Search

CN-122027487-A - Model construction method, device, equipment, storage medium and computer program product

CN122027487ACN 122027487 ACN122027487 ACN 122027487ACN-122027487-A

Abstract

The application discloses a model construction method which comprises the steps of obtaining a first data set of a communication network system to be modeled, wherein the first data set is related to real-time network communication of the communication network system to be modeled, carrying out model training on a model to be trained based on the first data set to obtain a first model of the communication network system to be modeled, and carrying out model derivatization processing on the first model based on the first data set to obtain a twin network performance model of the communication network system to be modeled. The application also discloses a model construction device, equipment, a storage medium and a computer program product.

Inventors

  • LI MEI
  • ZHOU CHENG
  • CHEN DANYANG
  • LI ZHIQIANG

Assignees

  • 中国移动通信有限公司研究院
  • 中国移动通信集团有限公司

Dates

Publication Date
20260512
Application Date
20241112

Claims (17)

  1. 1. A method of model construction, the method comprising: acquiring a first data set of a communication network system to be modeled, wherein the first data set is related to real-time network communication of the communication network system to be modeled; model training is carried out on the model to be trained based on the first data set, and a first model of the communication network system to be modeled is obtained; and carrying out model derivatization processing on the first model based on the first data set to obtain a twin network performance model of the communication network system to be modeled.
  2. 2. The method according to claim 1, wherein the architecture of the model to be trained belongs to a messaging neural network architecture, the model training is performed on the model to be trained based on the first data set, and a first model of the communication network system to be modeled is obtained, including: Acquiring data corresponding to the data transmission attribute from the first data set to obtain a second data set; And carrying out model training on the model to be trained by adopting the second data set to obtain a first model comprising the communication network system to be modeled, wherein the first model is used for indicating the cyclic dependency relationship among the transmission data stream, the network equipment output port queue state and the network topology link state in the communication network system to be modeled.
  3. 3. The method according to claim 2, characterized in that the message function used in the message passing phase in the model to be trained comprises at least a recurrent neural network RNN.
  4. 4. A method according to claim 3, wherein the messaging phase message delivery iterates a preset number of times.
  5. 5. A method according to claim 3, wherein the delivery message of the messaging phase comprises at least: Calculating the state of each first predicted data stream and a corresponding first predicted network topology link through the RNN to obtain a corresponding data stream to be input in the t+1st round of iteration, wherein when the state of each first predicted data stream and the corresponding first predicted network topology link is the t round of iteration, the information output by the model to be trained after the t-1 st round of iteration is updated; Calculating the accumulated sum value of each first prediction network equipment output port queue state and all the first prediction data streams through the RNN to obtain a corresponding to-be-input network equipment output port queue state in the t+1th round of iteration, wherein each first prediction network equipment output port queue state is information output by the to-be-trained model after the t-1th round of iteration when the t-1 th round of iteration is performed, and t is stepped by taking 1 as a value, and the value is taken from 1 until the value is taken to a preset number of times; and calculating the queue state of the output port of each network device to be input through the RNN to obtain the corresponding network topology link state to be input in the t+1st round of iteration.
  6. 6. The method according to claim 2, characterized in that in the readout phase of the model to be trained, the method further comprises: Calculating queuing delay and transmission delay of the first model; if the queuing delay is smaller than the transmission delay, determining a delay parameter of the first model as the transmission delay; And if the queuing delay is greater than or equal to the transmission delay, determining the delay parameter of the first model as the queuing delay.
  7. 7. The method of claim 6, wherein said calculating queuing delay and transmission delay of said first model comprises: Calculating a first accumulated sum of all of said transport data streams predicted by said first model; calculating the ratio of the first accumulated sum value to the link bandwidth feature vector of the communication network system to be modeled to obtain the queuing delay; calculating a second accumulated sum of feature vectors of the data packets included in the second data set; and calculating the ratio of the second accumulated sum value to the link bandwidth characteristic vector to obtain the transmission delay.
  8. 8. The method according to claim 1, wherein the performing model derivatization on the first model based on the first data set to obtain a twin network performance model of the communication network system to be modeled includes: acquiring data corresponding to the application scene characteristic attribute from the first data set to obtain a third data set; and carrying out model derivatization processing on the first model based on the third data set to obtain the twin network performance model.
  9. 9. The method of claim 8, wherein model derivatizing the first model based on the third data set to obtain the twin network performance model comprises: Determining p application scene characteristic parameters included in the third data set, wherein p is an integer greater than or equal to 1; Calculating mutual information parameters between each application scene characteristic parameter and other application scene characteristic parameters based on the third data set to obtain p groups of first mutual information sets; selecting target scene characteristic parameters from p application scene characteristic parameters based on p groups of the first mutual information sets; And carrying out model derivatization processing on the first model based on the target scene characteristic parameters to obtain the twin network performance model.
  10. 10. The method according to claim 9, wherein the method further comprises: determining q data attribute characteristic parameters in a second data set, wherein q is an integer greater than or equal to 1; calculating mutual information parameters between each data attribute characteristic parameter and other data attribute characteristic parameters based on the second data set to obtain q groups of second mutual information sets; Selecting target data characteristic parameters from q data attribute characteristic parameters based on q groups of the second mutual information sets; Correspondingly, the performing model derivatization processing on the first model based on the target scene characteristic parameters to obtain the twin network performance model includes: And carrying out model derivatization processing on the first model based on the target scene characteristic parameters and the target data characteristic parameters to obtain the twin network performance model.
  11. 11. The method according to claim 9 or 10, wherein calculating mutual information parameters between each of the application scene feature parameters and other application scene feature parameters based on the third data set, to obtain p groups of first mutual information sets, includes: Carrying out standardization processing on the data of each application scene characteristic parameter in the third data set to obtain a first standardized data set; and calculating mutual information parameters between each application scene characteristic parameter and other application scene characteristic parameters based on the first standardized data set to obtain p groups of first mutual information sets.
  12. 12. The method of claim 10, wherein calculating mutual information parameters between each of the data attribute feature parameters and other data attribute feature parameters based on the second data set to obtain q sets of second mutual information sets includes: carrying out standardization processing on the data of each data attribute characteristic parameter in the second data set to obtain a second standardized data set; And calculating mutual information parameters between each data attribute characteristic parameter and other data attribute characteristic parameters based on the second standardized data set to obtain q groups of second mutual information sets.
  13. 13. The method according to claim 1, wherein the method further comprises: Performing simulation analysis on a data set to be analyzed through the twin network performance model to obtain a simulation analysis performance index, wherein the data set to be analyzed is related to the communication network system to be modeled; Or outputting the twin network performance model to perform simulation analysis on the data set to be analyzed through the twin network model, so as to obtain a simulation analysis performance index.
  14. 14. The model construction device is characterized by comprising an acquisition unit, a training unit and a processing unit, wherein: The acquisition unit is used for acquiring a first data set of the communication network system to be modeled, wherein the first data set is related to real-time network communication of the communication network system to be modeled; The training unit is used for carrying out model training on the model to be trained based on the first data set to obtain a first model of the communication network system to be modeled; The processing unit is configured to perform model derivatization processing on the first model based on the first data set, so as to obtain a twin network performance model of the communication network system to be modeled.
  15. 15. An electronic device comprising a communication interface, a memory, a processor, and a communication bus, wherein: a memory for storing executable information; A communication bus for implementing communication connections between the communication interface, the processor and the memory; A processor for executing a model building program stored in a memory, implementing the steps in the model building method according to any one of claims 1 to 13.
  16. 16. A storage medium, characterized in that the storage medium has stored thereon a model building program which, when executed, is adapted to carry out the steps of the model building method according to any one of claims 1 to 13.
  17. 17. A computer program product comprising a computer program, characterized in that the computer program, when being executed by a processor, realizes the steps of the model building method according to any one of claims 1 to 13.

Description

Model construction method, device, equipment, storage medium and computer program product Technical Field The present application relates to the field of digital twin networks, and in particular, to a method, an apparatus, a device, a storage medium, and a computer program product for model construction. Background The digital twin Network (DIGITAL TWIN Network, DTN) builds a real-time mirror image of the physical Network, which can enhance the systematic simulation, optimization, verification and control capabilities that are lacking in the physical Network. The DTN can be used as a safe, economical and efficient performance evaluation environment for network operation and maintenance personnel to evaluate the network performance of various virtual scenes. Current network performance modeling methods based on artificial neural network modules, such as graph neural networks (Graph Neural Network, GNN), are developed and trained primarily using simulation data. While a simulation environment typically represents an ideal network scenario, the incoming data traffic is generated by a smooth random process and statistical features of these distributions, such as mean, variance, etc., are used as features describing the traffic. However, because the simulation environment data are used and the authenticity of the real network communication data is lacking, when the digital twin network model constructed according to the model environment is applied to the real network, the constructed model has the problems of insufficient precision, lower performance, limited scene and the like. Content of the application In order to solve the technical problems, the application provides a model construction method, a device, equipment, a storage medium and a computer program product, solves the problem that the modeling of network performance can only be developed and trained according to model data at present, provides a method for constructing a model based on actual network communication data, constructs a corresponding digital twin network model for a real communication network based on the characteristics aiming at the network communication data, and ensures the reliability of the generated digital twin network model. The technical scheme of the application is realized as follows: the application provides a model construction method, which comprises the following steps: acquiring a first data set of a communication network system to be modeled, wherein the first data set is related to real-time network communication of the communication network system to be modeled; model training is carried out on the model to be trained based on the first data set, and a first model of the communication network system to be modeled is obtained; and carrying out model derivatization processing on the first model based on the first data set to obtain a twin network performance model of the communication network system to be modeled. In the above solution, the architecture of the model to be trained belongs to a message passing neural network architecture, and the performing model training on the model to be trained based on the first data set to obtain a first model of the communication network system to be built includes: Acquiring data corresponding to the data transmission attribute from the first data set to obtain a second data set; And carrying out model training on the model to be trained by adopting the second data set to obtain a first model comprising the communication network system to be modeled, wherein the first model is used for indicating the cyclic dependency relationship among the transmission data stream, the network equipment output port queue state and the network topology link state in the communication network system to be modeled. In the above scheme, the message function used in the message transfer stage in the model to be trained at least comprises a recurrent neural network RNN. In the above scheme, the message transfer iteration in the message transfer stage is repeated for a preset number of times. In the above solution, the message delivery phase at least includes: Calculating the state of each first predicted data stream and a corresponding first predicted network topology link through the RNN to obtain a corresponding data stream to be input in the t+1st round of iteration, wherein when the state of each first predicted data stream and the corresponding first predicted network topology link is the t round of iteration, the information output by the model to be trained after the t-1 st round of iteration is updated; Calculating the queue state of each first prediction network equipment output port and the accumulated sum value of all the first prediction data flows through the RNN to obtain the queue state of the corresponding network equipment output port to be input when the t+1st round of iteration is performed, wherein the queue state of each first prediction network equipment output port is information