategy generation module in real time. This ensures that the method can continuously guarantee the matching degree between the assessment model and the current network and service characteristics, and can correct deviations in real time when they occur, ensuring the dynamic alignment between the assessment benchmark and the optimization target, thereby guaranteeing the accuracy of network quality perception and the timeliness of decision-making.

[0018] 3. In this invention, when optimizing and adjusting network quality, a twin simulation environment integrating historical data and topology is constructed. In this environment, based on dynamic quality scores and real-time feature data, a deep reinforcement learning-driven dynamic optimization strategy generation module is used to perform multi-objective iterative optimization to generate the optimal network quality adjustment scheme. This enables the method to achieve refined and collaborative strategy calculation for complex and ever-changing network conditions, effectively avoiding the impact of strategy conflicts and secondary performance fluctuations on the network, thereby improving the final effect of network quality adjustment and overall network stability. Attached Figure Description

[0019] Figure 1 This is a flowchart of a real-time network quality assessment and adjustment method according to the present invention. Detailed Implementation

[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] Please see Figure 1 This method for real-time network quality assessment and adjustment includes the following steps: S1. Collect real-time network quality multimodal data; S2. Perform data fusion and standardization processing on the collected multimodal data to generate a network quality fusion feature dataset; S3. Based on the network quality fusion feature dataset, a pre-trained real-time quality assessment model is used to calculate and generate multi-dimensional real-time quality index data. S4. Based on multi-dimensional real-time quality indicator data and combined with preset real-time quality scoring rules, generate dynamic quality scores. S5. Obtain historical network quality data and network topology data, and construct a twin simulation environment for network quality assessment; S6. Based on dynamic quality scoring, network quality fusion feature dataset and twin simulation environment, the optimal network quality adjustment scheme is generated by the dynamic optimization strategy generation module. S7. Based on the optimal network quality adjustment scheme, send real-time network configuration adjustment commands to the target network nodes and links; S8. Monitor and collect network quality feedback data in real time after executing network configuration adjustment commands; S9. Based on network quality feedback data, adaptive parameter optimization is performed on the real-time quality assessment model, real-time quality scoring rules, and dynamic optimization strategy generation module.

[0022] The process of collecting real-time network quality multimodal data in S1 includes the following steps: S11. By deploying probes on network nodes and user terminals, network traffic data packets are collected in real time. The network traffic data packets include transport layer protocol header information, application layer payload characteristic information, and data packet arrival time sequence information. S12. Using an active probing device, send probe data packets to the target service address at a preset period and collect network performance parameters, including end-to-end latency, jitter, packet loss rate and bandwidth availability data. S13. Collect network device operating status parameters in real time through the network infrastructure management interface. These parameters include port utilization, CPU load, memory usage, cache queue depth, and error packet count.

[0023] The steps involved in generating the network quality fusion feature dataset in S2 are as follows: S21. Perform timestamp alignment and data cleaning on network traffic data packets, network performance parameters, and network device operating status parameters respectively to remove outliers and noisy data. Formula expression: ; Character explanation: Indicates the first Original data set Indicates absolute timestamp A single data point collected at that location. and Representing datasets respectively The sample mean and sample standard deviation, A constant threshold factor, This represents the data cleaning operation function. This indicates the result after cleaning and alignment. Class-normalized datasets, For indexing, For indexing; S22. Map the cleaned network traffic data packets, network performance parameters, and network device operating status parameters to a unified spatiotemporal data coordinate system. The spatiotemporal data coordinate system uses the absolute timestamp as the primary index and the network node and link identifiers as the spatial index. Formula expression: ; Character explanation: Represents the spacetime mapping operator, Represents an absolute timestamp. Represents a spatial location identifier vector. Represents in spacetime coordinates The multimodal data vector defined at that point, This indicates the result after cleaning and alignment. Class-normalized datasets, For indexing; S23. A feature fusion algorithm is used to fuse heterogeneous data mapped to a spatiotemporal data coordinate system. The feature fusion algorithm includes feature weighted fusion based on attention weights and feature relationship extraction based on graph neural networks to generate a network quality fusion feature dataset. Formula expression: Feature weighting based on attention weights: ; ; Relation extraction based on graph neural network (GNN): ; Fusion Output: ; Character explanation: Indicates the first Multimodal data vectors of spatiotemporal coordinate points For learnable parameter matrices and vectors, Indicates the first Attention weights for each feature This represents the feature vector after attention-weighted aggregation. This represents the normalized network topology adjacency matrix. Represents the GNN No. Layer node feature representation, Represents the GNN No. The learnable weight matrix of the layer, Represents a non-linear activation function. This represents the total number of layers in the GNN. This represents a vector concatenation operation. This represents the final output network quality fusion feature vector. For indexing, For indexing, It is a transpose operator. The total number of samples.

[0024] Generating multidimensional real-time quality indicator data in S3 includes the following steps: S31. Input the network quality fusion feature dataset into the pre-trained real-time quality assessment model. The real-time quality assessment model is a hybrid neural network model based on the combination of temporal convolutional network and long short-term memory network. Formula expression: ; Character explanation: This represents the final output network quality fusion feature vector. The parameter is of Hybrid neural network model, This represents a multidimensional real-time quality index data vector output by the model. S32. The real-time quality assessment model outputs quality assessment results in multiple dimensions, including real-time throughput assessment value, real-time latency satisfaction, real-time service availability index, and real-time user experience score, which together constitute multi-dimensional real-time quality indicator data.

[0025] Generating dynamic quality scores in S4 involves the following steps: S41. Obtain the preset real-time quality scoring rules. The real-time quality scoring rules define the weight coefficient, scoring threshold and non-linear scoring function of each dimension in the multi-dimensional real-time quality indicator data. Formula expression: ; Character explanation: This represents the preset set of real-time quality scoring rules. Indicates the dimensions of quality assessment. Represents the set of indices for all evaluation dimensions. Indicates the first Weight coefficients for each dimension and They represent the first The lower and upper limits of the scoring thresholds for each dimension. Indicates the first A non-linear scoring function specific to each dimension; S42. Substitute the real-time throughput evaluation value, real-time latency satisfaction, real-time service availability index, and real-time user experience score into the non-linear scoring function to calculate the initial scores for each dimension. Formula expression: ; Character explanation: Indicates the first Real-time quality indicator data in multiple dimensions Indicates the first A non-linear scoring function specific to each dimension. Indicates the first The initial scores are calculated from each dimension. Indicates the dimensions of quality assessment. and They represent the first The lower and upper limits of the scoring thresholds for each dimension; S43. The initial scores of each dimension are weighted and summed according to the weight coefficients, and normalization is applied to generate a dynamic quality score. Formula expression: ; Character explanation: Indicates the first Weight coefficients for each dimension Indicates the first The initial scores are calculated from each dimension. Indicates a weighted sum. Represents the normalization function. This represents the final output dynamic quality score. Indicates the dimensions of quality assessment. This represents the set of indexes for all evaluation dimensions.

[0026] Building a twin simulation environment for network quality assessment in S5 includes the following steps: S51. Extract historical network quality data from the historical network quality database. The historical network quality data includes historical quality indicators and corresponding network status data under the same and similar network load conditions. S52. Obtain the current network topology data, which includes node connection relationships, link attributes, and deployment information of network service function chains; S53. Based on historical network quality data and network topology data, a twin simulation environment is built on the digital twin platform generator. The twin simulation environment can simulate the changes in multi-dimensional real-time quality index data and dynamic quality scores under different network configuration parameters. Formula expression: ; Character explanation: This represents a set of historical network quality data. Representing a network topology diagram, This represents a set of adjustable network configuration parameters. This represents the set of physical rules and service flow model functions that represent the simulated network behavior within the simulation environment. This represents the construction function of the digital twin platform generator. This represents the constructed twin simulation environment.

[0027] Generating the optimal network quality adjustment scheme in S6 includes the following steps: S61. With the goal of improving dynamic quality score, set an adjustable set of network configuration parameters in the twin simulation environment. The set of network configuration parameters includes routing weight, traffic scheduling strategy, service quality level identifier, congestion avoidance algorithm parameters and bandwidth reservation ratio. S62. The dynamic quality score, network quality fusion feature dataset and current network configuration status are taken as input. The dynamic optimization strategy generation module performs iterative simulation and optimization in the twin simulation environment. The dynamic optimization strategy generation module adopts a decision algorithm based on deep reinforcement learning. Formula expression: ; ; ; Symbol explanation: This represents the policy function of the dynamic optimization policy generation module. For parameters, Indicates the simulation time step The state vector, Representation Strategy In state The action of outputting below, Represents the reward function, Indicates the discount factor. Indicating a twin environment In the middle, follow the strategy Generate trajectory The probability distribution, This represents the optimal strategy obtained through deep reinforcement learning algorithms. Indicates time, This represents the current dynamic quality score, and the network quality fusion feature dataset. The current network configuration status; S63. The dynamic optimization strategy generation module outputs a combination of adjustment parameters for the set of network configuration parameters, which serves as the optimal network quality adjustment scheme.

[0028] Sending real-time network configuration adjustment commands in S7 includes the following steps: S71. Analyze the optimal network quality adjustment scheme and decompose it into a sequence of executable configuration commands for specific network devices; Formula expression: ; Character explanation: Represents the solution parsing function. This represents the optimal network quality adjustment scheme. Represents the target set of network devices. This indicates the parsed and generated data for the target device. Executable configuration commands; S72. Encapsulate the configuration command sequence into network configuration adjustment instructions through a secure encrypted channel, and send them to the corresponding target network node controller and link controller respectively. S73. Receive and verify the instruction execution confirmation signal returned by the target network node and link.

[0029] Real-time monitoring and collection of network quality feedback data in S8 includes the following steps: S81. Within one or more preset monitoring cycles after sending the network configuration adjustment command and receiving the command execution confirmation signal, S1 and S2 are re-executed to collect and generate the adjusted network quality fusion feature dataset. S82. Input the adjusted network quality fusion feature dataset into the real-time quality assessment model to generate the adjusted multi-dimensional real-time quality index data. S83. Based on the adjusted multidimensional real-time quality index data and real-time quality scoring rules, the adjusted dynamic quality score is calculated and used as a core component of the network quality feedback data. Formula expression: ; Character explanation: This represents the adjusted multidimensional real-time quality indicator data. This indicates the adjusted dynamic quality score. This represents the optimal network quality adjustment scheme. This represents the network quality feedback data after the final adaptive optimization.

[0030] Adaptive parameter optimization in S9 includes the following steps: S91. The network quality feedback data, the network quality fusion feature dataset before the adjustment, and the optimal network quality adjustment scheme are used together as training samples to form an experience playback dataset. Formula expression: ; Character explanation: This represents the network state vector before the adjustment. This represents the experience replay dataset that stores the training samples. This represents the optimal network quality adjustment scheme. This represents the network quality feedback data from the final adaptive optimization. S92. Using the experience replay dataset, periodically fine-tune the network weight parameters of the real-time quality assessment model to make its assessment results closer to the actual network state changes. Formula expression: ; Character explanation: and These represent the weight parameters of the real-time quality assessment model before and after fine-tuning, respectively. Indicates the learning rate. Indicates the parameter Find the gradient. Represents the loss function. This represents the adjusted multidimensional real-time quality indicator data. This represents the final output network quality fusion feature vector. express Hybrid neural network model; S93. Based on the long-term optimization effect reflected in the network quality feedback data, adaptively calibrate the weight coefficients and scoring thresholds in the real-time quality scoring rules. S94. Input the experience replay dataset into the deep reinforcement learning model of the dynamic optimization policy generation module for offline training, update its policy network parameters, and improve the accuracy and efficiency of future generation adjustment schemes. Formula expression: ; ; ; Character explanation: and These represent the parameters of the policy network before and after training, respectively. This represents the learning rate during the training of the policy network. This represents the objective function for training the policy network. Indicates from A transferred sample from the middle, in which It's a state. It's an action. It's an instant reward. The next state is... The parameter is Action value function, Indicates based on old parameters The calculated time difference target value, This represents the discount factor.

[0031] The operational steps of a real-time network quality assessment and adjustment method are as follows: Step 1: Multimodal data acquisition: This method first collects comprehensive real-time data from different layers of the network to form the basis of the evaluation. This includes: capturing traffic packets through network probes, analyzing their protocol headers and arrival times, periodically measuring end-to-end performance parameters through active probing devices, and directly reading the operational status of the infrastructure through the network management interface. This multi-source data collection aims to provide a three-dimensional and multi-dimensional input for the evaluation, overcoming the one-sidedness of a single data source.

[0032] Step 2: Data Fusion and Feature Extraction After being cleaned and aligned, the collected heterogeneous raw data is mapped to a unified "spatiotemporal data coordinate system" indexed by timestamps and network locations. Subsequently, weighted fusion based on attention mechanism and graph neural network technology are used to perform correlation analysis and feature extraction on multimodal data in the same spatiotemporal context, ultimately generating a "fusion feature dataset" that can comprehensively reflect the overall quality of the network. This step transforms the raw data into high-quality, standardized features that can be understood by the model.

[0033] Step 3: Real-time Intelligent Quality Assessment The fused feature dataset obtained in the previous step is input into a pre-trained hybrid neural network evaluation model. This model can comprehensively analyze temporal and spatial features and output quantitative evaluation results in multiple dimensions, including: real-time throughput evaluation value, real-time latency satisfaction, real-time service availability index, and real-time user experience score. This realizes multi-dimensional and intelligent measurement of network service quality from transmission performance to user experience.

[0034] Step 4: Generate a comprehensive quality score: Based on the preset scoring rules, the index values ​​of each dimension obtained in the third step are standardized. The rules define the weight, threshold and non-linear scoring function for each dimension. After each dimension index is converted into an initial score by the scoring function, it is then weighted and summed according to the weight, and finally normalized to generate a single, comprehensive "dynamic quality score". This score is a quantitative summary of the current quality status of the network, providing a clear target for subsequent optimization decisions.

[0035] Step 5: Construct a digital twin simulation environment: To test optimization strategies without affecting the real network, this method constructs a digital twin simulation environment that integrates historical network quality data, current network topology data, and network service flow models. In this virtual environment, the dynamic changes of quality indicators and comprehensive scores under different network configuration parameters are safely simulated, providing a "testing ground" for strategy optimization.

[0036] Step 6: Simulation optimization to generate adjustment scheme: With the goal of improving the "dynamic quality score", strategy optimization is carried out in the twin environment constructed in the fifth step. The current dynamic quality score, network fusion characteristics and configuration status are used as inputs and are calculated by a "dynamic optimization strategy generation module" based on deep reinforcement learning. This module performs a large number of iterative trials and errors and learning in the simulation environment, and finally outputs a set of "optimal network quality adjustment scheme" for routing weight, traffic scheduling and QoS parameter configuration.

[0037] Step 7: Issue and execute adjustment instructions: The optimal adjustment scheme generated in step six is ​​parsed into a sequence of executable configuration commands for specific network devices and controllers. These real-time network configuration adjustment commands are then sent to the target network nodes and link controllers via a secure encrypted channel, and confirmation signals are received to ensure that the commands are correctly received and executed, thereby applying the optimization strategy to the actual network.

[0038] Step 8: Collect feedback data on optimization results: After the adjustment command is executed, the system immediately enters a new monitoring cycle and re-executes the data acquisition and fusion process of the first and second steps to obtain the adjusted network quality fusion feature dataset. Then, through the models and rules of the third and fourth steps, the adjusted multidimensional quality indicators and new dynamic quality scores are calculated. These adjusted data and the implemented adjustment scheme together constitute "network quality feedback data" to evaluate the actual effect of the optimization action.

[0039] Step 9: Adaptive Optimization of Model and Rules Based on the feedback data collected in step eight, the system initiates a self-learning closed loop, using the feedback data and the state data before adjustment as training samples to periodically fine-tune the three core components: fine-tuning the parameters of the real-time quality assessment model to make its assessment more realistic; calibrating the weights and thresholds of the scoring rules based on long-term optimization results; and training a deep reinforcement learning policy network with new empirical data to update its decision model. Through this step, the entire system can continuously evolve with changes in the network environment and business, constantly improving assessment accuracy and optimization efficiency.

[0040] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0041] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for real-time network quality assessment and adjustment, characterized in that, The method includes the following steps: S1. Collect real-time network quality multimodal data; S2. Perform data fusion and standardization processing on the collected multimodal data to generate a network quality fusion feature dataset; S3. Based on the network quality fusion feature dataset, use a pre-trained real-time quality assessment model to calculate and generate multi-dimensional real-time quality index data. S4. Based on the multidimensional real-time quality index data and combined with the preset real-time quality scoring rules, generate a dynamic quality score. S5. Obtain historical network quality data and network topology data, and construct a twin simulation environment for network quality assessment; S6. Based on the dynamic quality score, the network quality fusion feature dataset, and the twin simulation environment, the optimal network quality adjustment scheme is generated by using the dynamic optimization strategy generation module. S7. Based on the optimal network quality adjustment scheme, send real-time network configuration adjustment instructions to the target network nodes and links; S8. Monitor and collect network quality feedback data in real time after executing the network configuration adjustment command; S9. Based on the network quality feedback data, adaptive parameter optimization is performed on the real-time quality assessment model, the real-time quality scoring rules, and the dynamic optimization strategy generation module.

2. The real-time network quality assessment and adjustment method according to claim 1, characterized in that, The process of collecting real-time network quality multimodal data in S1 includes the following steps: S11. By deploying probes on network nodes and user terminals, network traffic data packets are collected in real time. The network traffic data packets include transport layer protocol header information, application layer payload characteristic information, and data packet arrival time sequence information. S12. Using an active detection device, send probe data packets to the target service address at a preset period and collect network performance parameters, including end-to-end latency, jitter, packet loss rate and bandwidth availability data. S13. Real-time collection of network device operating status parameters through the network infrastructure management interface, including port utilization, CPU load, memory utilization, cache queue depth, and error packet count.

3. The real-time network quality assessment and adjustment method according to claim 2, characterized in that, The process of generating the network quality fusion feature dataset in S2 includes the following steps: S21. Perform timestamp alignment and data cleaning on the network traffic data packets, the network performance parameters, and the network device operating status parameters respectively to remove outliers and noisy data. S22. Map the cleaned network traffic data packets, network performance parameters, and network device operating status parameters to a unified spatiotemporal data coordinate system, wherein the spatiotemporal data coordinate system uses absolute timestamps as the primary index and network node and link identifiers as the spatial index. S23. A feature fusion algorithm is used to fuse heterogeneous data mapped to the spatiotemporal data coordinate system. The feature fusion algorithm includes feature weighted fusion based on attention weights and feature relationship extraction based on graph neural networks to generate the network quality fusion feature dataset.

4. The real-time network quality assessment and adjustment method according to claim 3, characterized in that, The process of generating multidimensional real-time quality index data in S3 includes the following steps: S31. Input the network quality fusion feature dataset into the pre-trained real-time quality assessment model, wherein the real-time quality assessment model is a hybrid neural network model based on the combination of temporal convolutional networks and long short-term memory networks. S32. The real-time quality assessment model outputs quality assessment results in multiple dimensions, including real-time throughput assessment value, real-time latency satisfaction, real-time service availability index, and real-time user experience score, which together constitute the multi-dimensional real-time quality indicator data.

5. The real-time network quality assessment and adjustment method according to claim 4, characterized in that, The generation of dynamic quality scores in S4 includes the following steps: S41. Obtain the preset real-time quality scoring rules, which define the weight coefficient, scoring threshold, and non-linear scoring function for each dimension in the multi-dimensional real-time quality index data. S42. Substitute the real-time throughput evaluation value, the real-time latency satisfaction, the real-time service availability index, and the real-time user experience score into the nonlinear scoring function to calculate the initial scores for each dimension. S43. The initial scores of each dimension are weighted and summed according to the weight coefficients, and normalization is applied to generate the dynamic quality score.

6. The real-time network quality assessment and adjustment method according to claim 5, characterized in that, The steps for constructing a twin simulation environment for network quality assessment in S5 are as follows: S51. Extract the historical network quality data from the historical network quality database. The historical network quality data includes historical quality indicators and corresponding network status data under the same and similar network load conditions. S52. Obtain the network topology data of the current network, the network topology data including node connection relationships, link attributes and deployment information of network service function chains; S53. Based on the historical network quality data and the network topology data, construct the twin simulation environment on the digital twin platform generator. The twin simulation environment can simulate the changes of the multidimensional real-time quality index data and the dynamic quality score under different network configuration parameters.

7. The real-time network quality assessment and adjustment method according to claim 6, characterized in that, The process of generating the optimal network quality adjustment scheme in S6 includes the following steps: S61. With the goal of improving the dynamic quality score, an adjustable set of network configuration parameters is set in the twin simulation environment. The set of network configuration parameters includes routing weight, traffic scheduling strategy, service quality level identifier, congestion avoidance algorithm parameters, and bandwidth reservation ratio. S62. The dynamic quality score, the network quality fusion feature dataset, and the current network configuration status are taken as inputs, and the dynamic optimization strategy generation module performs iterative simulation and optimization in the twin simulation environment. The dynamic optimization strategy generation module adopts a decision algorithm based on deep reinforcement learning. S63. The dynamic optimization strategy generation module outputs a combination of adjustment parameters for the set of network configuration parameters, which serves as the optimal network quality adjustment scheme.

8. The real-time network quality assessment and adjustment method according to claim 7, characterized in that, Sending the real-time network configuration adjustment command in S7 includes the following steps: S71. Analyze the optimal network quality adjustment scheme and decompose it into a sequence of executable configuration commands for specific network devices; S72. The configuration command sequence is encapsulated into network configuration adjustment instructions through a secure encrypted channel and sent to the corresponding target network node controller and link controller respectively. S73. Receive and verify the instruction execution confirmation signal returned by the target network node and link.

9. The real-time network quality assessment and adjustment method according to claim 8, characterized in that, The real-time monitoring and collection of network quality feedback data in S8 includes the following steps: S81. Within one or more preset monitoring cycles after sending the network configuration adjustment instruction and receiving the instruction execution confirmation signal, S1 and S2 are re-executed to collect and generate the adjusted network quality fusion feature dataset. S82. Input the adjusted network quality fusion feature dataset into the real-time quality assessment model to generate adjusted multi-dimensional real-time quality index data. S83. Based on the adjusted multidimensional real-time quality index data and the real-time quality scoring rules, the adjusted dynamic quality score is calculated and used as a core component of the network quality feedback data.

10. The real-time network quality assessment and adjustment method according to claim 9, characterized in that, The adaptive parameter optimization in S9 includes the following steps: S91. The network quality feedback data, the network quality fusion feature dataset before adjustment, and the optimal network quality adjustment scheme are used together as training samples to form an experience replay dataset. S92. Using the experience replay dataset, periodically fine-tune the network weight parameters of the real-time quality assessment model so that its assessment results are closer to the actual network state changes. S93. Based on the long-term optimization effect reflected in the network quality feedback data, adaptively calibrate the weight coefficients and scoring thresholds in the real-time quality scoring rules. S94. Input the experience replay dataset into the deep reinforcement learning model of the dynamic optimization policy generation module for offline training and update its policy network parameters.

Images