Network optimization method and device, equipment, storage medium and computer program product

Abstract

The invention discloses a network optimization method, the method is applied to a first node, and the method comprises the following steps: receiving a modified twin atom model sent by at least one second node; wherein the model precision of the twin atom model is greater than or equal to a first precision threshold value, and the twin atom model corresponds to a communication network application; performing combination processing on the at least one twin atom model to obtain a first twin network; and performing model optimization processing on the first twin network to obtain a second twin network of which the network precision is greater than or equal to a second precision threshold and the stability is greater than or equal to a stability threshold. The invention further discloses a network optimization device and system, a storage medium and a computer program product.

Description

Technical Field

[0001] The present application relates to the field of wireless communication technology, and in particular to a network optimization method, device, equipment, storage medium and computer program product. Background Art

[0002] With the development of communication technology, how to pre-verify the network strategy in the communication system has become a hot research topic. At present, a digital twin network technology has been proposed, which is to build a twin model based on multiple communication physical entities and arrange it to obtain the corresponding simulation environment to achieve real-time interaction and mapping with the physical entity. In this way, some network strategy pre-verification operations can be performed in the constructed communication simulation environment, namely the digital twin network, to determine the feasibility and reliability of the network strategy. In this way, how to ensure the accuracy of the constructed digital twin network is a technical problem that needs to be solved urgently.

[0003] Application Contents

[0004] In order to solve the above technical problems, the present application hopes to provide a network optimization method, device, equipment, storage medium and computer program product, which solves the problem of poor accuracy of the current digital twin network model, and proposes a method of optimizing the accuracy of a single model of the digital twin network model, and then optimizing the accuracy of the digital twin network model obtained by arranging multiple single models, thereby improving the accuracy of the digital twin network model and ensuring the reliability of the digital twin network model.

[0005] The technical solution of this application is implemented as follows:

[0006] The present application provides a network optimization method, which is applied to a first node and includes:

[0007] Receiving a corrected twin atom model sent by at least one second node; wherein the model accuracy of the twin atom model is greater than or equal to a first accuracy threshold, and the twin atom model corresponds to a communication network application;

[0008] Combining at least one of the twin atomic models to obtain a first twin network;

[0009] The first twin network is subjected to model optimization processing to obtain a second twin network having a network accuracy greater than or equal to a second accuracy threshold and a stability greater than or equal to a stability threshold.

[0010] In the above scheme, the model optimization processing is performed on the first twin network to obtain a second twin network whose network accuracy is greater than or equal to the second accuracy threshold and whose stability is greater than or equal to the stability threshold, including:

[0011] Predicting a first predicted state parameter set of at least one of the twin atomic models included in the first twin network at a first moment;

[0012] At the first moment, obtaining a first actual state parameter set of at least one of the twin atomic models included in the first twin network from at least one of the second nodes;

[0013] Based on the first predicted state parameter set and the first actual state parameter set, the first twin network model is optimized to obtain the second twin network.

[0014] In the above solution, based on the first predicted state parameter set and the first actual state parameter set, the first twin network model is optimized to obtain the second twin network, including:

[0015] Calculating the network accuracy of the first twin network based on the first predicted state parameter set and the first actual state parameter set;

[0016] If the network accuracy is greater than or equal to the second accuracy threshold, performing simulation operation on the first twin network to obtain a simulation operation result;

[0017] Based on the simulation operation result, a first probability that the network accuracy of the first twin network is greater than or equal to the second accuracy threshold is calculated;

[0018] If the first probability is greater than or equal to the stability threshold, the second twin network is determined to be the first twin network.

[0019] In the above solution, the calculating the network accuracy of the first twin network based on the first predicted state parameter set and the first actual state parameter set includes:

[0020] Determining static evaluation parameters, dynamic evaluation parameters, and associated evaluation parameters based on the first predicted state parameter set and the first actual state parameter set;

[0021] Based on the static evaluation parameters, the dynamic evaluation parameters and the associated evaluation parameters, the network accuracy of the first twin network is obtained.

[0022] In the above solution, the determining of static evaluation parameters, dynamic evaluation parameters and associated evaluation parameters based on the first predicted state parameter set and the first actual state parameter set includes:

[0023] Grouping the first prediction state parameter set to obtain a first static parameter set and a first dynamic parameter set;

[0024] Grouping the first actual state parameter set to obtain a second static parameter set and a second dynamic parameter set;

[0025] Performing similarity calculation on the first static parameter set and the second static parameter set to obtain the static evaluation parameter;

[0026] Determining the dynamic evaluation parameter based on the first dynamic parameter set and the second dynamic parameter set;

[0027] Based on the first predicted state parameter set, determining a parameter association relationship of at least one of the twin atom models to obtain a first association relationship parameter;

[0028] Based on the first actual state parameter set, determining a parameter association relationship between at least one physical entity node corresponding to the twin atom model to obtain a second association relationship parameter;

[0029] The association evaluation parameter is determined based on the first association relationship parameter and the second association relationship parameter.

[0030] In the above solution, determining the dynamic evaluation parameter based on the first dynamic parameter set and the second dynamic parameter set includes:

[0031] Based on the first dynamic parameter set, determining a change trend of each dynamic parameter to obtain a first dynamic change parameter set;

[0032] Based on the second dynamic parameter set, determining a change trend of each dynamic parameter to obtain a second dynamic change parameter set;

[0033] A consistency calculation is performed on the first dynamic parameter set, the second dynamic parameter set, the first dynamically changing parameter set, and the second dynamically changing parameter set to determine the dynamic evaluation parameter.

[0034] In the above solution, the performing consistency calculation on the first dynamic parameter set, the second dynamic parameter set, the first dynamically changing parameter set, and the second dynamically changing parameter set to determine the dynamic evaluation parameter includes:

[0035] Performing consistency calculation on the first dynamic parameter set and the second dynamic parameter set to obtain a first consistency parameter;

[0036] Performing consistency calculation on the first dynamically changing parameter set and the second dynamically changing parameter set to obtain a second consistency parameter;

[0037] Calculating a first product of a first weight coefficient and the first consistency parameter;

[0038] Calculating a second product of a second weight coefficient and the second consistency parameter;

[0039] The sum of the first product and the second product is calculated to obtain the dynamic evaluation parameter.

[0040] In the above solution, obtaining the network accuracy of the first twin network based on the static evaluation parameter, the dynamic evaluation parameter and the associated evaluation parameter includes:

[0041] Determine static weight coefficient, dynamic weight coefficient and associated evaluation weight coefficient;

[0042] Calculating the product of the static weight coefficient and the static evaluation parameter to obtain a weighted static evaluation parameter;

[0043] Calculating the product of the dynamic weight coefficient and the dynamic evaluation parameter to obtain a weighted dynamic evaluation parameter;

[0044] Calculating the product of the association evaluation weight coefficient and the association evaluation parameter to obtain a weighted association evaluation parameter;

[0045] Calculating the cumulative sum of the weighted static evaluation parameter, the weighted dynamic evaluation parameter and the weighted associated evaluation parameter to obtain a comprehensive parameter;

[0046] Calculating a first value of the static evaluation parameter;

[0047] Calculating a second value of the dynamic evaluation parameter;

[0048] Calculating a third value of the associated evaluation parameter;

[0049] Calculating the first value, the second value, and the third value to obtain a comprehensive value;

[0050] Calculating a fourth value of a ratio of the comprehensive parameter to the comprehensive value;

[0051] The network accuracy is obtained by calculating a ratio of the fourth value to the square root value of the number of parameters included in the comprehensive parameter.

[0052] In the above scheme, the method further includes:

[0053] If the network accuracy is less than the second accuracy threshold, or the first probability is less than the stability threshold, based on the overall accuracy threshold of the twin network and the probability threshold of the twin network accuracy reaching, determine the first accuracy threshold of each of the twin atomic models;

[0054] The first precision threshold of the twin atom model corresponding to each second node is sent to the corresponding second node, so that the second node corrects the corresponding twin atom model based on the first precision threshold.

[0055] In the above scheme, if the network accuracy is less than the second accuracy threshold, or the first probability is less than the stability threshold, based on the overall accuracy threshold of the twin network and the probability threshold of the twin network accuracy reaching, the first accuracy threshold of each of the twin atomic models is determined, including:

[0056] If the network accuracy is less than the second accuracy threshold, or the first probability is less than the stability threshold, a second probability that the model accuracy obtained by simulating and calculating the first twin network is greater than or equal to the overall accuracy threshold of the twin network is calculated within a preset time period;

[0057] If the second probability is greater than or equal to the twin network accuracy reaching probability threshold, determine the first accuracy threshold of each of the twin atom models as the current accuracy threshold of the corresponding twin atom model;

[0058] If the second probability is less than the twin network accuracy probability threshold, the first accuracy threshold of each of the twin atomic models is determined based on the overall accuracy threshold of the twin network, the twin network accuracy probability threshold and the first twin network.

[0059] In the above scheme, if the second probability is less than the twin network accuracy probability threshold, based on the twin network overall accuracy threshold, the twin network accuracy probability threshold and the first twin network, the first accuracy threshold of each twin atomic model is determined, including:

[0060] If the second probability is less than the probability threshold of the twin network accuracy, the conditional probability that the model accuracy of the first twin network is greater than or equal to the overall accuracy threshold of the twin network is calculated when the model accuracy is greater than or equal to the current accuracy threshold of each of the twin atomic models;

[0061] Calculating a priori probability when the model accuracy is greater than or equal to a current accuracy threshold of each of the twin atom models;

[0062] Based on the conditional probability and the prior probability, determine the posterior probability when the model accuracy is greater than or equal to the current accuracy threshold;

[0063] Optimizing the current precision threshold of each of the twin atom models to obtain an optimized precision threshold of each of the twin atom models;

[0064] After updating the current accuracy threshold of each of the twin atom models to the corresponding optimized accuracy threshold, repeatedly executing the step of calculating the conditional probability that the model accuracy of the first twin network is greater than or equal to the overall accuracy threshold of the twin network when the model accuracy is greater than or equal to the current accuracy threshold of each of the twin atom models, until the maximum a posteriori probability is determined;

[0065] Determine the first precision threshold of each of the twin atom models as the current precision threshold corresponding to the maximum a posteriori probability.

[0066] The embodiment of the present application provides a network optimization method, which is applied to a second node and includes:

[0067] Obtaining a first accuracy threshold of the atomic model to be corrected;

[0068] Determining the model accuracy of the atomic model to be corrected; wherein the atomic model to be corrected corresponds to a communication network application;

[0069] Based on the model accuracy and the first accuracy threshold, the to-be-corrected atomic model is subjected to model correction to obtain a twin atom model; wherein the model accuracy of the twin atom model is greater than or equal to the first accuracy threshold;

[0070] The twin atomic model is sent to a first node, so that the first node builds a corresponding first twin network based on the twin atomic model.

[0071] In the above solution, the step of obtaining the first accuracy threshold of the atomic model to be corrected includes:

[0072] The first precision threshold is determined as an initialization precision threshold.

[0073] In the above solution, the step of obtaining the first accuracy threshold of the atomic model to be corrected includes:

[0074] Receiving the first accuracy threshold sent by the first node; wherein the first accuracy threshold is an accuracy threshold corresponding to the twin atom model;

[0075] Correspondingly, after obtaining the first accuracy threshold of the atomic model to be corrected, the method further includes:

[0076] After updating the to-be-corrected atomic model to the twin atomic model, the step of determining the model accuracy of the to-be-corrected atomic model is repeated until the first accuracy threshold sent by the first node is not received.

[0077] In the above scheme, the step of determining the model accuracy of the atomic model to be corrected includes:

[0078] Predicting a second predicted state parameter set of the to-be-corrected atomic model at a second moment;

[0079] At the second moment, obtaining a second set of actual state parameters of the to-be-corrected atomic model;

[0080] The model accuracy of the atomic model to be corrected is calculated based on the second predicted state parameter set and the second actual state parameter set.

[0081] In the above solution, based on the model accuracy and the first accuracy threshold, the to-be-corrected atomic model is corrected to obtain a twin atom model, including:

[0082] If the model accuracy is greater than or equal to the first accuracy threshold, determining that the twin atom model is the atom model to be corrected;

[0083] If the model accuracy is less than the first accuracy threshold, calculating the difference between the first accuracy threshold and the model accuracy;

[0084] Calculating a third product of the difference and a regularization coefficient;

[0085] Calculating the sum of the loss value of the atomic model to be corrected and the third product to obtain an adjustment parameter;

[0086] Reversely adjusting the adjustable parameters in the to-be-corrected atomic model based on the adjustment parameters to obtain an intermediate atomic model;

[0087] The atomic model to be corrected is updated to the intermediate atomic model, and the steps of determining the model accuracy of the atomic model to be corrected are repeated until the twin atomic model is obtained.

[0088] In the above scheme, the method further includes:

[0089] Receiving terminal communication indicator data sent by the terminal device;

[0090] Receive a twin task request sent by the first node; wherein the twin task request includes at least: a model type, an overall accuracy threshold of the twin network, and a probability threshold of the twin network accuracy reaching;

[0091] At least one atomic model to be corrected is generated based on the terminal communication indicator data and the model type.

[0092] The embodiment of the present application provides a first network optimization device, which is applied to a first node. The device includes: a receiving unit, a first processing unit, and a second processing unit; wherein:

[0093] The receiving unit is configured to receive a corrected twin atom model sent by at least one second node; wherein the model accuracy of the twin atom model is greater than or equal to a first accuracy threshold, and the twin atom model corresponds to a communication network application;

[0094] The first processing unit is used to perform combination processing on at least one of the twin atomic models to obtain a first twin network;

[0095] The second processing unit is used to perform model optimization processing on the first twin network to obtain a second twin network whose network accuracy is greater than or equal to a second accuracy threshold and whose stability is greater than or equal to a stability threshold.

[0096] The embodiment of the present application provides a second network optimization device, which is applied to a second node. The device includes: an acquisition unit, a determination unit, a correction unit, and a sending unit; wherein:

[0097] The acquisition unit is used to acquire a first accuracy threshold of the atomic model to be corrected;

[0098] The determining unit is used to determine the model accuracy of the atomic model to be corrected; wherein the atomic model to be corrected corresponds to a communication network application;

[0099] The correction unit is used to perform model correction on the atomic model to be corrected based on the model accuracy and the first accuracy threshold to obtain a twin atom model;

[0100] The sending unit is used to send the twin atom model to the first node.

[0101] The embodiment of the present application provides a network optimization system, the system comprising at least: a first node and at least one second node; wherein:

[0102] The first node is used to implement the steps of the network optimization method as described in any one of the above items;

[0103] The second node is used to implement the steps of the network optimization method as described in any one of the above items.

[0104] An embodiment of the present application provides a storage medium, on which a resource conflict indication program is stored. When the resource conflict indication program is executed, it is used to implement the steps of any of the network optimization methods described above.

[0105] An embodiment of the present application provides a computer program product, including a computer program, which implements the steps of any of the above network optimization methods when executed by a processor.

[0106] The embodiments of the present application provide a network optimization method, apparatus, device, storage medium and computer program product. After obtaining the first accuracy threshold of the atomic model to be corrected through the second node, the model accuracy of the atomic model to be corrected is determined, and based on the model accuracy and the first accuracy threshold, the atomic model to be corrected is corrected. After obtaining the twin atomic model, the twin atomic model is sent to the first node. The first node receives the corrected twin atomic model sent by at least one second node, combines at least one twin atomic model to obtain a first twin network, and performs model optimization processing on the first twin network to obtain a second twin network with a network accuracy greater than or equal to the second accuracy threshold and stability greater than or equal to the stability threshold. In this way, the second node corrects the atomic model to be corrected according to the first accuracy threshold, and sends the corrected twin atomic model to the first node, so that the first node combines and processes the corrected twin atomic model sent by at least one second node according to network requirements to obtain a first twin network, and finally optimizes the model of the first twin network to obtain a second twin network with a network accuracy greater than or equal to the second accuracy threshold. In this way, through two-level model optimization, and different levels of model optimization are optimized at different nodes, the requirements on the computing power of the nodes are reduced, and the current problem of poor accuracy of digital twin network models is solved. A method is proposed to optimize the accuracy of a single model of the digital twin network model, and then optimize the accuracy of the digital twin network model obtained by arranging multiple single models, thereby improving the accuracy of the digital twin network model and ensuring the reliability of the digital twin network model. BRIEF DESCRIPTION OF THE DRAWINGS

[0107] Figure 1 A schematic diagram of a network optimization method provided in an embodiment of the present application Figure 1 ;

[0108] Figure 2 A schematic diagram of a network optimization method provided in an embodiment of the present application Figure 2 ;

[0109] Figure 3 A schematic diagram of a network optimization method provided in an embodiment of the present application Figure 3 ;

[0110] Figure 4 A schematic diagram of a network optimization method provided in an embodiment of the present application Figure 4 ;

[0111] Figure 5 A schematic diagram of an application implementation example provided for the present application;

[0112] Figure 6 A schematic diagram of the structure of a first network optimization device provided in an embodiment of the present application;

[0113] Figure 7 A schematic diagram of the structure of a second network optimization device provided in an embodiment of the present application;

[0114] Figure 8 A schematic diagram of the structure of a network optimization system provided in an embodiment of the present application. DETAILED DESCRIPTION

[0115] The technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application.

[0116] The embodiment of the present application provides a network optimization method, referring to Figure 1 As shown, the method is applied to the first node, and the method comprises the following steps:

[0117] Step 101: Receive a corrected twin atom model sent by at least one second node.

[0118] Among them, the model accuracy of the twin atom model is greater than or equal to a first accuracy threshold, and the twin atom model corresponds to a communication network application.

[0119] In an embodiment of the present application, the first node is a management node for service management and orchestration functions, for example, it can be a server node, or a computing node. The second node is a network service node with communication network services, for example, it can be a base station (g-NB) node. The twin atomic model corresponds to each communication network application in the communication network. The relationship between at least one second node is that at least one second node can provide a twin atomic module related to the first twin network. The first accuracy threshold is an empirical value of model accuracy obtained based on a large number of experiments, and it can also be an empirical value of model accuracy set according to actual requirements for the twin atomic model.

[0120] Step 102: Combining and processing at least one twin atom model to obtain a first twin network.

[0121] In an embodiment of the present application, the first node performs model combination and orchestration processing on at least one received twin atomic model to achieve model fusion, thereby obtaining a first twin network.

[0122] Step 103: Perform model optimization processing on the first twin network to obtain a second twin network whose network accuracy is greater than or equal to a second accuracy threshold.

[0123] In the embodiment of the present application, the first node performs model optimization processing on the first twin network obtained by the combination processing according to the model optimization requirements, and obtains a second twin network whose network accuracy is greater than or equal to the second accuracy threshold, that is, the second twin network is a twin network obtained by model optimization of the first twin network. The second accuracy threshold is an empirical value of network accuracy requirements obtained from a large number of experiments, and can also be an empirical value of network accuracy requirements set according to actual application requirements.

[0124] The network optimization method provided in the embodiment of the present application receives a twin atomic model sent by at least one second node through a first node, which is corrected according to a first accuracy threshold dynamically determined by the second node, and combines at least one twin atomic model to obtain a first twin network, and then performs model optimization processing on the first twin network to obtain a second twin network with a network accuracy greater than or equal to a second accuracy threshold and stability greater than or equal to a stability threshold. In this way, the second node dynamically determines the first accuracy threshold of the atomic model to be corrected according to the overall accuracy threshold of the twin network corresponding to the twin network corresponding to the atomic model to be corrected and the probability threshold of the twin network accuracy. Then, the atomic model to be corrected is corrected according to the first accuracy threshold, and the corrected twin atomic model is sent to the first node, so that the first node combines and processes the corrected twin atomic model sent by at least one second node according to network requirements to obtain a first twin network. Finally, the first twin network is model optimized to obtain a second twin network whose network accuracy is greater than or equal to the second accuracy threshold. In this way, through two-level model optimization, and different levels of model optimization are optimized at different nodes, the requirements on the computing power of the nodes are reduced, and the problem of poor accuracy of the current digital twin network model is solved. A method is proposed to optimize the accuracy of a single model of the digital twin network model, and then optimize the accuracy of the digital twin network model obtained by arranging multiple single models, thereby improving the accuracy of the digital twin network model and ensuring the reliability of the digital twin network model.

[0125] Based on the above embodiments, the embodiments of the present application provide a network optimization method, referring to Figure 2 As shown, the method is applied to the second node, and the method includes the following steps:

[0126] Step 201: Obtain a first accuracy threshold of the atomic model to be corrected.

[0127] In the embodiment of the present application, the first accuracy threshold of the atomic model to be corrected may be preset or determined according to actual application requirements.

[0128] Step 202: Determine the model accuracy of the atomic model to be corrected.

[0129] Among them, the atomic model to be modified corresponds to the communication network application.

[0130] In the embodiment of the present application, the second node determines the atomic model to be corrected included therein, and determines the model accuracy of the atomic model to be corrected according to the relevant characteristics of the second node. The model accuracy is used to indicate the reliability of the results of the model to be corrected in the simulation application process.

[0131] Step 203: Based on the model accuracy and the first accuracy threshold, the atomic model to be corrected is corrected to obtain a twin atom model.

[0132] Among them, the model accuracy of the twin atom model is greater than or equal to a first accuracy threshold.

[0133] In an embodiment of the present application, a comparison and analysis is performed between the model accuracy of the atomic model to be corrected and the first accuracy threshold to obtain an analysis result, and the atomic model to be corrected is corrected according to the obtained analysis result until a twin atomic model having a model accuracy greater than or equal to the first accuracy threshold is obtained.

[0134] Step 204: Send the twin atom model to the first node.

[0135] Among them, the twin atomic model is sent to the first node so that the first node establishes a corresponding first twin network based on the twin atomic model.

[0136] In an embodiment of the present application, the second node sends the corrected twin atom model to the first node through a communication method agreed upon with the first node, so that the first node performs subsequent related analysis operations based on the received twin atom model.

[0137] The network optimization method provided in the embodiment of the present application obtains the first accuracy threshold of the atomic model to be corrected through the second node, and determines the model accuracy of the atomic model to be corrected, then performs model correction on the atomic model to be corrected based on the model accuracy and the first accuracy threshold. After obtaining the twin atomic model, the twin atomic model is sent to the first node, so that the first node receives the corrected twin atomic model sent by at least one second node, combines at least one twin atomic model to obtain a first twin network, and performs model optimization processing on the first twin network to obtain a second twin network whose network accuracy is greater than or equal to the second accuracy threshold. In this way, the second node corrects the atomic model to be corrected according to the first accuracy threshold, and sends the corrected twin atomic model to the first node, so that the first node combines and processes the corrected twin atomic model sent by at least one second node according to network requirements to obtain a first twin network, and finally optimizes the model of the first twin network to obtain a second twin network with a network accuracy greater than or equal to the second accuracy threshold. In this way, through two-level model optimization, and different levels of model optimization are optimized at different nodes, the requirements on the computing power of the nodes are reduced, and the current problem of poor accuracy of digital twin network models is solved. A method is proposed to optimize the accuracy of a single model of the digital twin network model, and then optimize the accuracy of the digital twin network model obtained by arranging multiple single models, thereby improving the accuracy of the digital twin network model and ensuring the reliability of the digital twin network model.

[0138] Based on the above embodiments, the embodiments of the present application provide a network optimization method, referring to Figure 3 As shown, the method comprises the following steps:

[0139] Step 301: The second node determines a first accuracy threshold of the atomic model to be corrected.

[0140] Step 302: The second node determines the model accuracy of the atomic model to be corrected.

[0141] Among them, the atomic model to be modified corresponds to the communication network application.

[0142] In the embodiment of the present application, the second node is taken as a base station for example. The base station measures the accuracy of the atomic model to be corrected, and obtains the model accuracy of the atomic model to be corrected.

[0143] Step 303: The second node performs model correction on the atomic model to be corrected based on the model accuracy and the first accuracy threshold to obtain a twin atom model.

[0144] In an embodiment of the present application, the base station uses a model correction method to correct the atomic model to be corrected according to the relationship between the model accuracy of the atomic model to be corrected and the first accuracy threshold to obtain a twin atomic model.

[0145] Step 304: The second node sends the twin atom model to the first node.

[0146] Among them, the twin atomic model is sent to the first node so that the first node establishes a corresponding first twin network based on the twin atomic model.

[0147] Step 305: The first node receives a corrected twin atom model sent by at least one second node.

[0148] Among them, the model accuracy of the twin atom model is greater than or equal to a first accuracy threshold, and the twin atom model corresponds to a communication network application.

[0149] Step 306: The first node performs combination processing on at least one twin atomic model to obtain a first twin network.

[0150] In an embodiment of the present application, taking the first node as a service management server node with management orchestration functions as an example, the service management server node receives a corrected twin atomic model sent by at least one second node, obtains at least one twin atomic model, and then performs model orchestration and combination processing on at least one twin atomic model to obtain a first twin network.

[0151] In some application scenarios, the first node can first perform similar model aggregation processing on at least one twin atomic model to obtain a new model belonging to the same category with strong universality, and then arrange and combine the new models obtained by aggregating these similar models to obtain the first twin network.

[0152] Step 307: The first node performs model optimization processing on the first twin network to obtain a second twin network whose network accuracy is greater than or equal to the second accuracy threshold and whose stability is greater than or equal to the stability threshold.

[0153] In an embodiment of the present application, the service management server node performs model optimization processing on the first twin network obtained by the combination processing until a second twin network is obtained whose network accuracy is greater than or equal to a second accuracy threshold and whose stability is greater than or equal to a stability threshold.

[0154] Based on the above embodiments, in other embodiments of the present application, step 301 may be implemented by step 301a or step 301b:

[0155] Step 301a: The second node determines the first accuracy threshold as the initialization accuracy threshold.

[0156] In the embodiment of the present application, when the second node obtains the atomic model to be corrected for the first time, the first accuracy threshold of the atomic model to be corrected is determined as the initialization accuracy threshold. The initialization accuracy threshold is an empirical value obtained from a large number of experiments, and in some application scenarios, it can be set according to actual conditions.

[0157] Step 301b: The second node receives the first accuracy threshold sent by the first node.

[0158] Among them, the first accuracy threshold is the accuracy threshold corresponding to the twin atom model.

[0159] In an embodiment of the present application, the first accuracy threshold sent by the first node can be obtained by dynamically setting and allocating the first node according to the twin network including the atomic model to be corrected. In this way, the accuracy requirements of the twin atomic model can be guaranteed, and then the accuracy requirements of the corresponding twin network can be guaranteed.

[0160] Correspondingly, after the second node executes step 301b, it is also used to execute the following steps: after updating the atomic model to be corrected to a twin atomic model, execute the step of "determining the model accuracy of the atomic model to be corrected" until the first accuracy threshold sent by the first node is not received.

[0161] In an embodiment of the present application, when the second twin network obtained by the first node after optimizing the first twin network does not meet the requirements, the model accuracy of at least one twin atomic model is readjusted for the corresponding at least one twin atomic model according to the requirements such as the model accuracy of the twin network model, so that the model accuracy of each twin atomic model changes dynamically in different analysis processes, and by controlling the model accuracy of each twin atomic model, the network accuracy and other requirements of the corresponding twin network can be quickly guaranteed.

[0162] Based on the above embodiments, in other embodiments of the present application, step 302 may be implemented by steps 302a to 302c:

[0163] Step 302a: The second node predicts a second set of predicted state parameters of the atomic model to be corrected at a second moment.

[0164] In the embodiment of the present application, the second node performs prediction processing based on the atomic model to be corrected, determines the relevant state parameters corresponding to the atomic model to be corrected at a second moment in the future, and obtains a second predicted state parameter set.

[0165] The second moment may be, for example, a moment corresponding to a period of time from the current moment, such as half an hour, one hour, or one day from the current moment. The specific setting may be based on actual conditions and needs, and is not specifically limited here.

[0166] Step 302b: The second node obtains a second set of actual state parameters of the to-be-corrected atomic model at a second moment.

[0167] In the embodiment of the present application, when the second moment arrives, the second node obtains the actual state parameters corresponding to the model corresponding to the atomic model to be corrected during the actual operation process, and obtains a second actual state parameter set.

[0168] When the second moment arrives, the base station obtains the actual state parameters corresponding to the communication network application corresponding to the atomic model to be corrected during the actual operation, thereby obtaining a second actual state parameter set of the atomic model to be corrected.

[0169] Step 302c: The second node calculates the model accuracy of the atomic model to be corrected based on the second predicted state parameter set and the second actual state parameter set.

[0170] In an embodiment of the present application, the second node performs an accuracy analysis of the atomic model to be corrected according to a second preset state parameter set and a second actual state parameter set of the atomic model to be corrected to obtain the model accuracy of the atomic model to be corrected.

[0171] It should be noted that the specific implementation process of step 302c can refer to the subsequent process of determining the network accuracy of the first twin network based on the first preset state parameter set and the first actual state parameter set, which will not be repeated here.

[0172] Based on the above embodiments, in other embodiments of the present application, step 303 may be implemented by step 303a, or steps 303b to 303f:

[0173] Step 303a: If the model accuracy is greater than or equal to the first accuracy threshold, the second node determines that the twin atom model is the atom model to be corrected.

[0174] In an embodiment of the present application, when the model accuracy of the atomic model to be corrected is greater than or equal to the first accuracy threshold, it indicates that the atomic model to be corrected meets the current requirements. Therefore, it can be determined that the twin atomic model is the atomic model to be corrected itself.

[0175] Step 303b: If the model accuracy is less than the first accuracy threshold, the second node calculates the difference between the first accuracy threshold and the model accuracy.

[0176] In an embodiment of the present application, when the model accuracy of the atomic model to be corrected is less than the first accuracy threshold, it indicates that the atomic model to be corrected does not currently meet the requirements and the atomic model to be corrected needs to be corrected. At this time, the second node calculates the first accuracy threshold and the model accuracy, for example, calculates the difference between the first accuracy threshold and the model accuracy.

[0177] Step 303c: The second node calculates the third product of the difference and the regularization coefficient.

[0178] In the embodiment of the present application, the regularization coefficient is an empirical value obtained from a large number of experiments, or an empirical value set according to experience based on actual needs. The third product of the difference and the regularization coefficient can be calculated using the following formula: third product = difference * regularization coefficient.

[0179] Step 303d: The second node calculates the sum of the loss value of the atomic model to be corrected and the third product to obtain an adjustment parameter.

[0180] In an embodiment of the present application, after the second node uses the loss function corresponding to the atomic model to be corrected to calculate the loss value of the atomic model to be corrected, it continues to calculate the sum of the loss value of the atomic model to be corrected and the third product, thereby obtaining adjustment parameters for model correction of the atomic model to be corrected.

[0181] Step 303e: The second node reversely adjusts the adjustable parameters in the atomic model to be corrected based on the adjustment parameters to obtain an intermediate atomic model.

[0182] In the embodiment of the present application, the second node uses the calculated adjustment parameters to reversely adjust the adjustable parameters in the atomic model to be corrected to obtain an intermediate atomic model.

[0183] Step 303f, updating the atomic model to be corrected to an intermediate atomic model, and repeating the steps to determine the model accuracy of the atomic model to be corrected until a twin atomic model is obtained.

[0184] In an embodiment of the present application, after the atomic model to be corrected is updated to an intermediate atomic model, the process of step 302 (including steps 302a to 302c) and step 303 (including steps 303a to 303f) are repeated until a twin atom model with a model accuracy greater than a first accuracy threshold is obtained.

[0185] Based on the above embodiments, in other embodiments of the present application, refer to Figure 4 As shown, the second node is also used to execute steps 308 to 310:

[0186] Step 308: The second node receives the terminal communication indicator data sent by the terminal device.

[0187] In the embodiment of the present application, the terminal device is a device that uses the communication network service provided by the second node, for example, it can be an intelligent mobile terminal device that uses the communication network service, for example, it can be a smart phone, a smart car and other devices. The terminal communication indicator data is some parameters obtained by measuring the communication network service during the use of the communication network service by the terminal device, for example, it can be a channel quality parameter.

[0188] Step 309: The second node receives the twin task request sent by the first node.

[0189] Among them, the twin task request includes at least: model type, twin network overall accuracy threshold and twin network accuracy reaching probability threshold.

[0190] In an embodiment of the present application, the first node sends a twin task request to the second node to request the second node to provide a corresponding revised twin atomic model.

[0191] Step 310: The second node generates at least one atomic model to be corrected based on the terminal communication indicator data and the model type.

[0192] In an embodiment of the present application, the second node creates at least one atomic model to be corrected that matches the twin task request according to the terminal communication indicator data and data type reported by the terminal device. Further, when the second node creates at least one atomic model to be corrected, it can also refer to the node feature data of the second node to perform the operation of creating at least one atomic model to be corrected. At this time, the second node obtains the atomic model to be corrected for the first time. After some application scenarios, the second node executes step 310 and then executes step 301a. After obtaining the twin atomic model, it is sent to the first node, and after the first node obtains the first twin network according to the received twin atomic model combination processing, if the second twin network is not obtained, the first node dynamically determines the first precision threshold of the twin atomic model and sends it to the second node. The second node updates the corresponding atomic model to be corrected to the corresponding twin atomic model, and updates its precision threshold to the first precision threshold, and then updates the updated atomic model to be corrected according to the first precision threshold sent by the first node. Model update, obtain a twin atomic model, so that the first node finally builds a second twin network that meets the requirements according to the twin atomic model provided by the second node.

[0193] Based on the above embodiments, in other embodiments of the present application, step 307 can be implemented by steps 307a to 307c:

[0194] Step 307a: The first node predicts a first predicted state parameter set of at least one twin atomic model included in the first twin network at a first moment.

[0195] In an embodiment of the present application, the first node predicts a first predicted state parameter set of at least one twin atomic model included in the first twin network at a first moment in the future.

[0196] Step 307b: At a first moment, the first node obtains a first actual state parameter set of at least one twin atomic model included in the first twin network from at least one second node.

[0197] In an embodiment of the present application, when the first moment is reached, actual state parameters of at least one twin atom model included in each second node are obtained from each second node, thereby obtaining a first actual state parameter set.

[0198] Step 307c: The first node performs model optimization processing on the first twin network model based on the first predicted state parameter set and the first actual state parameter set to obtain a second twin network.

[0199] In an embodiment of the present application, the first node performs calculation and analysis on the first and predicted state parameter sets and the first actual state parameter set to obtain calculation and analysis results, and performs model optimization processing on the first twin network model according to the calculation and analysis results until a second twin network that meets the requirements is obtained.

[0200] Based on the above embodiment, in other embodiments of the present application, step 307c can be implemented by steps a11 to a15:

[0201] Step a11: The first node calculates the network accuracy of the first twin network based on the first predicted state parameter set and the first actual state parameter set.

[0202] In an embodiment of the present application, the first node calculates a first preset state parameter set and a first actual state parameter set to obtain a network accuracy of the first twin network.

[0203] After the first node executes step a11, it can choose to execute steps a12 to a13 or steps a14 to a15. If the network accuracy is greater than or equal to the second accuracy threshold, choose to execute steps a12 to a13; if the network accuracy is less than the second accuracy threshold, or the first probability is less than the stability threshold, choose to execute step a15.

[0204] Step a12: If the network accuracy is greater than or equal to the second accuracy threshold, a simulation operation is performed on the first twin network to obtain a simulation operation result.

[0205] Step a13: Based on the simulation calculation results, a first probability that the network accuracy of the first twin network is greater than or equal to the second accuracy threshold is calculated.

[0206] In an embodiment of the present application, when the network accuracy of the first twin network is greater than or equal to the second accuracy threshold, the stability of the first twin network is statistically analyzed, that is, the probability value that the network accuracy is greater than or equal to the second accuracy threshold is statistically analyzed to obtain a first probability.

[0207] Step a14: If the first probability is greater than or equal to the stability threshold, determine that the second twin network is the first twin network.

[0208] In an embodiment of the present application, when the first probability is greater than or equal to the stability threshold, it indicates that the stability of the first twin network is higher. At this time, the network accuracy of the first twin network is greater than or equal to the second accuracy threshold, and the first probability is greater than or equal to the stability threshold, indicating that the network accuracy and stability of the first twin network meet the requirements. At this time, the second twin network can be determined to be the first twin network.

[0209] Step a15: If the network accuracy is less than the second accuracy threshold, or the first probability is less than the stability threshold, the first node determines the first accuracy threshold of each twin atomic model based on the overall accuracy threshold of the twin network and the probability threshold of the twin network accuracy.

[0210] In the embodiment of the present application, when the network accuracy is less than the second accuracy threshold, or the first accuracy is less than the stability threshold, the first node determines the overall accuracy threshold of the twin network set for the twin network and the probability threshold of the twin network accuracy, and then determines the first accuracy threshold of each twin atomic model according to the overall accuracy threshold of the twin network and the probability threshold of the twin network accuracy. The accuracy threshold of the twin atomic model is dynamically determined.

[0211] Based on the above embodiment, in other embodiments of the present application, step a11 can be implemented by steps a111 to a112:

[0212] Step a111: The first node determines static evaluation parameters, dynamic evaluation parameters, and associated evaluation parameters based on the first predicted state parameter set and the first actual state parameter set.

[0213] In the embodiment of the present application, the static evaluation parameters of the first twin network are obtained by analyzing and calculating the static parameters representing the static attributes and indicators of the first twin network, wherein the static parameters are the inherent attributes of the first twin network, which do not change with the network state of the first twin network. The dynamic evaluation parameters of the first twin network are obtained by analyzing the dynamic parameters representing the dynamic attributes and indicators of the first twin network, wherein the dynamic parameters change with the change of the network state. The associated evaluation parameters of the first twin network are obtained by analyzing and calculating the explicit or implicit relationship description parameters between the parameters representing the first twin network and the associated network node parameters, that is, the associated relationship parameters. The first node classifies the parameters in the first predicted state parameter set and the first actual state parameter set, obtains the corresponding static parameters, dynamic parameters and associated relationship parameters, calculates the static parameters, obtains the static evaluation parameters, and calculates the dynamic parameters to obtain the dynamic evaluation parameters. The first node analyzes the first predicted state parameter set and the second actual state parameter set, determines the corresponding associated relationship parameters, and calculates to obtain the associated evaluation parameters.

[0214] Step a112: The first node obtains the network accuracy of the first twin network based on the static evaluation parameters, the dynamic evaluation parameters and the associated evaluation parameters.

[0215] In an embodiment of the present application, the first node uses a preset calculation and analysis method to perform calculation and analysis on the calculated static evaluation parameters, dynamic evaluation parameters and associated evaluation parameters to obtain the network accuracy of the first twin network.

[0216] Based on the above embodiments, in other embodiments of the present application, step a111 can be implemented by steps b11 to b17:

[0217] Step b11: The first node groups the first prediction state parameter set to obtain a first static parameter set and a first dynamic parameter set.

[0218] In an embodiment of the present application, the first node performs dynamic parameter and static parameter grouping processing on the first prediction state parameter set to obtain a first static parameter set and a first dynamic parameter set.

[0219] Step b12: The first node groups the first actual state parameter set to obtain a second static parameter set and a second dynamic parameter set.

[0220] In an embodiment of the present application, the first node performs grouping processing of dynamic parameters and static parameters on the first actual state parameter set to obtain a second static parameter set and a second dynamic parameter set.

[0221] Step b13: The first node performs similarity calculation on the first static parameter set and the second static parameter set to obtain static evaluation parameters.

[0222] In an embodiment of the present application, the first node uses a similarity calculation method to perform similarity calculation on the first static parameter set and the second static parameter set to obtain static evaluation parameters.

[0223] Step b14: The first node determines a dynamic evaluation parameter based on the first dynamic parameter set and the second dynamic parameter set.

[0224] In the embodiment of the present application, there are two analysis indicators, namely, parameters and parameter changes, in the dynamic parameter change process. Therefore, the first dynamic parameter set and the second dynamic parameter set are calculated and analyzed from the two analysis indicator directions to obtain dynamic evaluation parameters.

[0225] Step b15: The first node determines a parameter association relationship of at least one twin atom model based on the first predicted state parameter set to obtain a first association relationship parameter.

[0226] In an embodiment of the present application, the first node analyzes the first predicted state parameter set and obtains the first association relationship parameter of the first twin network based on the parameter association relationship between at least one twin atomic model, such as the previous and subsequent operation impact relationship, the business impact relationship, etc.

[0227] Step b16: The first node determines a parameter association relationship between physical entity nodes corresponding to at least one twin atom model based on the first actual state parameter set to obtain a second association relationship parameter.

[0228] In an embodiment of the present application, the first node analyzes the first actual state parameter set to obtain a second association relationship parameter that matches the first association relationship parameter.

[0229] Step b17: The first node determines an association evaluation parameter based on the first association relationship parameter and the second association relationship parameter.

[0230] In the embodiment of the present application, the first node performs parameter calculation and analysis on the first association relationship parameter and the second association relationship parameter to obtain an association evaluation parameter.

[0231] Based on the above embodiments, in other embodiments of the present application, step b14 can be implemented by steps b141 to b143:

[0232] Step b141: The first node determines the change trend of each dynamic parameter based on the first dynamic parameter set to obtain the first dynamic change parameter set.

[0233] In an embodiment of the present application, the first node analyzes the parameters in the first dynamic parameter set, determines the changing trend of each dynamic parameter in the first dynamic parameter set over time, and obtains the first dynamic change parameter set.

[0234] Step b142: The first node determines the change trend of each dynamic parameter based on the second dynamic parameter set to obtain a second dynamic change parameter set.

[0235] In an embodiment of the present application, the first node analyzes the parameters in the second dynamic parameter set, determines the changing trend of each dynamic parameter in the second dynamic parameter set over time, and obtains the second dynamically changing parameter set.

[0236] Step b143: The first node performs consistency calculation on the first dynamic parameter set, the second dynamic parameter set, the first dynamically changing parameter set, and the second dynamically changing parameter set to determine a dynamic evaluation parameter.

[0237] In an embodiment of the present application, the first node uses consistency calculation to calculate the first dynamic parameter set, the second dynamic parameter set, the first dynamic change parameter, and the second dynamic change parameter to determine the dynamic evaluation parameter.

[0238] Based on the above embodiment, in other embodiments of the present application, step b143 can be implemented by steps c11 to c15:

[0239] Step c11: The first node performs consistency calculation on the first dynamic parameter set and the second dynamic parameter set to obtain a first consistency parameter.

[0240] Step c12: The first node performs consistency calculation on the first dynamically changing parameter set and the second dynamically changing parameter set to obtain a second consistency parameter.

[0241] Step c13: The first node calculates a first product of a first weight coefficient and a first consistency parameter.

[0242] In the embodiment of the present application, the first weight coefficient is an empirical value obtained in advance based on a large number of experiments. In some application scenarios, it can also be set according to actual conditions.

[0243] Step c14: The first node calculates a second product of the second weight coefficient and the second consistency parameter.

[0244] In the embodiment of the present application, the second weight coefficient is an empirical value obtained in advance based on a large number of experiments. In some application scenarios, it can also be set according to actual conditions.

[0245] Step c15: The first node calculates the sum of the first product and the second product to obtain a dynamic evaluation parameter.

[0246] Based on the above embodiments, in other embodiments of the present application, step a112 can be implemented by steps d11 to d21:

[0247] Step d11: The first node determines a static weight coefficient, a dynamic weight coefficient and an associated evaluation weight coefficient.

[0248] In the embodiment of the present application, the static weight coefficient, the dynamic weight coefficient and the associated evaluation weight coefficient are empirical values ​​obtained in advance based on a large number of experiments. In some application scenarios, they can also be set according to actual conditions.

[0249] Step d12: The first node calculates the product of the static weight coefficient and the static evaluation parameter to obtain a weighted static evaluation parameter.

[0250] Step d13: The first node calculates the product of the dynamic weight coefficient and the dynamic evaluation parameter to obtain a weighted dynamic evaluation parameter.

[0251] Step d14: The first node calculates the product of the association evaluation weight coefficient and the association evaluation parameter to obtain a weighted association evaluation parameter.

[0252] Step d15: The first node calculates the cumulative sum of the weighted static evaluation parameter, the weighted dynamic evaluation parameter and the weighted associated evaluation parameter to obtain a comprehensive parameter.

[0253] Step d16: The first node calculates a first value of a static evaluation parameter.

[0254] Step d17: The first node calculates a second value of the dynamic evaluation parameter.

[0255] Step d18: The first node calculates a third value of the associated evaluation parameter.

[0256] Step d19: The first node calculates the first value, the second value and the third value to obtain a comprehensive value.

[0257] Step d20: The first node calculates a fourth value of the ratio of the comprehensive parameter to the comprehensive value.

[0258] Step d21: The first node calculates the ratio of the fourth value to the root value of the number of parameters included in the comprehensive parameter to obtain the network accuracy.

[0259] Based on the above embodiments, in other embodiments of the present application, step a15 can be implemented by steps a151 to a153:

[0260] Step a151: The first node counts the second probability that the model accuracy obtained by simulating and calculating the first twin network within a preset time period is greater than or equal to the overall accuracy threshold of the twin network.

[0261] In the embodiment of the present application, the preset duration is an empirical duration obtained from a large number of experiments, or it may be an empirical duration set according to actual needs. Within the preset duration, the first node runs the first twin network, performs simulation operations on the first twin network, and then calculates the second probability that the model accuracy obtained during the simulation operation of the first twin network is greater than or equal to the overall accuracy threshold of the twin network within the preset duration.

[0262] After the first node executes step a151, it may choose to execute step a152 or step a153.

[0263] Step a152: If the second probability is greater than or equal to the probability threshold of the twin network accuracy, the first node determines the first accuracy threshold of each twin atomic model as the current accuracy threshold of the corresponding twin atomic model.

[0264] Step a153: If the second probability is less than the probability threshold of the twin network accuracy, the first node determines the first accuracy threshold of each twin atomic model based on the overall accuracy threshold of the twin network, the probability threshold of the twin network accuracy and the first twin network.

[0265] In an embodiment of the present application, the first node determines the overall accuracy threshold of the twin network and the probability threshold of the twin network accuracy corresponding to the first twin network, wherein the overall accuracy threshold of the twin network and the probability threshold of the twin network accuracy can be accuracy thresholds pre-set according to the requirements for the corresponding first twin network. The overall accuracy threshold of the twin network represents the accuracy of the corresponding result when the corresponding twin network is applied, and the probability threshold of the twin network accuracy is the number of times that the corresponding result of the corresponding twin network reaches the overall accuracy threshold of the twin network within a certain number of times, which must exceed a certain proportional experience value.

[0266] The first node adjusts and analyzes the accuracy threshold of each twin atomic model according to the overall accuracy threshold of the twin network corresponding to the first twin network and the probability threshold of the twin network accuracy, and calculates the first accuracy threshold of each twin atomic model to match the first accuracy threshold of the twin atomic model with the requirements of the corresponding twin network, so that the final twin atomic model is more in line with the actual requirements, and the final first twin network can quickly meet the requirements, thereby reducing the consumption of computing resources.

[0267] Based on the above embodiment, in other embodiments of the present application, step a153 can be implemented by steps e11 to e16:

[0268] Step e11: If the second probability is less than the probability threshold of the twin network accuracy, when the accuracy of the first node calculation model is greater than or equal to the current accuracy threshold of each twin atomic model, the conditional probability that the model accuracy of the first twin network is greater than or equal to the overall accuracy threshold of the twin network.

[0269] In an embodiment of the present application, when the second probability is less than the probability threshold of the twin network accuracy, the first node uses a conditional probability calculation method to calculate the condition: when the model accuracy is greater than or equal to the current accuracy threshold of each twin atomic model, the conditional probability that the model accuracy of the first twin network is greater than or equal to the overall accuracy threshold of the twin network.

[0270] Step e12: The first node calculates the prior probability when the model accuracy is greater than or equal to the current accuracy threshold of each twin atom model.

[0271] In an embodiment of the present application, the first node uses a priori algorithm to calculate the prior probability that the model accuracy is greater than or equal to the current accuracy threshold of each twin atom model.

[0272] Step e13: The first node determines the posterior probability when the model accuracy is greater than or equal to the current accuracy threshold of each twin atom model based on the conditional probability and the prior probability.

[0273] In an embodiment of the present application, the second node uses a posterior probability algorithm to calculate the calculated conditional probability and prior probability, and determines the posterior probability corresponding to when the model accuracy is greater than or equal to the current accuracy threshold of each twin atom model.

[0274] Step e14: The first node optimizes the current accuracy threshold of each twin atom model to obtain the optimized accuracy threshold of each twin atom model.

[0275] In an embodiment of the present application, the first node may use methods such as gradient ascent or stochastic gradient ascent to adjust the current accuracy threshold to obtain an optimized accuracy threshold.

[0276] Step e15, after the first node updates the current accuracy threshold of each twin atomic model to the corresponding optimized accuracy threshold, repeat the step of calculating the conditional probability that the model accuracy of the first twin network is greater than or equal to the overall accuracy threshold of the twin network when the model accuracy is greater than or equal to the current accuracy threshold of each twin atomic model, until the maximum a posteriori probability is determined.

[0277] In an embodiment of the present application, after the first node updates the current accuracy threshold of each twin atom model to the optimized accuracy threshold, step a151 and steps e11 to e15 are repeatedly executed, and the current accuracy threshold of each twin atom model is repeatedly iterated until the current accuracy threshold of each twin atom model obtained maximizes the posterior probability, or the gradient approaches zero.

[0278] Step e16: The first node determines that the first accuracy threshold of each twin atom model is the current accuracy threshold corresponding to the maximum a posteriori probability.

[0279] In the embodiment of the present application, the current accuracy threshold corresponding to the maximum delay probability is used as the first accuracy threshold of the atomic model of each twin atomic model. In this way, the accuracy threshold of each atomic model is dynamically set according to the accuracy requirements of the task for the overall digital twin network, so that the accuracy requirements of the overall digital twin network are decomposed into each atomic model, ensuring the accuracy requirements of the final overall digital twin network.

[0280] Based on the aforementioned embodiment, the embodiment of the present application provides a network optimization method for improving the overall accuracy of the digital twin network through hierarchical measurement accuracy and closed-loop correction. The corresponding application scenario is a terminal device (User, Equipment, UE), a centralized twin task node with operation and maintenance management (Service Management and Orchestration, OAM) function, corresponding to the aforementioned first node, and a digital twin network application scenario between at least one base station g-NB. Correspondingly, the implementation process of this method can refer to Figure 5 As shown, the specific steps include:

[0281] Step f101, UE reports terminal communication indicator data to g-NB.

[0282] Among them, the terminal communication indicator data may be, for example, UE data such as channel state information (CSI), load update interval (LUI), measurement report (MR), and terminal capability (UE capability).

[0283] Step f102: Concentrate the twin task nodes to build a twin scenario, specify the digital twin network type (Date Models type) involved in the twin environment, and the overall accuracy Min (P) and stability requirements Min (S) for the digital twin network.

[0284] Step f103: The centralized twin task node sends a task request including DT Modelstype, Min(P) and Min(S) to the g-NB involved in the digital twin network.

[0285] Among them, the OAM in the centralized twin task node can send a task request to the g-NB involved in the task through the E2 interface set by OAM for communicating with the g-NB.

[0286] Step f104: g-NB constructs a twin model based on the received task request, the terminal communication indicator data and the base station related data of g-NB, and obtains the atomic model to be corrected.

[0287] Among them, the atomic model to be corrected can be, for example, a UE behavior model, a channel twin model, a network topology twin model, a resource allocation model, and other application models corresponding to communication applications.

[0288] Step f105, g-NB performs model correction on the atomic model to be corrected to obtain a corrected twin atom model.

[0289] The g-NB is provided with an accuracy measurement module for measuring the accuracy of each atomic model to be corrected, a threshold setting module for determining the model threshold of the atomic model to be corrected, and an intelligent verification model for model correction.

[0290] In some application scenarios, step f105 can be implemented by the following steps:

[0291] Step f1051, g-NB starts the accuracy measurement module, performs accuracy measurement on the atomic model to be corrected, and obtains the model accuracy c of the atomic model to be corrected.

[0292] In an embodiment of the present application, the accuracy value of the atomic model to be corrected can be comprehensively calculated based on the accuracy measurement values ​​of three indicators, namely, static parameter accuracy (corresponding to the aforementioned static evaluation parameters), dynamic parameter accuracy (corresponding to the aforementioned dynamic evaluation parameters) and parameter association relationship accuracy (corresponding to the aforementioned association evaluation parameters), so as to more accurately evaluate the model accuracy of the atomic model to be corrected.

[0293] Specifically, static parameters are a variety of static properties and indicators of the atomic model to be corrected. Static parameters are related to the inherent properties of the atomic model to be corrected and do not change with the state of the atomic model to be corrected. Static parameters can be represented by defining a static parameter eigenvector. Define sp m represents the static parameters of the atomic model to be modified, sp r Represents the static parameters of the entity corresponding to the atomic model to be corrected. The accuracy of the static parameters can be measured by finding the similarity of the eigenvectors of the two. One calculation method can be: Among them, “·” represents the dot product between vectors, ‖‖ represents the second-order norm, and c s The closer it is to 1, the higher the accuracy.

[0294] Dynamic parameters are a variety of dynamic properties and indicators of the atomic model to be corrected. The dynamic parameters are related to the state of the atomic model to be corrected and can change with the state of the atomic model to be corrected. They also include two indicators: parameter value and parameter change trend. Dynamic parameters can be represented by defining dynamic parameter feature vectors. Define dp m represents the dynamic parameters of the atomic model to be corrected, dp r Represents the dynamic parameters of the entity corresponding to the atomic model to be corrected. The parameter change trend can be characterized by the first-order time-domain derivative and the second-order time-domain derivative of the dynamic parameter, which respectively represent the dynamic parameter change trend and future direction, that is, define dpt m Indicates the dynamic parameter change trend of the atomic model to be corrected, dpt r Indicates the dynamic parameter change trend of the entity corresponding to the atomic model to be corrected, and They respectively represent the state change trend and future direction of the nth dynamic parameter of the atomic model to be corrected, and They respectively represent the state change trend and future direction of the nth dynamic parameter of the entity corresponding to the atomic model to be corrected. For example, when the model to be corrected is a channel model, its dynamic parameters may be, for example, signal strength, signal-to-noise ratio, path loss, etc. The first-order time-domain derivative of the path loss represents the change trend of the loss, and the second-order time-domain derivative represents the future direction of the loss.

[0295] Therefore, the accuracy of the dynamic parameters can be expressed by the consistency of the dynamic parameter feature vector of the atomic model to be corrected and the dynamic parameter feature vector of the entity corresponding to the atomic model to be corrected, as well as the consistency of the dynamic parameter change trend. An implementation algorithm can be expressed as: Among them, c d The closer it is to 1, the higher the accuracy. a and b are weight coefficients, which can be predetermined empirical values.

[0296] The parameter association relationship is the explicit or implicit relationship between the parameters of the atomic model to be modified and the parameters of the associated model. The parameter association relationship can be represented by defining an associated adjacency vector, for example, It represents the parameter association relationship between the parameters in the atomic model u to be corrected and the parameters in the atomic model v to be corrected, and defines It represents the parameter association relationship between the parameters in the entity corresponding to the atomic model u to be corrected and the parameters in the entity corresponding to the atomic model v to be corrected.

[0297] Therefore, the accuracy of the parameter association relationship can be expressed by the similarity between the association neighbor vector of the atomic model to be corrected and the association neighbor vector of the corresponding entity of the atomic model to be corrected. A calculation algorithm can be expressed as: Among them, c pl The closer it is to 1, the higher the accuracy.

[0298] Finally, the model accuracy of the atomic model to be corrected can be measured by the consistency of the results of the atomic model to be corrected and the entity corresponding to the atomic model to be corrected under the same configuration strategy. One calculation method can be expressed as: Among them, w s 、w d 、w pl The values ​​are all in the range of 0 to 1, including two endpoints, which are weight coefficients corresponding to static parameter accuracy, dynamic parameter accuracy, and parameter correlation accuracy, which can be predetermined empirical values. The closer the c value is to 1, the stronger the consistency is, and the higher the accuracy of the atomic model to be corrected is.

[0299] The atomic model to be corrected is constructed for the first time. Therefore, after g-NB executes f1051, it executes step f1052:

[0300] Step f1052, the g-NB accuracy measurement module assigns an initialization model accuracy threshold to the atomic model to be corrected.

[0301] Step f1053: If the model accuracy of the atomic model to be corrected is greater than or equal to the initialization model accuracy threshold, the twin atom model is determined to be the atomic model to be corrected.

[0302] Step f1054: If the model accuracy of the atomic model to be corrected is less than the initialization model accuracy threshold, g-NB performs model correction on the atomic model to be corrected until a twin atom model is obtained.

[0303] Among them, the model correction process is: the intelligent verification module starts the atomic model adaptive learning mechanism, and determines |min(c-θ,0)| for the atomic model i to be corrected, that is, after calculating the difference between the model accuracy c of the atomic model to be corrected and the initialization model accuracy threshold θ, the minimum value between the difference between the model accuracy c of the atomic model to be corrected and the initialization model accuracy threshold θ and 0 is determined, and the absolute value of the minimum value is determined as the atomic model regularization factor to be added to the loss (Loss) function to correct the atomic model to be corrected, which can be recorded as: Lcorrected = Loriginal + λ(|min(c-θ,0)|), where λ is the regularization coefficient, which is used to control the influence of the regularization term on the atomic model to be corrected, thereby optimizing the model parameters or model structure until the accuracy reaches the threshold, Lcorrected represents the Loss function after the model is corrected, and Loriginal represents the Loss function before the model is corrected. In this way, the Lcorrected value can be calculated to optimize the model parameters or model structure of the atomic model to be corrected, obtain an intermediate atomic model, and then update the atomic model to be corrected to the intermediate atomic model, and repeat steps f1051 to f1054 until a twin atom model with a model accuracy greater than or equal to the initialization model accuracy threshold is calculated.

[0304] Step f106, g-NB sends the twin atomic model to the centralized twin task node.

[0305] Step f107, the centralized twin task node receives the twin atomic model sent by at least one g-NB.

[0306] Step f108: The centralized twin task node arranges and combines at least one twin atomic model to obtain a first twin network.

[0307] Among them, the OAM in the centralized twin task node can orchestrate at least one twin atomic model received, such as the UE behavior digital model, the channel digital model, the resource allocation digital model, etc., based on the topology model to construct an overall twin network and obtain the first twin network.

[0308] Step f109: The centralized twin task node starts its intelligent verification module to perform intelligent verification on the first twin network to obtain the predicted state of the first twin network.

[0309] Among them, constructing the twin snapshot of the first twin network specifically refers to deducing the predicted state of each twin atomic model at time t+Δt in advance based on the first twin network, and obtaining the predicted state of the first twin network, where time t represents the current time. Among them, the predicted state of the first twin network deduced in advance includes the predicted state of each twin atomic model at time t+Δt, which is recorded as n represents the number of all twin atomic models received by the centralized twin task node.

[0310] Step f110: At time t+Δt, the centralized twin task node collects the actual operating status of the corresponding twin atomic model from each g-NB to obtain the actual operating status of the first twin network.

[0311] Among them, the actual operating state corresponding to each twin atom model can be recorded as

[0312] Step f111, the centralized twin task node determines the twin network accuracy of the first twin network based on the predicted state and actual operating state of the first twin network.

[0313] Among them, the twin network accuracy of the first twin network can be identified by the consistency accuracy of the overall digital twin network, which can be recorded as: C DT =||C(P m ,P r )||, where ‖‖ represents the second-order norm calculation. The specific process can refer to the implementation process of determining the model accuracy of the atomic model to be corrected at step f1051, which will not be described in detail here.

[0314] Step f112: If the twin network accuracy is greater than or equal to Min(P), the centralized twin task nodes determine the probability that the twin network accuracy is greater than or equal to Min(P) to obtain a first probability.

[0315] Step f113: If the first probability is greater than or equal to Min(S), the centralized twin task node determines that the first twin network is the final constructed twin network, and the twin network construction ends.

[0316] Step f114: If the twin network accuracy is less than Min(P), or the first probability is less than Min(S), the centralized twin task node determines the accuracy threshold of each twin atomic model based on Min(P) and Min(S).

[0317] Among them, first, calculate the conditional probability: combine the first twin network to calculate the model accuracy of the twin atomic model i when it meets the threshold requirement (c i ≥θ i ), the first twin network meets the threshold requirement (c total ≥Min(P)) total ≥Min(P)c i ≥θ i ); Second, calculate the prior probability: Combined with the model simulation results of the twin atom model i, the calculation accuracy meets c i ≥θ i The probability p(c i ≥θ i ); Third, calculate the posterior probability: use the conditional probability and prior probability to calculate whether the atomic model i to be corrected meets the accuracy threshold θ i The posterior probability f(c i ≥θ i |c total ≥Min(P))=f(c total ≥Min(P)|c i ≥θ i )·p(c i ≥θ i ) / p(c total ≥Min(P)); Fourth, maximum a posteriori probability estimation: optimize θ i , iterate the first to third steps to maximize the posterior probability and determine the final accuracy threshold θ of the atomic model to be corrected i Among them, in the fourth step, optimize θ i One implementation process can be: (1) using methods such as gradient ascent or stochastic gradient ascent for the current accuracy threshold θ i Adjust to optimize the posterior probability in the direction of increasing the value. (2) Repeat the iteration of the current precision threshold until the posterior probability is maximized or the gradient approaches zero, at which point the optimized precision threshold is obtained. Specifically, adjust θ i The maximum a posteriori probability estimation formula f(c i ≥θ i |c total ≥Min(P)) can be specifically recorded as: for f(c i ≥θ i |c total ≥Min(P)) function for θ iTo calculate the gradient, the gradient is Determine the current gradient ascent direction (i.e., the positive gradient direction), and adjust θ by adjusting the step size to α i , so that the maximum a posteriori probability formula iterates in the direction of increasing: Iterate the first two steps, when the gradient When it approaches 0 or is less than a very small threshold ε, the iteration ends. At this time, it is considered that the posterior probability has reached the maximum, and the optimized accuracy threshold θ is obtained. i .

[0318] Step f115: The centralized twin task node sends the accuracy threshold of each twin atomic model to the corresponding g-NB.

[0319] Step f116: After receiving the precision threshold of the twin atomic model, the g-NB updates the atomic model to be corrected to the twin atomic model, and updates the corresponding precision threshold to the precision threshold sent by the centralized twin task node. Then execute f105 to f116 until a first twin network that meets the requirements is obtained.

[0320] In this way, the twin task is analyzed, and the participating models are organized to perform model correction according to the analysis results. The overall measurement and verification of the twin environment composed of multiple models are carried out, and the joint optimization of multiple models is carried out to make the overall twin environment meet user requests. In this way, the overall accuracy of the twin network is hierarchically improved through two closed-loop processes. Among them, the atomic model is measured and corrected before orchestration to ensure the accuracy of the atomic model. After orchestration, the twin network formed by the combination of multiple models is measured and corrected as a whole to ensure the overall accuracy of the twin network. The threshold is set dynamically, and the threshold for triggering correction of each atomic model is calculated based on the expected twin network accuracy, so as to quickly achieve the twin network accuracy.

[0321] It should be noted that, for the description of the same steps and the same contents in this embodiment as those in other embodiments, reference can be made to the description in other embodiments and will not be repeated here.

[0322] The network optimization method provided in the embodiment of the present application determines the model accuracy of the atomic model to be corrected through the second node, determines the first accuracy threshold of the atomic model to be corrected based on the overall accuracy threshold of the twin network and the probability threshold reached by the twin network accuracy, and performs model correction on the atomic model to be corrected based on the model accuracy and the first accuracy threshold. After obtaining the twin atomic model, the twin atomic model is sent to the first node so that the first node receives the corrected twin atomic model sent by at least one second node, combines the at least one twin atomic model to obtain a first twin network, and performs model optimization processing on the first twin network to obtain a second twin network whose network accuracy is greater than or equal to the second accuracy threshold. In this way, the second node dynamically determines the first accuracy threshold of the atomic model to be corrected according to the overall accuracy threshold of the twin network corresponding to the twin network corresponding to the atomic model to be corrected and the probability threshold of the twin network accuracy. Then, the atomic model to be corrected is corrected according to the first accuracy threshold, and the corrected twin atomic model is sent to the first node, so that the first node combines and processes the corrected twin atomic model sent by at least one second node according to network requirements to obtain a first twin network. Finally, the first twin network is model optimized to obtain a second twin network with a network accuracy greater than or equal to the second accuracy threshold and stability greater than or equal to the stability threshold. In this way, through two-level model optimization, and different levels of model optimization are optimized at different nodes, the requirements on the computing power of the nodes are reduced, and the problem of poor accuracy of the current digital twin network model is solved. A method is proposed to optimize the accuracy of a single model of the digital twin network model, and then optimize the accuracy of the digital twin network model obtained by arranging multiple single models, thereby improving the accuracy of the digital twin network model and ensuring the reliability of the digital twin network model.

[0323] Based on the above embodiments, the embodiments of the present application provide a first network optimization device, which can be applied to Figure 1 and Figures 3-4 In the network optimization method provided in the corresponding embodiment, refer to Figure 6 As shown, the first network optimization device 4 may include: a receiving unit 41, a first processing unit 42 and a second processing unit 43; wherein:

[0324] A receiving unit 41 is configured to receive a corrected twin atom model sent by at least one second node; wherein the model accuracy of the twin atom model is greater than or equal to a first accuracy threshold, and the twin atom model corresponds to a communication network application;

[0325] A first processing unit 42, configured to perform combination processing on at least one twin atom model to obtain a first twin network;

[0326] The second processing unit 43 is used to perform model optimization processing on the first twin network to obtain a second twin network whose network accuracy is greater than or equal to the second accuracy threshold and whose stability is greater than or equal to the stability threshold.

[0327] In other embodiments of the present application, the second processing unit includes: a first prediction module, a first acquisition module and a processing module; wherein:

[0328] A first prediction module, used to predict a first predicted state parameter set of at least one twin atomic model included in the first twin network at a first moment;

[0329] A first acquisition module, configured to acquire, at a first moment, from at least one second node, a first actual state parameter set of at least one twin atomic model included in the first twin network;

[0330] A processing module is used to perform model optimization processing on the first twin network model based on the first predicted state parameter set and the first actual state parameter set to obtain a second twin network.

[0331] In other embodiments of the present application, the processing module is used to implement the following steps:

[0332] Calculating the network accuracy of the first twin network based on the first predicted state parameter set and the first actual state parameter set;

[0333] If the network accuracy is greater than or equal to the second accuracy threshold, a simulation operation is performed on the first twin network to obtain a simulation operation result;

[0334] Based on the simulation operation result, a first probability that the network accuracy of the first twin network is greater than or equal to the second accuracy threshold is calculated;

[0335] If the first probability is greater than or equal to the stability threshold, the second twin network is determined to be the first twin network.

[0336] In other embodiments of the present application, when the processing module executes the step of calculating the network accuracy of the first twin network based on the first predicted state parameter set and the first actual state parameter set, it can be implemented by the following steps:

[0337] Determining static evaluation parameters, dynamic evaluation parameters, and associated evaluation parameters based on the first predicted state parameter set and the first actual state parameter set;

[0338] Based on the static evaluation parameters, dynamic evaluation parameters and associated evaluation parameters, the network accuracy of the first twin network is obtained.

[0339] In other embodiments of the present application, when the processing module executes the step of determining the static evaluation parameters, the dynamic evaluation parameters, and the associated evaluation parameters based on the first predicted state parameter set and the first actual state parameter set, it can be implemented by the following steps:

[0340] Grouping the first prediction state parameter set to obtain a first static parameter set and a first dynamic parameter set;

[0341] Grouping the first actual state parameter set to obtain a second static parameter set and a second dynamic parameter set;

[0342] Perform similarity calculation on the first static parameter set and the second static parameter set to obtain static evaluation parameters;

[0343] Determining a dynamic evaluation parameter based on the first dynamic parameter set and the second dynamic parameter set;

[0344] Based on the first predicted state parameter set, determining a parameter association relationship of at least one twin atom model to obtain a first association relationship parameter;

[0345] Based on the first actual state parameter set, determining a parameter association relationship between physical entity nodes corresponding to at least one twin atom model to obtain a second association relationship parameter;

[0346] Based on the first association relationship parameter and the second association relationship parameter, an association evaluation parameter is determined.

[0347] In other embodiments of the present application, when the processing module performs the step of determining the dynamic evaluation parameter based on the first dynamic parameter set and the second dynamic parameter set, it can be implemented by the following steps:

[0348] Based on the first dynamic parameter set, determining a change trend of each dynamic parameter to obtain a first dynamic change parameter set;

[0349] Based on the second dynamic parameter set, determining a change trend of each dynamic parameter to obtain a second dynamic change parameter set;

[0350] Consistency calculation is performed on the first dynamic parameter set, the second dynamic parameter set, the first dynamically changing parameter set, and the second dynamically changing parameter set to determine a dynamic evaluation parameter.

[0351] In other embodiments of the present application, the processing module executes the steps of performing consistency calculation on the first dynamic parameter set, the second dynamic parameter set, the first dynamically changing parameter set, and the second dynamically changing parameter set to determine the dynamic evaluation parameters, which can be achieved by the following steps:

[0352] Performing consistency calculation on the first dynamic parameter set and the second dynamic parameter set to obtain a first consistency parameter;

[0353] Performing consistency calculation on the first dynamically changing parameter set and the second dynamically changing parameter set to obtain a second consistency parameter;

[0354] Calculating a first product of a first weight coefficient and a first consistency parameter;

[0355] Calculating a second product of a second weight coefficient and a second consistency parameter;

[0356] The sum of the first product and the second product is calculated to obtain a dynamic evaluation parameter.

[0357] In other embodiments of the present application, when the processing module executes the steps based on the static evaluation parameters, the dynamic evaluation parameters and the associated evaluation parameters to obtain the network accuracy of the first twin network, it can be achieved by the following steps:

[0358] Determine static weight coefficient, dynamic weight coefficient and associated evaluation weight coefficient;

[0359] Calculate the product of the static weight coefficient and the static evaluation parameter to obtain the weighted static evaluation parameter;

[0360] Calculate the product of the dynamic weight coefficient and the dynamic evaluation parameter to obtain a weighted dynamic evaluation parameter;

[0361] Calculate the product of the association evaluation weight coefficient and the association evaluation parameter to obtain the weighted association evaluation parameter;

[0362] Calculate the cumulative sum of the weighted static evaluation parameter, the weighted dynamic evaluation parameter and the weighted associated evaluation parameter to obtain a comprehensive parameter;

[0363] Calculating a first value of a static evaluation parameter;

[0364] Calculating a second value of the dynamic evaluation parameter;

[0365] Calculating a third value of the associated evaluation parameter;

[0366] Calculate the first value, the second value, and the third value to obtain a comprehensive value;

[0367] calculating a fourth value of a ratio of the comprehensive parameter to the comprehensive value;

[0368] The ratio of the fourth value to the root value of the number of parameters included in the comprehensive parameter is calculated to obtain the network accuracy.

[0369] In other embodiments of the present application, the processing module is further used to implement the following steps:

[0370] If the network accuracy is less than the second accuracy threshold, or the first probability is less than the stability threshold, the first accuracy threshold of each twin atomic model is determined based on the overall accuracy threshold of the twin network and the probability threshold of the twin network accuracy;

[0371] The first accuracy threshold of the twin atom model corresponding to each second node is sent to the corresponding second node, so that the second node corrects the corresponding twin atom model based on the first accuracy threshold.

[0372] In other embodiments of the present application, the processing module is used to implement the step of: if the network accuracy is less than the second accuracy threshold, or the first probability is less than the stability threshold, based on the overall accuracy threshold of the twin network and the twin network accuracy reaching the probability threshold, determining the first accuracy threshold of each twin atomic model, which can be implemented by the following steps:

[0373] If the network accuracy is less than the second accuracy threshold, or the first probability is less than the stability threshold, the second probability that the model accuracy obtained by simulating the first twin network is greater than or equal to the overall accuracy threshold of the twin network within the preset time period is calculated;

[0374] If the second probability is greater than or equal to the twin network accuracy reaching probability threshold, determine the first accuracy threshold of each twin atom model as the current accuracy threshold of the corresponding twin atom model;

[0375] If the second probability is less than the probability threshold of the twin network accuracy, the first accuracy threshold of each twin atomic model is determined based on the overall accuracy threshold of the twin network, the probability threshold of the twin network accuracy and the first twin network.

[0376] In other embodiments of the present application, the processing module is used to execute the step: if the second probability is less than the probability threshold of the twin network accuracy, based on the overall accuracy threshold of the twin network, the probability threshold of the twin network accuracy, and the first twin network, the first accuracy threshold of each twin atomic model is determined, which can be achieved by the following steps:

[0377] If the second probability is less than the probability threshold of the twin network accuracy, when the calculation model accuracy is greater than or equal to the current accuracy threshold of each twin atomic model, the conditional probability that the model accuracy of the first twin network is greater than or equal to the overall accuracy threshold of the twin network;

[0378] Calculate the prior probability that the model accuracy is greater than or equal to the current accuracy threshold of each twin atom model;

[0379] Based on the conditional probability and the prior probability, determine the posterior probability when the model accuracy is greater than or equal to the current accuracy threshold of each twin atom model;

[0380] Optimizing the current accuracy threshold of each twin atom model to obtain an optimized accuracy threshold;

[0381] After updating the current accuracy threshold of each twin atom model to the corresponding optimized accuracy threshold, repeat the steps to calculate the conditional probability that the model accuracy of the first twin network is greater than or equal to the overall accuracy threshold of the twin network when the model accuracy is greater than or equal to the current accuracy threshold of each twin atom model, until the maximum a posteriori probability is determined;

[0382] The first accuracy threshold of each twin atom model is determined to be the current accuracy threshold corresponding to the maximum a posteriori probability.

[0383] It should be noted that the specific implementation process of the interaction between the units and modules in the embodiment of the present application can be referred to Figure 1 and Figures 3-4 The implementation process of the network optimization method provided in the corresponding embodiment will not be repeated here.

[0384] The first network optimization device provided in the embodiment of the present application receives through a first node a twin atomic model sent by at least one second node, which is corrected according to a first accuracy threshold dynamically determined by the second node, and combines at least one twin atomic model to obtain a first twin network, and then performs model optimization processing on the first twin network to obtain a second twin network whose network accuracy is greater than or equal to the second accuracy threshold. In this way, the second node dynamically determines the first accuracy threshold of the atomic model to be corrected according to the overall accuracy threshold of the twin network corresponding to the twin network corresponding to the atomic model to be corrected and the probability threshold of the twin network accuracy. Then, the atomic model to be corrected is corrected according to the first accuracy threshold, and the corrected twin atomic model is sent to the first node, so that the first node combines and processes the corrected twin atomic model sent by at least one second node according to network requirements to obtain a first twin network. Finally, the first twin network is model optimized to obtain a second twin network whose network accuracy is greater than or equal to the second accuracy threshold. In this way, through two-level model optimization, and different levels of model optimization are optimized at different nodes, the requirements on the computing power of the nodes are reduced, and the problem of poor accuracy of the current digital twin network model is solved. A method is proposed to optimize the accuracy of a single model of the digital twin network model, and then optimize the accuracy of the digital twin network model obtained by arranging multiple single models, thereby improving the accuracy of the digital twin network model and ensuring the reliability of the digital twin network model.

[0385] Based on the above embodiments, the embodiments of the present application provide a second network optimization device, which can be applied to Figures 2 to 4 In the network optimization method provided in the corresponding embodiment, refer to Figure 7 As shown, the second network optimization device 5 may include: an acquisition unit 51, a determination unit 52, a correction unit 53 and a sending unit 54; wherein:

[0386] An acquisition unit 51 is used to acquire a first accuracy threshold of the atomic model to be corrected;

[0387] A determination unit 52, configured to determine the model accuracy of the atomic model to be corrected; wherein the atomic model to be corrected corresponds to a communication network application;

[0388] A correction unit 53 is used to perform model correction on the atomic model to be corrected based on the model accuracy and the first accuracy threshold to obtain a twin atom model; wherein the model accuracy of the twin atom model is greater than or equal to the first accuracy threshold;

[0389] The sending unit 54 is used to send the twin atom model to the first node.

[0390] In other embodiments of the present application, the acquisition unit is used to implement the following steps:

[0391] A first precision threshold is determined as an initialization precision threshold.

[0392] In other embodiments of the present application, the acquisition unit is further used to implement the following steps:

[0393] Receiving a first accuracy threshold sent by the first node; wherein the first accuracy threshold is an accuracy threshold corresponding to the twin atom model;

[0394] Correspondingly, after the acquisition unit, the second network optimization device further includes: a processing unit; wherein:

[0395] The processing unit is used to update the atomic model to be corrected to a twin atomic model, and then repeatedly execute the steps to determine the model accuracy of the atomic model to be corrected until the first accuracy threshold sent by the first node is not received.

[0396] In other embodiments of the present application, the determination unit includes: a second prediction module, a second acquisition module and a first calculation module; wherein:

[0397] A second prediction module, used for predicting a second set of predicted state parameters of the atomic model to be corrected at a second moment;

[0398] A second acquisition module, used for acquiring, at a second moment, a second set of actual state parameters of the atomic model to be corrected;

[0399] The first calculation module is used to calculate the model accuracy of the atomic model to be corrected based on the second predicted state parameter set and the second actual state parameter set.

[0400] In other embodiments of the present application, the correction unit includes: a first determination module, a second calculation module, an adjustment module and an update module; wherein:

[0401] A first determination module, configured to determine that the twin atom model is an atom model to be corrected if the model accuracy is greater than or equal to a first accuracy threshold;

[0402] A second calculation module, configured to calculate a difference between the first accuracy threshold and the model accuracy if the model accuracy is less than the first accuracy threshold;

[0403] A second calculation module, used for calculating a third product of the difference and the regularization coefficient;

[0404] A second calculation module is used to calculate the sum of the loss value of the atomic model to be corrected and the third product to obtain an adjustment parameter;

[0405] An adjustment module, used for reversely adjusting the adjustable parameters in the atomic model to be corrected based on the adjustment parameters to obtain an intermediate atomic model;

[0406] The updating module updates the atomic model to be corrected to an intermediate atomic model, and repeats the steps to determine the model accuracy of the atomic model to be corrected until a twin atomic model is obtained.

[0407] In other embodiments of the present application, the second network optimization device further includes: a receiving unit and a generating unit; wherein:

[0408] A receiving unit, used to receive terminal communication indicator data sent by a terminal device;

[0409] The receiving unit is further used to receive the twin task request sent by the first node; wherein the twin task request at least includes: a model type, an overall accuracy threshold of the twin network, and a probability threshold of the twin network accuracy;

[0410] The generating unit is further used to generate at least one atomic model to be corrected based on the terminal communication indicator data and the model type.

[0411] It should be noted that the specific implementation process of the interaction between the units and modules in the embodiment of the present application can be referred to Figures 2 to 4 The implementation process of the network optimization method provided in the corresponding embodiment will not be repeated here.

[0412] The second network optimization device provided in the embodiment of the present application determines the model accuracy of the atomic model to be corrected through at least one second node, determines the first accuracy threshold of the atomic model to be corrected based on the overall accuracy threshold of the twin network and the probability threshold reached by the twin network accuracy, and performs model correction on the atomic model to be corrected based on the model accuracy and the first accuracy threshold. After obtaining the twin atomic model, the twin atomic model is sent to the first node so that the first node receives the corrected twin atomic model sent by at least one second node, combines at least one twin atomic model to obtain a first twin network, and performs model optimization processing on the first twin network to obtain a second twin network whose network accuracy is greater than or equal to the second accuracy threshold. In this way, the second node dynamically determines the first accuracy threshold of the atomic model to be corrected according to the overall accuracy threshold of the twin network corresponding to the twin network corresponding to the atomic model to be corrected and the probability threshold of the twin network accuracy. Then, the atomic model to be corrected is corrected according to the first accuracy threshold, and the corrected twin atomic model is sent to the first node, so that the first node combines and processes the corrected twin atomic model sent by at least one second node according to network requirements to obtain a first twin network. Finally, the first twin network is model optimized to obtain a second twin network whose network accuracy is greater than or equal to the second accuracy threshold. In this way, through two-level model optimization, and different levels of model optimization are optimized at different nodes, the requirements on the computing power of the nodes are reduced, and the problem of poor accuracy of the current digital twin network model is solved. A method is proposed to optimize the accuracy of a single model of the digital twin network model, and then optimize the accuracy of the digital twin network model obtained by arranging multiple single models, thereby improving the accuracy of the digital twin network model and ensuring the reliability of the digital twin network model.

[0413] Based on the above embodiments, the embodiments of the present application provide a network optimization system, which can be applied to Figures 1 to 4 In the network optimization method provided in the corresponding embodiment, refer to Figure 8 As shown, the network optimization system 6 may include: a first node 61 and at least one second node 62; wherein:

[0414] The first node 61 is used to implement Figure 1 and Figures 3-4 The implementation process of the network optimization method provided by the corresponding embodiment will not be described in detail here;

[0415] The second node 62 is used to implement Figures 2 to 4 The implementation process of the network optimization method provided by the corresponding embodiment will not be described in detail here.

[0416] Based on the foregoing embodiments, the embodiments of the present application provide a computer-readable storage medium, referred to as a storage medium, which stores one or more programs, and the one or more programs can be executed by one or more processors to implement reference Figure 1 and Figures 3-4 ,or Figures 2 to 4 The implementation process of the network optimization method provided in the corresponding embodiment will not be repeated here.

[0417] Based on the foregoing embodiments, an embodiment of the present application further provides a computer program product, including a computer program, which can be executed by a first node or at least one second node to complete the steps of any of the foregoing methods.

[0418] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of hardware embodiments, software embodiments, or embodiments in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage and optical storage, etc.) that contain computer-usable program code.

[0419] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0420] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0421] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0422] The above description is only a preferred embodiment of the present application and is not intended to limit the protection scope of the present application.

Claims

1. A network optimization method, characterized in that: The method is applied to a first node, and the method comprises: Receiving a corrected twin atom model sent by at least one second node; wherein the model accuracy of the twin atom model is greater than or equal to a first accuracy threshold, and the twin atom model corresponds to a communication network application; Combining at least one of the twin atomic models to obtain a first twin network; The first twin network is subjected to model optimization processing to obtain a second twin network having a network accuracy greater than or equal to a second accuracy threshold and a stability greater than or equal to a stability threshold.

2. The method according to claim 1, characterized in that The performing model optimization processing on the first twin network to obtain a second twin network whose network accuracy is greater than or equal to a second accuracy threshold and whose stability is greater than or equal to a stability threshold includes: Predicting a first predicted state parameter set of at least one of the twin atomic models included in the first twin network at a first moment; At the first moment, obtaining a first actual state parameter set of at least one of the twin atomic models included in the first twin network from at least one of the second nodes; Based on the first predicted state parameter set and the first actual state parameter set, the first twin network model is optimized to obtain the second twin network.

3. The method according to claim 2, characterized in that The step of performing model optimization processing on the first twin network model based on the first predicted state parameter set and the first actual state parameter set to obtain the second twin network includes: Calculating the network accuracy of the first twin network based on the first predicted state parameter set and the first actual state parameter set; If the network accuracy is greater than or equal to the second accuracy threshold, performing simulation operation on the first twin network to obtain a simulation operation result; Based on the simulation operation result, a first probability that the network accuracy of the first twin network is greater than or equal to the second accuracy threshold is calculated; If the first probability is greater than or equal to the stability threshold, the second twin network is determined to be the first twin network.

4. The method according to claim 3, characterized in that The calculating the network accuracy of the first twin network based on the first predicted state parameter set and the first actual state parameter set includes: Determining static evaluation parameters, dynamic evaluation parameters, and associated evaluation parameters based on the first predicted state parameter set and the first actual state parameter set; Based on the static evaluation parameters, the dynamic evaluation parameters and the associated evaluation parameters, the network accuracy of the first twin network is obtained.

5. The method according to claim 4, characterized in that The determining of static evaluation parameters, dynamic evaluation parameters, and associated evaluation parameters based on the first predicted state parameter set and the first actual state parameter set includes: Grouping the first prediction state parameter set to obtain a first static parameter set and a first dynamic parameter set; Grouping the first actual state parameter set to obtain a second static parameter set and a second dynamic parameter set; Performing similarity calculation on the first static parameter set and the second static parameter set to obtain the static evaluation parameter; Determining the dynamic evaluation parameter based on the first dynamic parameter set and the second dynamic parameter set; Based on the first predicted state parameter set, determining a parameter association relationship of at least one of the twin atom models to obtain a first association relationship parameter; Based on the first actual state parameter set, determining a parameter association relationship between at least one physical entity node corresponding to the twin atom model to obtain a second association relationship parameter; The association evaluation parameter is determined based on the first association relationship parameter and the second association relationship parameter.

6. The method according to claim 5, characterized in that The determining the dynamic evaluation parameter based on the first dynamic parameter set and the second dynamic parameter set includes: Based on the first dynamic parameter set, determining a change trend of each dynamic parameter to obtain a first dynamic change parameter set; Based on the second dynamic parameter set, determining a change trend of each dynamic parameter to obtain a second dynamic change parameter set; A consistency calculation is performed on the first dynamic parameter set, the second dynamic parameter set, the first dynamically changing parameter set, and the second dynamically changing parameter set to determine the dynamic evaluation parameter.

7. The method according to claim 6, characterized in that The performing consistency calculation on the first dynamic parameter set, the second dynamic parameter set, the first dynamically changing parameter set, and the second dynamically changing parameter set to determine the dynamic evaluation parameter includes: Performing consistency calculation on the first dynamic parameter set and the second dynamic parameter set to obtain a first consistency parameter; Performing consistency calculation on the first dynamically changing parameter set and the second dynamically changing parameter set to obtain a second consistency parameter; Calculating a first product of a first weight coefficient and the first consistency parameter; Calculating a second product of a second weight coefficient and the second consistency parameter; The sum of the first product and the second product is calculated to obtain the dynamic evaluation parameter.

8. The method according to claim 4, characterized in that The obtaining the network accuracy of the first twin network based on the static evaluation parameter, the dynamic evaluation parameter and the associated evaluation parameter includes: Determine static weight coefficient, dynamic weight coefficient and associated evaluation weight coefficient; Calculating the product of the static weight coefficient and the static evaluation parameter to obtain a weighted static evaluation parameter; Calculating the product of the dynamic weight coefficient and the dynamic evaluation parameter to obtain a weighted dynamic evaluation parameter; Calculating the product of the association evaluation weight coefficient and the association evaluation parameter to obtain a weighted association evaluation parameter; Calculating the cumulative sum of the weighted static evaluation parameter, the weighted dynamic evaluation parameter and the weighted associated evaluation parameter to obtain a comprehensive parameter; Calculating a first value of the static evaluation parameter; Calculating a second value of the dynamic evaluation parameter; Calculating a third value of the associated evaluation parameter; Calculating the first value, the second value, and the third value to obtain a comprehensive value; Calculating a fourth value of a ratio of the comprehensive parameter to the comprehensive value; The network accuracy is obtained by calculating a ratio of the fourth value to the square root value of the number of parameters included in the comprehensive parameter.

9. The method according to claim 3, characterized in that: The method further comprises: If the network accuracy is less than the second accuracy threshold, or the first probability is less than the stability threshold, based on the overall accuracy threshold of the twin network and the probability threshold of the twin network accuracy reaching, determine the first accuracy threshold of each of the twin atomic models; The first precision threshold of the twin atom model corresponding to each second node is sent to the corresponding second node, so that the second node corrects the corresponding twin atom model based on the first precision threshold.

10. The method according to claim 9, characterized in that If the network accuracy is less than the second accuracy threshold, or the first probability is less than the stability threshold, based on the overall accuracy threshold of the twin network and the probability threshold of the twin network accuracy reaching, determining the first accuracy threshold of each of the twin atomic models, including: If the network accuracy is less than the second accuracy threshold, or the first probability is less than the stability threshold, a second probability that the model accuracy obtained by simulating and calculating the first twin network is greater than or equal to the overall accuracy threshold of the twin network is calculated within a preset time period; If the second probability is greater than or equal to the twin network accuracy reaching probability threshold, determine the first accuracy threshold of each of the twin atom models as the current accuracy threshold of the corresponding twin atom model; If the second probability is less than the twin network accuracy probability threshold, the first accuracy threshold of each of the twin atomic models is determined based on the overall accuracy threshold of the twin network, the twin network accuracy probability threshold and the first twin network.

11. The method according to claim 10, characterized in that If the second probability is less than the twin network accuracy probability threshold, based on the twin network overall accuracy threshold, the twin network accuracy probability threshold and the first twin network, determining the first accuracy threshold of each of the twin atomic models, including: If the second probability is less than the probability threshold of the twin network accuracy, the conditional probability that the model accuracy of the first twin network is greater than or equal to the overall accuracy threshold of the twin network is calculated when the model accuracy is greater than or equal to the current accuracy threshold of each of the twin atomic models; Calculating a priori probability when the model accuracy is greater than or equal to a current accuracy threshold of each of the twin atom models; Based on the conditional probability and the prior probability, determining a posterior probability when the model accuracy is greater than or equal to the current accuracy threshold of each twin atom model; Optimizing the current precision threshold of each of the twin atom models to obtain an optimized precision threshold of each of the twin atom models; After updating the current accuracy threshold of each of the twin atom models to the corresponding optimized accuracy threshold, repeatedly executing the step of calculating the conditional probability that the model accuracy of the first twin network is greater than or equal to the overall accuracy threshold of the twin network when the model accuracy is greater than or equal to the current accuracy threshold of each of the twin atom models, until the maximum a posteriori probability is determined; Determine the first precision threshold of each of the twin atom models as the current precision threshold corresponding to the maximum a posteriori probability.

12. A network optimization method, characterized in that: The method is applied to the second node, and the method includes: Obtaining a first accuracy threshold of the atomic model to be corrected; Determining the model accuracy of the atomic model to be corrected; wherein the atomic model to be corrected corresponds to a communication network application; Based on the model accuracy and the first accuracy threshold, the to-be-corrected atomic model is subjected to model correction to obtain a twin atom model; wherein the model accuracy of the twin atom model is greater than or equal to the first accuracy threshold; The twin atomic model is sent to a first node, so that the first node builds a corresponding first twin network based on the twin atomic model.

13. The method according to claim 12, characterized in that The step of obtaining a first accuracy threshold of the atomic model to be corrected includes: The first precision threshold is determined as an initialization precision threshold.

14. The method according to claim 12, characterized in that The step of obtaining a first accuracy threshold of the atomic model to be corrected includes: Receiving the first accuracy threshold sent by the first node; wherein the first accuracy threshold is an accuracy threshold corresponding to the twin atom model; Correspondingly, after obtaining the first accuracy threshold of the atomic model to be corrected, the method further includes: After updating the to-be-corrected atomic model to the twin atomic model, the step of determining the model accuracy of the to-be-corrected atomic model is repeated until the first accuracy threshold sent by the first node is not received.

15. The method according to claim 12, characterized in that The step of determining the model accuracy of the atomic model to be corrected comprises: Predicting a second predicted state parameter set of the to-be-corrected atomic model at a second moment; At the second moment, obtaining a second set of actual state parameters of the to-be-corrected atomic model; The model accuracy of the atomic model to be corrected is calculated based on the second predicted state parameter set and the second actual state parameter set.

16. The method according to claim 12, characterized in that The step of performing model correction on the to-be-corrected atomic model based on the model accuracy and the first accuracy threshold to obtain a twin atom model includes: If the model accuracy is greater than or equal to the first accuracy threshold, determining that the twin atom model is the atom model to be corrected; If the model accuracy is less than the first accuracy threshold, calculating the difference between the first accuracy threshold and the model accuracy; Calculating a third product of the difference and a regularization coefficient; Calculating the sum of the loss value of the atomic model to be corrected and the third product to obtain an adjustment parameter; Reversely adjusting the adjustable parameters in the to-be-corrected atomic model based on the adjustment parameters to obtain an intermediate atomic model; The atomic model to be corrected is updated to the intermediate atomic model, and the steps of determining the model accuracy of the atomic model to be corrected are repeated until the twin atomic model is obtained.

17. The method according to claim 13, characterized in that The method further comprises: Receiving terminal communication indicator data sent by the terminal device; Receive a twin task request sent by the first node; wherein the twin task request includes at least: a model type, an overall accuracy threshold of the twin network, and a probability threshold of the twin network accuracy reaching; At least one atomic model to be corrected is generated based on the terminal communication indicator data and the model type.

18. A first network optimization device, characterized in that: The device is applied to a first node, and comprises: a receiving unit, a first processing unit, and a second processing unit; wherein: The receiving unit is configured to receive a corrected twin atom model sent by at least one second node; wherein the model accuracy of the twin atom model is greater than or equal to a first accuracy threshold, and the twin atom model corresponds to a communication network application; The first processing unit is used to perform combination processing on at least one of the twin atomic models to obtain a first twin network; The second processing unit is used to perform model optimization processing on the first twin network to obtain a second twin network whose network accuracy is greater than or equal to a second accuracy threshold and whose stability is greater than or equal to a stability threshold.

19. A second network optimization device, characterized in that: The device is applied to the second node, and comprises: an acquisition unit, a determination unit, a correction unit and a sending unit; wherein: The acquisition unit is used to acquire a first accuracy threshold of the atomic model to be corrected; The determining unit is used to determine the model accuracy of the atomic model to be corrected; wherein the atomic model to be corrected corresponds to a communication network application; The correction unit is used to perform model correction on the atomic model to be corrected based on the model accuracy and the first accuracy threshold to obtain a twin atom model; The sending unit is used to send the twin atom model to the first node.

20. A network optimization system, characterized in that: The system comprises at least: a first node and at least one second node; wherein: The first node is used to implement the steps of the network optimization method according to any one of claims 1 to 11; The second node is used to implement the steps of the network optimization method according to any one of claims 12 to 17.

21. A storage medium, characterized in that: The storage medium stores a resource conflict indication program, which, when executed, is used to implement the steps of the network optimization method according to any one of claims 1 to 11 or claims 12 to 17.

22. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the computer program implements the steps of the network optimization method according to any one of claims 1 to 11 or claims 12 to 17.

Images