A method and device for updating a digital twin model

Abstract

The present invention relates to a method and device for updating a digital twin model. The method mainly includes: evaluating the input and output parameters of the model according to the input points of the digital twin model parameters, and classifying the model parameters; constructing a model flow graph based on the candidate points between models, and evaluating the model connection weights; calculating the association degree between models according to the model flow graph and the connection weights, and determining the model automatic operation mechanism according to the characteristics of the model instance; selecting the model combination according to the selection weights by means of election to determine the most satisfactory model combination, and using the feedback of the operation of the model combination as the weight basis for updating the parameters of each model. The present invention can provide a basis for which models can operate automatically and which models cannot. In the case of limited resources, the least number of models can support more application scenarios.

Description

Technical Field

[0001] The present invention relates to the field of communication technologies, and in particular, to a method and device for updating a digital twin model. Background Art

[0002] The development of the digital economy requires the support of information technology and communication technology, and the demand for digital transformation in the industry also promotes the evolution of the traditional communication network architecture towards a "cloud-network integration" architecture. It is very difficult for the communication network architecture constructed by the traditional "physical device plus professional network management" mode to meet the digital transformation requirements of various industries such as "resources on demand, flexible control, and secure and reliable". Moreover, with the continuous development of the informatization process of the whole society, the demands for network latency, capacity, bandwidth, etc. are also showing a rapid growth trend, and the traditional network is facing pressures such as large-scale and high-capacity data exchange and processing.

[0003] Digital twin originates from the CPS system (Cyber-Physical Systems, information physical system). CPS is a multi-dimensional complex system integrating the physical environment, information network, and computing. Through the deep integration and collaboration of control, communication, and computer technologies, it completes adaptive control, comprehensive perception, and information service supply. It consists of multiple elements, including a complex network composed of physical devices and digital components.

[0004] As the core part of the digital twin, how to effectively measure the model, determine the connection relationship between models, and provide a basis for the accuracy of the model and the update of model parameters is a problem to be solved in this field. On the other hand, in the case of limited resources, how to select as few models as possible to support as much work as possible is also a problem to be solved in this field.

[0005] In view of the above situation, how to overcome the defects of the existing technology is a difficult problem to be solved in this technical field. Summary of the Invention

[0006] In view of the above-mentioned defects or improvement requirements of the existing technology, the present invention provides a method and device for updating a digital twin model. The method includes model evaluation, directed graph construction, automatic operation mechanism, and model parameter update. First, evaluate the input and output parameters of the model, secondly, construct a flow chart of the user / requirement model, evaluate the connection weights of the model, and finally, construct an automatic operation mechanism of the model and update the model parameters.

[0007] The embodiments of the present invention adopt the following technical solutions:

[0008] In a first aspect, the present invention provides a method for updating a digital twin model, including:

[0009] Evaluate the input and output parameters of the model according to the input points of the digital twin model parameters, and classify the model parameters;

[0010] Construct a model flow diagram based on the candidate points between models, and evaluate the model connection weights;

[0011] Calculate the correlation degree between models according to the model flow diagram and the connection weights, and determine the model automatic operation mechanism according to the characteristics of the model instance;

[0012] According to the election method, select the model combination according to the selection weights to determine the most satisfactory model combination, and use the feedback of the operation of this model combination as the weight basis for updating the parameters of each model.

[0013] Further, the evaluation of the input and output parameters of the model according to the input points of the digital twin model parameters and the classification of the model parameters specifically include: classifying the model parameters into active points, candidate points, edge points, and outlier points; among them, the active points include the data reported from the network management and the device side; the candidate points include the data provided by other models; the edge points include the sensor device data collected through the network management message queue class interface; the outlier points include the data measured through experiments when the model is not deployed to the digital twin system.

[0014] Further, the construction of the model flow diagram according to the candidate points between models specifically includes:

[0015] Evaluate the requirements and find the minimum level of the model;

[0016] Analyze the variation law of different network conditions corresponding to the requirements over time, and describe the changes of different networks in different time periods, as well as the resource changes of the minimum-level model under the network;

[0017] Obtain the physical topologies of two similar regions with the highest degree of association with the currently selected network, and construct a digital twin virtual topology according to the topological hierarchical relationship;

[0018] According to the constructed digital twin virtual topology and the minimum level of the model, find the functional models that calculate the same physical characteristic parameters at the same level, and construct a functional model cluster;

[0019] Based on the candidate points of the model, construct a model flow diagram relying on the model input and output relationships between the functional model clusters and the functional model clusters.

[0020] Further, the connection weight represents the influence factor of the output of the previous model on the subsequent model, and the influence factor includes the degree of influencing the output parameters in the previous model and the similarity between the total output parameters of the previous model and the total input parameters of the subsequent model.

[0021] Further, the calculation of the correlation degree between models according to the model flow diagram and the connection weights, and the determination of the model automatic operation mechanism according to the characteristics of the model instance specifically include:

[0022] Calculate the association degree between the model and other models in the model flow diagram according to the connection weight calculation model;

[0023] Select the model with the highest importance level of the model when resources permit;

[0024] Find the corresponding functional model according to the selected model, and select the model with the highest ranking in the instantiation of the probability model or the simulation model instantiation in the functional model.

[0025] Furthermore, the higher the association degree between the model and other models, the deeper the connection between the model and other models, and the current model can support more models. Among them, the association degree between the model and other models is linearly proportional to the connection weight of the model.

[0026] Furthermore, the importance level of the model is proportional to the association degree of the model and the accuracy of the resources required by the model each time.

[0027] Furthermore, the finding the corresponding functional model according to the selected model specifically includes: if the selected model is a simulation model, then select the corresponding probability model; if the selected model is a probability model, then select the corresponding simulation model.

[0028] Furthermore, according to the election method, select the model combination according to the selection weight to determine the most satisfactory model combination, and use the feedback of the operation of the model combination as the basis for updating the parameter weights of each model, which specifically includes:

[0029] All models sign up for the combination model campaign of the target model, and the target model assigns the same selection weight to all competing models;

[0030] After each task call is completed, compare the output of the target's own model in this task environment with the output of the competing model combination and the output of the target model alone, and adjust the model selection weight until the most satisfactory model combination is found;

[0031] Use the feedback of the operation of the most satisfactory model combination in the current task as the basis for updating the parameter weights of each model.

[0032] In a second aspect, the present invention also provides a digital twin model update architecture, including a model evaluation module, a directed graph construction module, an automatic operation mechanism module, and a model parameter update module, where:

[0033] The model evaluation module classifies the model parameters into active points, candidate points, edge points, and outlier points according to the input points of the digital twin model parameters;

[0034] The directed graph construction module constructs a model flow graph based on the candidate points between models and evaluates the connection weights of the models;

[0035] The automatic operation mechanism module calculates the association degree between models according to the model flow graph and the connection weights, and determines the automatic operation situation of the models according to the characteristics of the model instances;

[0036] The model parameter update module selects the model combination according to the selection weights by means of election to determine the most satisfactory model combination, and uses the feedback of the operation of the model combination as the weight basis for updating the parameters of each model.

[0037] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0038] (1) By establishing the relationship between the input and output of the digital twin model, the present invention constructs a user / requirement model operation flow graph, establishes a weight connection between models, and provides a basis for determining which models can operate automatically and which cannot. In the case of limited resources, the least number of models can support more application scenarios.

[0039] (2) According to the usage of the models called for each task, the present invention corrects and combines the model parameters, so that when the previous model is used in combination with the subsequent model, the calculation error caused by theory / statistics between the two models can be reduced, making the combined use of models more efficient in the current environment, and thus greatly improving the model accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required to be used in the embodiments of the present invention will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention, and those of ordinary skill in the art can obtain other drawings according to these drawings without creative efforts.

[0041] Figure 1 It is a flowchart of a digital twin model update method provided in Embodiment 1 of the present invention;

[0042] Figure 2 It is a schematic diagram of model parameter classification provided in Embodiment 1 of the present invention;

[0043] Figure 3 It is an expanded flowchart of step 200 provided in Embodiment 1 of the present invention;

[0044] Figure 4 It is a schematic diagram of a model flow graph provided in Embodiment 1 of the present invention;

[0045] Figure 5It is the expanded flowchart of step 300 provided by Embodiment 1 of the present invention;

[0046] Figure 6 It is the expanded flowchart of step 400 provided by Embodiment 1 of the present invention;

[0047] Figure 7 It is a schematic diagram of a digital twin model update architecture provided by Embodiment 2 of the present invention. Detailed implementation manners

[0048] In order to make the objectives, technical solutions and advantages of the present invention clearer and more understandable, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0049] The present invention is an architecture of a specific function system. Therefore, in specific embodiments, the functional logic relationships of each structural module are mainly described, and the specific software and hardware implementation manners are not limited.

[0050] In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other. The present invention will be described in detail below with reference to the drawings and embodiments.

[0051] With the process of digitalization and intelligentization, traditional communication networks will realize automatic, intelligent management and control of optical networks by machines in the future. However, the management and control solutions formed based on machine intelligence may not be reassuring, and instructions cannot be directly issued to the device side. Based on this situation, it is necessary to establish a twin optical network based on digital twin technology, so that it can accurately simulate the operation mechanism of the optical network, synchronize the operation state of the optical network in real time, verify the management and control solutions constructed by machines in the twin network, and then issue them to the physical device after being accurate. In this way, the risks brought by the management and control solutions formed entirely by machine intelligence can be avoided. Based on this, the following embodiments of the present invention provide a digital twin model update method and architecture. It realizes the classification of digital twin model parameter types, constructs the update flow chart of model users / requirement models in the digital twin platform, and realizes the automatic update process of models in the digital twin platform.

[0052] Embodiment 1:

[0053] As Figure 1 shown, this Embodiment 1 proposes a digital twin model update method, and this method includes the following steps.

[0054] Step 100: Evaluate the input and output parameters of the model according to the digital twin model parameter input points, and classify the model parameters.

[0055] Step 200: Construct a model flow diagram based on candidate points between models, and evaluate model connection weights.

[0056] Step 300: Calculate the correlation degree between models according to the model flow diagram and connection weights, and determine the model automatic operation mechanism according to model instance characteristics.

[0057] Step 400: Select the model combination according to the selection weight based on the election method to determine the most satisfactory model combination, and use the feedback of the operation of this model combination as the weight basis for updating the parameters of each model.

[0058] Reference Figure 2 Referring to the schematic diagram of model parameter classification shown in , in step 100 of this preferred embodiment, when classifying model parameters, the model parameters are classified into active points, candidate points, marginalized points, and outlier points; among them, the active points include data reported from the network management and device sides, and the acquisition frequency is related to the communication time interval between the network management and the device sides; the candidate points include data provided by other models, and when the model runs, the output parameters of each step may be candidate points of other models; the marginalized points include sensor device data collected through the network management message queue class interface. At present, some sensing devices in the network do not run in real time. For example, for the OPM disk, when OPM disk data needs to be collected, it is obtained by issuing commands through the network management interface. The main use of the marginalized points is to call the marginalized point data to check the model calculation error and model optimization after the model runs; the outlier points include data collected through experiments in the laboratory when the model is not deployed to the digital twin system. For example, the outlier points of the non-linear model include back-to-back data, constellation diagram data, etc. These data generally do not change or increase during the operation of the live network, and are mostly used to compare the output parameters of other models. For example, when the optical module model outputs constellation diagram / back-to-back data, it can be compared with the outlier data of this model.

[0059] As Figure 3 shown, in step 200 of this preferred embodiment, the constructing a model flow diagram according to candidate points between models specifically includes the following steps.

[0060] Step 201: Evaluate the user / requirements and find the minimum level of the model. For example, when the user / requirements calculate the non-linear situation in an optical link, by evaluating the accuracy requirements of the user / requirements, select the smallest granularity model. When the accuracy requirement is about 1.5 dB, select the network and network element level models. When the accuracy requirement is about 1.0 dB, select the single-disk or module level models. When the accuracy requirement is within 0.5 dB, select the device and material level models.

[0061] Step 202: Analyze the variation law of different network conditions corresponding to the requirements over time, describe the changes in different networks during different time periods, and the resource changes of the minimum-level model under the network. This step is used to analyze the scope fluctuation of users / requirements over a period of time, compare the changes in different time intervals, and describe the changes in the use of model resources in different regions during different time periods. For example, when calculating the non-linear situation in an optical link L for user requirements, not only the current optical link L needs to be calculated, but also multiple links in this region need to be considered uniformly, because the network is mutually influential and dynamically changing. Only by fully considering the information of other links that affect the optical link L can the non-linear change situation of the optical link L be accurately known.

[0062] In the current network environment, users / requirements change periodically over time. For example, during the day, because people are concentrated in companies and factories, the demand for personnel access, bandwidth, and computing resources in industrial parks is high, and the usage frequency of network elements and single-disk models in this region is high. After work at night, there is almost no personnel access, and the demand for bandwidth and computing resources in industrial parks is extremely low, and the usage frequency of network elements and single-disk models in industrial parks is low. However, the personnel access, bandwidth, and computing resources in communities will double compared to during the day. Therefore, this embodiment statistically analyzes the usage situation in the region over one year and constructs a matrix of regional changes over time. The time interval is the same as the time interval for synchronizing network management data in the digital twin system.

[0063] The regional time change table is as follows:

[0064]

[0065] Referring to the above table, this embodiment constructs a two-dimensional matrix according to the regional time table , where the rows represent regions and the columns represent time. For example, represents the number of task / requirement invocations in Region 1 during the time period . Perform singular value decomposition on the matrix to find the first two regions corresponding to the maximum and the second maximum singular values, that is, the two most similar regions with the highest network correlation.

[0066] Step 203: Obtain the physical topologies of the two most similar regions with the highest degree of network association to the currently selected network, and construct a digital twin virtual topology according to the topological hierarchy relationship. The two most similar regions with the highest degree of network association selected in this step are also the first two regions corresponding to the maximum and second maximum singular values found in the previous step, and these two regions have the greatest impact on the above optical link L. Then, according to the hierarchy relationship selected in Step 201, a digital twin network is fabricated. This digital twin network includes two similar regions and the user / task focus area. For example, if the single-disk level model is selected in Step 201, the optical link can first be divided into multiple OTSs, and further the OTSs can be divided into multiple OMSs plus WSS disk models. Each OMS includes multiple OCHs plus multiplexer / demultiplexer disk models, and each OCH corresponds to multiple optical fiber models and OA amplifier disk models. Finally, the above single-disk level models are combined to form a physical entity network.

[0067] Step 204: According to the constructed digital twin virtual topology and the minimum level of the model, find the functional models that calculate the same physical characteristic parameters at the same level, and construct a functional model cluster. For example, to calculate the output OSNR of the first OA disk, the following functional models are included: OSNR monitoring functional model, OSNR prediction functional model, OA amplification mechanism functional model, etc. All the above functional models together constitute the model cluster of the first OA. There are overlapping functional models between model clusters. For example, the OSNR monitoring model is included in almost all OA clusters.

[0068] Step 205: Based on the candidate points of the model, construct a model flow diagram relying on the model input and output relationships between the functional model clusters. For example, the digital twin virtual topology diagram contains two types of functional models. In the functional model one, there are model one and model three, and in the functional model two, there are model two and model four. The output parameters of model one . They are respectively the input parameters of model two and model four. Model two outputs . Among them are respectively the input parameters of model one, three, and four. Then, a directed graph of the model is constructed through the connection of candidate points as Figure 4 described.

[0069] In this preferred embodiment, the connection weight evaluated in Step 200 represents the influence factor of the output of the previous model on the subsequent model, and the influence factor includes the degree of influence on the output parameters in the previous model and the similarity between the total output parameters of the previous model and the total input parameters of the subsequent model.

[0070] For example, the output of model one to model two. Then its influence factor consists of two parts. 1) At which step in model one the output parameter is, used to evaluate the degree of influence on the output parameters in model one ; 2) The similarity degree between the total output parameters of Model 1 and the total input parameters of Model 2 ; Among them:

[0071] ;

[0072] ;

[0073] ;

[0074] ;

[0075] .

[0076] Take , the first two parameters of , , then .

[0077] The meaning of each parameter and each term in the above formula is as follows:

[0078] represents the influence factor from the output of Model 1 to the input of Model 2.

[0079] Evaluate the degree of influence on the output parameters in Model 1.

[0080] represents the similarity degree between the total output parameters of Model 1 and the total input parameters of Model 2.

[0081] represents in Model 1 which is the nth output parameter.

[0082] represents the total number of output parameters in Model 1.

[0083] 、 respectively represent performing singular value decomposition on the output of Model 1 and the input of Model 2.

[0084] , represents the second-order diagonal matrix composed of the first two parameters of the singular value decomposition of the output of Model 1 and the input of Model 2.

[0085] represents the operation of the diagonal matrix.

[0086] For example, Model 2 is an optical signal-to-noise ratio monitoring model, and Model 3 is an optical signal-to-noise ratio prediction model. Calculate The weight of the link where it is located. The output parameters of the optical signal-to-noise ratio (OSNR) monitoring include the OSNR of 1 to 96 channels, arranged in ascending order. The input parameters of the OSNR prediction model include the OSNR of 1 to 96 channels, also arranged in ascending order. Represents the OSNR of the first channel. Then the degree of influence on the output parameters in Model 1 Is 1 / 96, which means that the OSNR of the first waveguide is the first among the 96 output parameters of the OSNR monitoring model and is also the most accurate parameter of the OSNR monitoring model (each digital twin model in this embodiment is accurate when used alone for calculation). On the other hand, the output of the OSNR monitoring model is completely the input of the OSNR prediction model. Then the similarity degree between the total output parameters of Model 1 and the total input parameters of Model 2 = 1. The output of Model 1 To Model 2, that is, the connection weight of the OSNR of the first channel output by the OSNR monitoring model to the OSNR prediction model is 1 / 96.

[0087] Such as Figure 5 As shown, in this preferred embodiment, step 300 specifically includes the following steps:

[0088] Step 301: Calculate the correlation degree between the model and other models in the model flow diagram according to the connection weight. The correlation degree The higher it is, the deeper the connection between the model and other models, and the more models the current model can support. The model correlation degree Is linearly proportional to the model connection weight. The greater the connection weight of the input and output parameters between the model and other models, the deeper the connection between the model and other models, and the Greater the correlation degree. For example: . 1 represents the correlation of the model with itself, and 0.5 represents the ratio of the connection weight value to the output model and the input model, generally using equal division.

[0089] Taking Model 3 as an example, given the situations of Model 1, 2, 3, and 4, only Model 2 and Model 3 establish a link connection. Then the correlation degree of Model 3 is 1 + 0.5 * 1 / 96 = 1.0052.

[0090] Step 302: Select the model with the highest importance degree of the model when resources permit; among them, the importance degree of the model is proportional to the correlation degree of the model and the accuracy of the resources required by the model each time . When the resources consumed by the model are closer to the resources it requires and can support more models, the importance degree of the model is higher.

[0091] Step 303: Find the corresponding functional model according to the selected model, and select the model with the highest ranking in the instantiation of the probability model or the simulation model in the functional model. For example, if the simulation model is selected in the previous step, then select the corresponding probability model; if the probability model is selected, then select the corresponding simulation model. The selection criteria for the probability model and the simulation model are different. The selection of the probability model pays more attention to how many models can be supported, and the selection of the simulation model pays more attention to how much the calculation error can be reduced. , In the probability model , in the simulation model .

[0092]

[0093] For example = 0.3, = 0.7 corresponds to selecting the simulation model, while when selecting the probability model = 0.7, = 0.3. represents the selection criterion, , represents the model correlation and the accuracy of the required resources of the selection ratio. represents when selecting the corresponding probability or simulation model, for and the attention ratios are inconsistent, represents and the sum of the attention ratios is 1.

[0094] As Figure 6 shown, in this preferred embodiment, step 400 specifically includes the following steps:

[0095] Step 401: All models sign up to participate in the combined model campaign of the target model, and the target model assigns the same selection weight to all competing models. In this step, each output parameter of each alternative model i is assigned a weight , and when selecting the alternative model for the first time, the output parameter weights of all alternative models are the same (1, 0.5, or any same value is acceptable).

[0096] Step 402: After each task call is completed, compare the combined output of the target model itself and the competing models in the task environment with the output of the target model alone, and adjust the model selection weight until the most satisfactory model combination is found. In this step, the weight of the user / task requirement is denoted as , define , in the t-th round according to Make a decision. The meanings of each item in the above formula are as follows: respectively represent the combination situations of models B and C in the previous rounds with the model combination, represents that the weight of model B in this t-th round is the sum of the weights when model B is called in the previous 1 to t - 1 rounds, represents that in the previous i-th round, model B was called in combination. At the t-th round, if the weight of model B in this t-th round is greater than the weights of all other models in this round ( ), then model B is selected in the t-th round ( ).

[0097] Step 403: Use the feedback of running the most satisfactory model combination in the current task as the basis for updating the weight of each model parameter.

[0098] For example, the optical same-route model publicly announces to recruit a combined model. At this time, the optical signal-to-noise ratio monitoring model and the optical signal-to-noise ratio prediction model that meet the candidate value conditions participate in the competition. Then the optical same-route model first assigns the same weights of 0.5 and 0.5 to the signal-to-noise ratio monitoring model and the optical signal-to-noise ratio prediction model. When called in the first round, the task scenario at this time is M, and the requirements of M are (M1, M2,...), =(M1, M2,...). Because the weights of the optical signal-to-noise ratio monitoring model and the optical signal-to-noise ratio prediction model are the same, the optical same-route model randomly selects one of the models to participate in the combination. For example, it selects the optical signal-to-noise ratio monitoring model. It is found that when used in combination with the optical signal-to-noise ratio monitoring model, the model accuracy drops by 0.2. Then at this time, the weight of the optical signal-to-noise ratio monitoring model in the current task scenario is reduced to 0.5 * (1 - 0.2) = 0.4, and so on.

[0099] When selecting task M next time, at this time the weight of the optical signal-to-noise ratio monitoring model is 0.4, and the weight of the optical signal-to-noise ratio prediction model is 0.5. Then the optical same-route model should select the optical signal-to-noise ratio prediction model with a higher model weight as the combined model. Then, according to the running situation of the model combination in task M, the weight of the optical signal-to-noise ratio prediction model is updated.

[0100] In summary, in this embodiment, through the relationship between the input and output of the digital twin model, a running flow chart of the user / requirement model is constructed, a weight connection is established between models, so that there are rules to follow for which model to run automatically and which model not to run. In the case of limited resources, the fewest models support more application scenarios. In addition, in this embodiment, according to the usage of the model called in each task, the model parameters are corrected and combined, so that when the previous model is used in combination with the subsequent model, the calculation error caused by theory / statistics between the two models can be reduced, making the combination of models more efficient in the current environment, and thus greatly improving the model accuracy.

[0101] Embodiment 2:

[0102] Based on the digital twin model update method provided in Embodiment 1, Embodiment 2 further provides a digital twin model update architecture. As Figure 7 shown, the digital twin model update architecture of this embodiment includes a model evaluation module, a directed graph construction module, an automatic operation mechanism module, and a model parameter update module. The above four modules are set as a model update layer between the probability model and the simulation model in the digital twin platform to interact with the probability model and the simulation model. When interacting, these four modules first evaluate the model input and output parameters, secondly construct the user / requirement model flow graph, evaluate the model connection weights, and finally construct the model automatic operation mechanism and update the model parameters.

[0103] Specifically, the model evaluation module is used to classify model parameters into active points, candidate points, edge points, and outlier points according to the digital twin model parameter input points. The model evaluation module unifies the four types of input and output parameters of the model. For example, the optical same-route model includes four types of parameters. Active points (data reported from network management and devices) are similar to OTDR data, candidate points (data provided by other models) such as the performance data of optical fibers like OSNR, edge points (sensing data collected and configured through network management) include the input and output power of the current link, and outlier points (data measured in the laboratory based on the current link) include the back-to-back OSNR of the link.

[0104] The directed graph construction module is used to construct a model flow graph based on candidate points between models and evaluate the connection weights of the models. The automatic operation mechanism module calculates the association degree between models according to the model flow graph and the connection weights, and determines the automatic operation situation of the models according to the characteristics of the model instances. The model parameter update module selects a model combination according to the selection weights based on an election method to determine the most satisfactory model combination, and uses the feedback of the operation of this model combination as the weight basis for updating the parameters of each model. Specifically, after the directed graph construction module constructs the user / requirement model update flow graph, the relationship between the models and the model input and output can be determined through the determination of the model flow graph. Therefore, when using the model combination, the impact of the output of the previous model on the input of the next model can be evaluated, and this is used as the connection weight of the model. Finally, when the connection weights between models in the flow graph are determined, the automatic operation mechanism module constructs an automatic operation mechanism for the models. The automatic operation mechanism of the models refers to how to use the fewest models to support more scenarios when resources are limited. Finally, when a task / requirement comes, the fewest models are also selected to support the operation according to the current resource situation. How to update the model parameters based on the execution situation of the models under the current task is the problem that the model parameter update module needs to solve. Through the model combination method, when comparing the combination of the previous and the next models with the usage situation of a single model, if the error increases compared with the usage of a single model, the two-model combination scheme is excluded; otherwise, the two-model combination scheme is considered valid.

[0105] Specifically, the directed graph construction module takes the virtual topology corresponding to the physical entity as a blueprint, takes the model candidate points as a benchmark, connects model to model, and constructs a directed graph of the model. The directed graph construction module of this embodiment specifically includes an evaluation sub-module, an analysis sub-module, a virtual topology generation sub-module, a functional model cluster construction sub-module, a model flow graph construction sub-module, and a connection weight calculation module, where: The evaluation sub-module is used to evaluate requirements and find the minimum level of the model; The analysis sub-module is used to analyze the changing rules of different network conditions corresponding to the user / requirements over time, describe the changes of different networks in different time periods, and the resource changes of the minimum-level model under the network; The virtual topology generation sub-module is used to obtain the physical topologies of two similar regions with the highest degree of association (i.e., the most representative) with the currently selected network, and construct a digital twin virtual topology according to the topological hierarchy relationship; The functional model cluster construction sub-module is used to find functional models that calculate the same physical characteristic parameters at the same level based on the constructed digital twin virtual topology and the minimum level of the model, and construct a functional model cluster; The model flow graph construction sub-module is used to construct a model flow graph of the user / requirements based on the candidate points of the model, relying on the model input and output relationships between the functional model clusters and the functional model clusters; The connection weight calculation module is used to calculate the connection weights between each model and other models. Specific examples of directed graph construction and connection weights can refer to step 200 of Embodiment 1, which will not be elaborated here.

[0106] The automatic operation mechanism module calculates the correlation degree between models according to the user / requirements model flow graph, determines the automatic operation situation of the models according to the characteristics of the model instances, enables the calculation results of the probability model to support more models, and enables the calculation results of the simulation model to correct the calculation results of the probability model. The automatic operation mechanism module of this embodiment specifically includes a correlation degree calculation module, an importance calculation module, and a model selection module, where: The correlation degree calculation module is used to calculate the correlation degree between each model and other models in the model flow graph according to the connection weights; The importance calculation module is used to calculate the importance of each model according to the correlation degree of each model and the accuracy of the resources required by the model each time; The model selection module is used to select the model with the highest importance, and select the functional model with the highest ranking in the instantiation of the corresponding probability model or the instantiation of the simulation model. Specific examples of the automatic operation mechanism can refer to step 300 of Embodiment 1, which will not be elaborated here.

[0107] For the model parameter update module, for any one of the input parameters in a model, multiple other models may output values with the same physical meaning. Such models are called backup models. However, due to the limitations of the operating mechanism and probability analysis, the output values are different from each other. How to select the most suitable input parameter for the model from these alternative values, which can not only support the operation of the model but also reduce the calculation error of the model corresponding to the alternative value, is the core of the model parameter update module. The model parameter update module of this embodiment specifically includes a campaign module, a selection weight adjustment module, and a feedback module, where: the campaign module is used to obtain all the models participating in the combined model campaign of the target model and assign the same selection weight to all the competing models; the selection weight adjustment module is used to, after each task call is completed, compare the combined output of the target model itself and the competing model combination with the output of the target model alone in the task environment, and adjust the model selection weight until the most satisfactory model combination is found; the feedback module is used to use the feedback of the most satisfactory model combination running in the current task as the basis for updating the parameter weights of each model. For a specific example of model parameter update, refer to step 400 of Embodiment 1, which will not be elaborated here.

[0108] In summary, through the relationship between the input and output of the digital twin model in this embodiment, a flow chart of the user / requirement model operation is constructed, a weight connection is established between models, and there are rules to determine which models run automatically and which do not. In the case of limited resources, the fewest models support more application scenarios. According to the usage of the model in each task call in this embodiment, the model parameters are corrected and combined, so that when the previous model is used in combination with the subsequent model, the calculation error caused by theory / statistics between the two models can be reduced, making the combined use of models more efficient in the current environment and thus greatly improving the model accuracy.

[0109] Those of ordinary skill in the art can understand that all or part of the steps in the various methods of the embodiment can be completed by instructing relevant hardware through a program, and the program can be stored in a computer-readable storage medium. The storage medium can include: Read Only Memory (ROM for short), Random Access Memory (RAM for short), magnetic disk or optical disk, etc.

[0110] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, and improvements made within the spirit and principle of the present invention shall be included in the protection scope of the present invention. The content not detailedly described in this specification belongs to the prior art well-known to those skilled in the art.

Claims

1. A method for updating a digital twin model, characterized in that, it includes: Evaluating the input and output parameters of the model based on the digital twin model parameter input points, and classifying the model parameters; including: classifying the model parameters into active points, candidate points, edge points, and outlier points; wherein, the active points include data reported from the network management and the device side; the candidate points include data provided by other models, and the candidate points include the optical fiber performance data OSNR; the edge points include sensor device data collected through the network management message queue class interface; the outlier points include data measured through experiments when the model is not deployed to the digital twin system, realizing the classification of digital twin model parameter types; Constructing a model flow diagram based on the candidate points between models, and evaluating the model connection weights; Calculating the association degree between models according to the model flow diagram and the connection weights, and determining the model automatic operation mechanism according to the model instance characteristics, so as to support more application scenarios with the least number of models under limited resources; According to the election method, selecting the model combination according to the selection weight to determine the most satisfactory model combination, and using the feedback of the operation of this model combination as the weight basis for updating the parameters of each model; realizing the automatic update process of the models in the digital twin platform, so as to correct and combine the model parameters according to the usage of the models called by each task, so that the previous model can reduce the calculation error caused by theory / statistics when combined with the latter model, making the model combination more efficient in the current environment and improving the model accuracy; Establishing a twin optical network based on digital twin technology through the foregoing process, enabling it to accurately simulate the operation mechanism of the optical network and synchronize the optical network operation state in real time.

2. The digital twin model update method according to claim 1, characterized in that, The constructing a model flow diagram according to the candidate points between models specifically includes: Evaluating the requirements and finding the minimum level of the model; Analyzing the variation law of different network conditions corresponding to the requirements over time, and describing the changes of different networks in different time periods, as well as the resource changes of the minimum level model under the network; Obtaining the physical topologies of two similar regions with the highest degree of association with the currently selected network, and constructing a digital twin virtual topology according to the topological hierarchical relationship; According to the constructed digital twin virtual topology and the minimum level of the model, finding the functional models that calculate the same physical characteristic parameters at the same level, and constructing a functional model cluster; Based on the candidate points of the model, constructing a model flow diagram relying on the model input and output relationships between the functional model clusters and the functional model clusters.

3. The digital twin model update method according to claim 2, characterized in that, The connection weight represents the influence factor of the output of the previous model on the next model, and the influence factor includes the degree of influencing the output parameters in the previous model and the similarity between the total output parameters of the previous model and the total input parameters of the next model.

4. The digital twin model update method according to claim 3, characterized in that, Calculating the degree of association between models according to the model flow diagram and connection weights, and determining the specific model automatic operation mechanism according to the characteristics of model instances specifically includes: Calculating the degree of association between a model and other models in the model flow diagram according to the connection weights; Selecting the model with the highest degree of importance of the model when resources permit; Finding the corresponding functional model according to the selected model, and selecting the model with the highest ranking in the instantiation of the probability model or the simulation model in the functional model.

5. The digital twin model update method according to claim 4, wherein, The higher the degree of association between the model and other models, the deeper the connection between the model and other models, and the current model can support more models. Among them, the degree of association between the model and other models is linearly proportional to the connection weight of the model.

6. The digital twin model update method according to claim 4, wherein, The degree of importance of the model is proportional to the degree of association of the model and the accuracy of the resources required by the model each time.

7. The digital twin model update method according to claim 4, wherein, Finding the corresponding functional model according to the selected model specifically includes: if the selected model is a simulation model, then selecting the corresponding probability model; if the selected model is a probability model, then selecting the corresponding simulation model.

8. The digital twin model update method according to claim 1, wherein, Selecting the model combination according to the selection weight according to the election method to determine the most satisfactory model combination, and using the feedback of the operation of the model combination as the basis for updating the parameter weights of each model specifically includes: All models sign up for the combined model campaign of the target model, and the target model assigns the same selection weight to all competing models; After each task call is completed, the model selection weight is adjusted according to the comparison between the output of the target model itself in the environment of the task and the output of the competing model combination and the output of the target model alone until the most satisfactory model combination is found; Using the feedback of the operation of the most satisfactory model combination in the current task as the basis for updating the parameter weights of each model.

9. A digital twin model update device, wherein, It includes a model evaluation module, a directed graph construction module, an automatic operation mechanism module, and a model parameter update module, among which: The model evaluation module classifies model parameters into active points, candidate points, edge points, and outlier points according to the input points of digital twin model parameters; including: classifying model parameters into active points, candidate points, edge points, and outlier points; among them, the active points include data reported from the network management and the device side; the candidate points include data provided by other models, and the candidate points include the performance data of the optical fiber OSNR; the edge points include sensor device data collected through the network management message queue class interface; the outlier points include data measured through experiments when the model is not deployed to the digital twin system, realizing the classification of digital twin model parameter types; The directed graph construction module constructs a model flow diagram according to the candidate points between models and evaluates the connection weights of the models; The automatic operation mechanism module calculates the association degree between models according to the model flow diagram and connection weights, and determines the automatic operation of the models according to the characteristics of model instances, so as to support more application scenarios with the least models under limited resources; The model parameter update module selects the model combination according to the selection weight based on the election method to determine the most satisfactory model combination, and uses the feedback of the operation of the model combination as the weight basis for updating the parameters of each model; realizes the automatic update process of the models in the digital twin platform, so as to correct and combine the model parameters according to the usage of the models called by each task, so that the previous model can reduce the calculation error caused by theory / statistics when combined with the latter model, making the combined use of models more efficient in the current environment and improving the model accuracy; Through the foregoing modules, a twin optical network based on digital twin technology is established to accurately simulate the operation mechanism of the optical network and synchronize the operation state of the optical network in real time.

Images