CN115051926B

CN115051926B - Digital twin device, model evaluation system and model operation method

Info

Publication number: CN115051926B
Application number: CN202210713618.7A
Authority: CN
Inventors: 李文超; 徐安然; 尹山; 匡立伟; 谢秋红
Original assignee: Fiberhome Telecommunication Technologies Co Ltd
Current assignee: Wuhan Changjiang Computing Technology Co ltd; Fiberhome Telecommunication Technologies Co Ltd
Priority date: 2022-06-22
Filing date: 2022-06-22
Publication date: 2023-06-09
Anticipated expiration: 2042-06-22
Also published as: WO2023246568A1; CN115051926A

Abstract

The invention relates to a digital twin device, a model evaluation system and a model operation method. The digital twin device mainly comprises a data model interaction module, a model updating module and a resource calling module, wherein: the data model interaction module is used for storing the probability model and the simulation model data into the data pool, the model updating module is used for updating the probability model and the simulation model, and the resource calling module is used for calling the probability model and the simulation model. The invention can realize establishment of physical properties such as hierarchy, functions and the like of the digital twin model, enables a communication equipment manufacturer to carry out simulation modeling on physical entity equipment while researching and producing the physical entity equipment, and simultaneously research and produce the digital twin model, and guides and verifies research and development of the physical entity equipment by carrying out experimental and even destructive experiments on the virtual digital twin model.

Description

Digital twin device, model evaluation system and model operation method

Technical Field

The invention relates to the technical field of communication, in particular to a digital twin device, a model evaluation system and a model operation method. The effects of realizing the integration isomorphism of the twinning world and the real data and the mutual cooperation of the twinning world and the real data are achieved by the real mapping deficiency, the real control deficiency and the virtual mapping real.

Background

The development of digital economy requires information technology and communication technology support, and the industry digital transformation requirement also promotes the evolution of the traditional communication network architecture to a cloud network convergence architecture. The traditional communication network architecture constructed by the mode of physical equipment and professional network management is difficult to meet the digital transformation requirements of 'resources on demand, flexible management and control, safety and reliability' in various industries. And with the continuous development of the informatization process of the whole society, the demands on network time delay, capacity, bandwidth and the like are rapidly increased, and the traditional network faces large-scale and large-capacity data exchange, processing and the like.

Digital twinning is derived from a CPS system (Cyber-Physical Systems), wherein the CPS is a multi-dimensional complex system integrating Physical environment, information network and operation, and the self-adaptive control, comprehensive perception and information service supply are completed through deep fusion and cooperation of control, communication and computer technology. It consists of a complex network of many elements, including physical devices and digital components.

But CPS belongs to the scientific research category, and digital twinning is an important bridge bringing the CPS scientific field into the industrial field. The digital twin concept can be traced back to Apollo project of the aviation and aerospace agency in the United states at the earliest time, when the aerospace craft performs various emergency tasks in space, ground scientists and maintainers need to perform monitoring and prediction on the ground by using the virtual model which is identical to the digital twin concept, and the scientificity and the accuracy of the emergency decision of the astronaut are improved. Digital twinning is a method for describing and modeling physical objective physical characteristics, behaviors, running processes, performance performances and the like by using a digital simulation technology.

However, in the field of intelligent optical networks of communication technology, digital twinning is not yet applied to the field of intelligent optical networks of communication technology, and how to use digital twinning in the field of intelligent optical networks of communication technology is an urgent problem to be solved in the whole industry based on the large-scale and large-capacity data exchange, processing and the like faced by the traditional network.

Disclosure of Invention

Aiming at the above defects or improvement demands of the prior art, the invention provides a digital twin device, a model evaluation system and a model operation method, which comprise three newly added digital twin basic function modules, a set of digital twin model evaluation system and two model operation feedback scheduling strategies, so that the establishment of physical properties such as digital twin model hierarchy, functions and the like can be realized, a communication equipment manufacturer can carry out simulation modeling on physical entity equipment while researching and producing the physical entity equipment, researching and producing the digital twin model, and the research and the development of the physical entity equipment are guided through experimental and even destructive experiments on a virtual digital twin model; in addition, the invention can also enable the digital twin model to select the running state according to the user demand, dynamically adjust the calculation, storage and transmission resources, can support the automatic running of the digital twin model with more application scenes when no task demand exists, and can accurately schedule the required model when the task demand exists, thereby meeting the task demand.

The embodiment of the invention adopts the following technical scheme:

in a first aspect, the present invention provides a digital twin device, including a data model interaction module, a model update module, and a resource call module, wherein:

the data model interaction module is used for storing the probability model and the simulation model data into a data pool, specifically, the data model interaction module comprehensively considers heterogeneous source data and stores the distributed double-link confidentiality into the data pool;

the model updating module is used for updating the probability model and the simulation model, specifically, the model updating module carries out accurate evaluation on the model, builds different directed graphs according to the logic execution relationship between the models, builds an automatic model operation mechanism and realizes the self-optimization and self-healing of model parameters;

the resource calling module is used for calling the probability model and the simulation model, specifically, the resource calling module predicts the model output under the current environment from the matching of the model and the user information, finally reasonably distributes the model according to storage, network and calculation resources, and finally establishes a reward and punishment mechanism for the model by verifying the performance of the model under the resource distribution scheme.

Further, the data model interaction module includes a data access layer, a weak centralization layer, and a blockchain layer, wherein:

four types of parameters needed by the data access layer storage model are as follows: the model outputs intermediate quantity and result, task demand data and provides a data table;

the weak centralization layer takes respective servers as centers to construct a distributed database, and is divided into a root node, a first sub-node and a second sub-node, wherein the root node is used for exchanging data with other servers in a blockchain layer, the first sub-node is used for storing data according to a demand model flow graph, and the second sub-node is used for asking the first sub-node for data;

the block chain layer comprises a data application layer specification formed by authorized access control and a safety protection mechanism.

Further, in the weakly centralized layer, a model consensus mechanism of digital twinning is commonly established by the root node, the first child node and the second child node, in particular:

the digital twin owners and participants jointly specify root nodes;

setting all the virtual machines deployed by the models in the region as second child nodes;

screening out nodes with high centrality from the second sub-nodes as first sub-nodes through the model flow diagram;

When the second sub-node B has data change, the data is sent to the first sub-node A to which the second sub-node B belongs, the first sub-node A sends the data information of the second sub-node B to all the first sub-nodes, and each first sub-node distributes the data information of the second sub-node B to all the second sub-nodes in the first sub-node.

The model updating module comprises a model evaluation, a directed graph construction, an automatic operation mechanism and model parameter updating, and specifically, the model evaluation evaluates four types of parameters required by a model, a demand model flow graph is constructed in the directed graph construction through a model input-output relation, the automatic operation mechanism of the model is selected according to the flow graph, and finally, the model parameters are comprehensively evaluated in a model parameter updating layer.

Further, the building the demand model flow graph in the directed graph building specifically includes:

evaluating the requirements and finding the minimum level of the model;

analyzing the change rule of different network conditions corresponding to the requirements along with time, and describing the change conditions of different networks in different time periods and the resource change conditions of a minimum hierarchy model under the network;

acquiring two similar regional physical topologies with highest association degree with a currently selected network, and constructing a digital twin virtual topology according to a topology hierarchy relation;

According to the constructed digital twin virtual topology and the minimum level of the model, a functional model with the same physical characteristic parameter calculated in the same level is found, and a functional model cluster is constructed;

based on the candidate points of the model, a model flow diagram is constructed by depending on the model input and output relations between the functional model clusters.

Further, the resource calling module comprises information collection, credit calculation, allocation strategy and punishment mechanism, specifically, the information collection, the credit calculation and the evaluation of the demand are carried out, the credit calculation is carried out on the model under the current demand, the allocation mechanism is used for obtaining an optimal model set according to the demand combination model, and when the demand is over, punishment is carried out on the model combination according to feedback comments and the model combination is stored in the data pool.

Further, the credit calculation for calculating the credit of the model under the current demand specifically includes:

acquiring resource requirements related to the aspects of transmission, storage and calculation required by the model, wherein if the model does not need any resource, the resource is represented by 0, and if any resource in the resource vector exceeds the currently available resource of the model, the model is considered to be not applicable to the current network environment;

acquiring the accuracy of the model when calculating a plurality of indexes of the current task, and considering that the model accuracy is limited in the network environment if the model accuracy cannot reach the requirements of the task in theory;

When the resource requirement and the accuracy meet the current network environment requirement, calculating the credit label of the model at the moment; the credit label of the model is related to the input and output of the model at the historical moment, the output of the model in the current network environment state is predicted through the output-input relation of the model at the historical moment, the predicted output of the model is compared with the task demand difference value, and the situation that the model can meet the task demand is estimated, namely the credit of the model is obtained.

Further, the data pool, the probability model, and the simulation model form a base component, wherein:

the data pool is used for storing digital twin data and comprises the following three types of data: the performance data directly reported from the equipment unit; service, configuration, alarm and fault data uploaded from the management and control unit; digital twin model operation data;

each of the probability model and the simulation model is composed of a plurality of atomic level models below or at the level, and each atomic level model corresponds to a model input and a model output.

In a second aspect, the present invention provides a digital twin model evaluation system, comprising a hierarchy property module, an instance property module, and a credit module, wherein:

The hierarchical characteristic module is used for defining the hierarchical characteristics of the digital twin model so as to realize establishment of physical properties of the digital twin model, and particularly is used for layering the digital twin model and comprises a network model layer, a network element model layer, a single-disk model layer, a module model layer, a device model layer and a material model layer, wherein each hierarchy is divided into a plurality of functional models according to different functions, the smallest particle model of each functional model is an atomic model, and each demand model is completed by a plurality of hierarchies and a plurality of functional model combinations;

the example characteristic module is used for dividing a digital twin model into a probability model and a simulation model according to different digital twin construction mechanisms, and specifically, the simulation model is formed by translating the internal mechanism through a mathematical mode after knowing the internal mechanism of the physical space communication equipment in detail; the probability model is obtained by modeling and fitting technology of the characterization data according to the input and output data of the measured communication equipment when the internal mechanism of the physical space communication equipment cannot be known;

the credit module is used for determining the attribute of the model which is good and bad in the corresponding model combination in the digital twin model instantiation combination process, if the model is good, the credit is high, if the model is bad, the credit is low, and the model parameters are corrected and combined according to the feedback of the credit, so that when the former model is combined with the latter model, the calculation errors of the two models caused by theory/statistics are reduced.

In a third aspect, the present invention provides a digital twin model operation method, including a demand model flow strategy, where the demand model flow strategy is: the resource calling layer is used for selecting different simulation models and probability model combination schemes according to different scenes; and the model updating layer constructs a directed graph according to the combination scheme selected by the resource calling layer, completes user task calculation, and finally feeds back simulation model and probability model output results to the resource calling layer.

Further, the resource calling layer selects different simulation models and probability model combination schemes according to different scenes, and the method specifically comprises the following steps:

evaluating the requirements and finding the minimum level of the model;

dividing the hierarchical model relationship according to the requirements, and constructing a virtual model topology corresponding to the physical network;

and finding out a model for calculating the same physical characteristic parameters according to the atomic level model, and constructing a model cluster.

Further, the model updating layer builds a directed graph according to the combination scheme selected by the resource calling layer, and the method specifically comprises the following steps: and constructing a demand model flow chart according to the model input-output relation in the virtual model topology.

In a fourth aspect, the present invention provides a digital twin model running method, including a model instance automatic optimization strategy, specifically:

refining the indexes of the demand tasks, and finding out all model sets capable of meeting task demands by adopting a step-by-step time-sharing calculation mode;

analyzing the flow direction weight according to the flow chart of the demand model, and finding out the optimal set in all sets;

and according to the model running condition of the demand feedback under the current environment, adjusting model weights of the demand model flow diagram.

Further, in the process of analyzing the flow direction weights according to the demand model flow chart and finding the optimal set in all sets, the selected simulation model can support as many tasks as possible while the calculation parameter accuracy is high, and the selected probability model uses the least resources to meet the task demands as soon as possible.

Further, according to the model running condition of the demand feedback in the current environment, the adjusting the model weight of the demand model flow chart specifically includes:

when the two models are combined for use, the situation that the error accumulation is overlarge is caused, and the connection between the models is deleted;

when two model combinations are used, the error accumulation is smaller or the error of the previous model is reduced, the weight between model connecting lines is increased, and the model combinations are recorded into a data pool in the form of block chains.

Compared with the prior art, the invention has the beneficial effects that:

(1) The invention provides a digital twin model evaluation system, which realizes establishment of physical properties such as hierarchy, functions and the like of the digital twin model, enables a communication equipment manufacturer to carry out simulation modeling on physical entity equipment while researching and producing the physical entity equipment, and guides and verifies research and development of the physical entity equipment by carrying out experimental and even destructive experiments on the virtual digital twin model.

(2) The invention provides two different model operation strategies and three new digital twin function connection modules, so that the digital twin model can select an operation state according to user requirements, dynamically adjust calculation, storage and transmission resources, can support the automatic operation of the digital twin model of more application scenes when no task is required, and can accurately schedule the required model when the task is required to meet the task requirement.

Drawings

In order to more clearly illustrate the technical solution of the embodiments of the present invention, the drawings that are required to be used in the embodiments of the present invention will be briefly described below. It is evident that the drawings described below are only some embodiments of the present invention and that other drawings may be obtained from these drawings without inventive effort for a person of ordinary skill in the art.

Fig. 1 is a schematic diagram of a theoretical CPS framework for digital twinning in the communication field provided in embodiment 1 of the invention;

FIG. 2 is a schematic diagram of a digital twin device according to embodiment 1 of the present invention;

FIG. 3 is a diagram illustrating an example of data stored in a data pool according to embodiment 1 of the present invention;

FIG. 4 is a schematic diagram of a probabilistic model/simulation model provided in embodiment 1 of the present invention;

FIG. 5 is a schematic diagram of the operation steps of the digital twin device according to embodiment 1 of the present invention;

FIG. 6 is a layered schematic diagram of a digital twin model provided in embodiment 2 of the present invention;

FIG. 7 is a schematic diagram of a demand model flow strategy provided in embodiment 3 of the present invention;

FIG. 8 is a flowchart of a method for operating a digital twin model according to embodiment 3 of the present invention;

FIG. 9 is a schematic diagram of an example automatic model optimization strategy provided in embodiment 4 of the present invention;

FIG. 10 is a flowchart of another method for operating a digital twin model according to embodiment 4 of the present invention.

Detailed Description

The present invention will be described in further detail with reference to the drawings and examples, in order to make the objects, technical solutions and advantages of the present invention more apparent. It should be understood that the specific embodiments described herein are for purposes of illustration only and are not intended to limit the scope of the invention.

The present invention is an architecture of a specific functional system, so that in a specific embodiment, functional logic relationships of each structural module are mainly described, and specific software and hardware implementations are not limited.

In addition, the technical features of the embodiments of the present invention described below may be combined with each other as long as they do not collide with each other. The invention will be described in detail below with reference to the drawings and examples.

Along with the process of digitization and intellectualization, the traditional communication network will realize automatic and intelligent management and control of the optical network of the machine in the future, but the management and control scheme formed based on the intelligent machine cannot be relieved or dared to directly issue instructions to the equipment side. Based on the situation, it is necessary to establish a twin optical network constructed based on a digital twin technology, so that the alarm degree simulates an optical network operation mechanism, synchronizes the optical network operation state in real time, verifies a management and control scheme constructed by a machine in the twin network, and issues the management and control scheme to entity equipment accurately, so that risk sharing caused by the management and control scheme formed by the machine intelligence can be avoided. Based on this, the following embodiments of the present invention provide a digital twin apparatus, a digital twin model evaluation system, and two digital twin model operation methods. The effects of realizing the integration isomorphism of the twinning world and the real data and the mutual cooperation of the twinning world and the real data are achieved by the real mapping deficiency, the real control deficiency and the virtual mapping real.

Example 1:

as shown in fig. 1, the present embodiment first teaches a theoretical CPS framework for digital twinning in the field of communication. The digital twin framework provided by the embodiment is applied to the field of communication and comprises an application space, an information space and a physical space. For example, as shown in the left side of fig. 1, in the field of cloud network collaboration, a physical space refers to computing, storing and transmitting resources, an information space includes an orchestrator and a controller, and an application space includes distributed machine learning and cloud virtual reality. The physical space reports the equipment state to the information space, the information space digitally describes the cloud network capacity through the virtual computing resource, the virtual storage resource, the virtual forwarding resource and the virtual bearing resource, and the application space analyzes the user demand according to the cloud network capacity. Based on the slicing, intention and other modes, the user demand and the resource allocation strategy are issued to the information space, and the information space utilizes the orchestrator and the controller to issue a management and control instruction to the physical space. For example, as shown in the right side of fig. 1, in the field of optical transmission, the physical space refers to an optical transmission link, which includes devices, links, network elements, and the like, the information space corresponds to a large management and control side, includes a conventional management and control unit, a digital twin engine, and the like, the application space corresponds to an application APP, includes optical network performance monitoring, SOPA, and the like, the optical transmission link reports an optical network operation state to the large management and control, analyzes the network operation state, uploads a result to the application APP, the application APP issues a user demand to the large management and control, and the large management and control issues a management and control instruction to the optical transmission link according to the demand.

Based on the above-mentioned theoretical CPS framework, this embodiment 1 proposes a digital twin device, and the architecture schematic diagram is shown in fig. 2. It should be noted that digital twinning includes a service layer, a network layer, and a connection layer in the communication field. The service layer is used for realizing the cooperation of the routing and control strategies for the user differentiated experience perception, and the connection layer fuses the resource supply. The digital twin device functional module comprises a device unit corresponding to the connection layer, an application unit corresponding to the service layer, a management and control unit corresponding to the network layer and an intelligent unit. The application unit corresponds to the independent APP in FIG. 2, and is closely related to the user demand, including intelligent operation and maintenance, planning simulation and the like. The management and control unit corresponds to the network management and control system in fig. 2, and is responsible for network management and control, including management and control interaction, network management, service management, network element management, southbound interfaces, and the like. In addition, it is also important that the digital twin platform in fig. 2 is a core part of the digital twin device, and it includes basic components (data pool, probability model, simulation model) and connection function components (resource call layer, model update layer, data model interaction layer).

For connecting functional components, the embodiment of the invention provides three new functional modules in the digital twin framework: a resource call module (corresponding to the resource call layer in fig. 2), a model update module (corresponding to the model update layer in fig. 2), and a data model interaction module (corresponding to the data model interaction layer in fig. 2). The traditional digital twin device only comprises a data pool, a model and an engine, and can not clearly describe how the model is mobilized in the digital twin to respond to the demands of users/tasks; an instantiation of how to update the model according to changes in the real-world environment; the method does not relate to data retrieval and storage in the digital twin model operation and feedback optimization process. According to the embodiment of the invention, the digital twin model reasonably schedules the model according to the user demands through the resource calling module, the model updating module and the resource calling module, adjusts the model instantiation process according to the current network environment, and finally realizes the function of efficiently storing the model data.

Specifically, in the preferred embodiment, the resource calling module is configured to call the probability model and the simulation model. And the resource calling module starts from the matching of the model and the user information, predicts the output of the model in the current environment, finally reasonably distributes the model according to storage, network and calculation resources, and finally establishes a reward and punishment mechanism for the model by verifying the performance of the model under the resource distribution scheme. The resource calling module comprises four functions of information acquisition, credit calculation, allocation strategy and punishment mechanism, wherein the information acquisition evaluates user demands, the credit calculation calculates the credit of a model under the current task/demand, the allocation mechanism obtains an optimal set of the model according to a user/demand combination model, and when the task/demand is finished, punishment is carried out on the model combination according to feedback opinion and the punishment is stored in a data pool. When the user/APP has tasks to be executed, a task request is sent to a resource calling layer, and an information acquisition module of the resource calling layer constructs an LFM (Looking For Model, model lookup table) which comprises task requirements of the user, a calculation scene, time and end time of starting the resources, various indexes of the task and index accuracy. After the credit calculation module receives the LFM, firstly judging which models output meet task indexes, judging whether various combinations of the models can execute task demands, if so, the credit calculation module drags resources, historical accuracy and the like required by the current model from a data pool, and constructs an information label and transmits the information label to the allocation strategy module. The distribution strategy module selects a combination of multiple models according to model information in the information label constructed by the credit calculation, determines how to finish tasks step by step in a time-sharing way under the condition of limited resources, then sends the tasks and the model information to the probability model and the simulation model module, invokes the model combination to finish task calculation, and finally compares the model credit with model information label information according to the actual model use information in the reward and punishment mechanism module.

The calculating the credit of the model under the current demand through credit calculation specifically comprises the following steps: acquiring resource requirements related to the aspects of transmission, storage and calculation required by the model, wherein if the model does not need any resource, the resource is represented by 0, and if any resource in the resource vector exceeds the currently available resource of the model, the model is considered to be not applicable to the current network environment; acquiring the accuracy of the model when calculating a plurality of indexes of the current task, and considering that the model accuracy is limited in the network environment if the model accuracy cannot reach the requirements of the task in theory; when the resource requirement and the accuracy meet the current network environment requirement, calculating the credit label of the model at the moment; the credit label of the model is related to the input and output of the model at the historical moment, the output of the model in the current network environment state is predicted through the output-input relation of the model at the historical moment, the predicted output of the model is compared with the task demand difference value, and the situation that the model can meet the task demand is estimated, namely the credit of the model is obtained.

In the preferred embodiment, the model updating module is used for updating the probability model and the simulation model, the model updating module carries out accurate evaluation on the model, different directed graphs are built according to the logic execution relation between the models, and an automatic model operation mechanism is built to realize the self-optimizing and self-healing of the model parameters. The model updating module is divided into four functions of model evaluation, directed graph construction, automatic operation mechanism and model parameter updating. The model evaluation evaluates four types of data required by the model, a user/demand model flow diagram is constructed in the directed diagram construction through the model input-output relation, a model automatic operation mechanism is selected according to the flow diagram, finally, comprehensive evaluation is carried out on model parameters in a model parameter updating layer, the model parameters are corrected and combined according to the use condition of the model called after each task, and when the former model is combined with the latter model, calculation errors caused by theory/statistics of the former model and the latter model are reduced, so that model combination is more efficient in the current environment, and model precision is greatly improved.

The building the demand model flow chart in the directed graph construction specifically includes: evaluating the requirements and finding the minimum level of the model; analyzing the change rule of different network conditions corresponding to the requirements along with time, and describing the change conditions of different networks in different time periods and the resource change conditions of a minimum hierarchy model under the network; acquiring two similar regional physical topologies with highest association degree with a currently selected network, and constructing a digital twin virtual topology according to a topology hierarchy relation; according to the constructed digital twin virtual topology and the minimum level of the model, a functional model with the same physical characteristic parameter calculated in the same level is found, and a functional model cluster is constructed; based on the candidate points of the model, a model flow diagram is constructed by depending on the model input and output relations between the functional model clusters. Wherein the connection weight represents an influence factor of the front model output on the rear model, and the influence factor comprises a degree of influence on the output parameters in the front model and a degree of similarity between the total output parameters of the front model and the total input parameters of the rear model.

In the preferred embodiment, the data model interaction module is used for storing the probability model and the simulation model data into a data pool, and the data model interaction module comprehensively considers heterogeneous source data and stores the distributed double-chain confidentiality into the data pool. The data model interaction module is divided into a data access layer, a weak centralization layer and a block chain layer. The data access layer stores four types of parameters required by the model, and the model outputs intermediate quantity and results, task demand data and provides a data table. At the weak central layer, each carrier, manufacturer builds a distributed database centered on each (e.g., centered on its own server). The method comprises the steps of dividing the method into a root node, a first child node and a second child node. And respectively storing the data into the first child nodes according to the user/demand model flow diagram. The second child node requests data from the first child node. The root node is responsible for exchanging data with the rest of the company, vendor, at the blockchain level. The block chain layer contains data application layer specifications formed by authorized access control and security protection mechanisms.

In the weak centralization layer, a digital twin model consensus mechanism is commonly established by a root node, a first child node and a second child node, and the method is specifically: the digital twin owners and participants jointly specify root nodes; setting all the virtual machines deployed by the models in the region as second child nodes; screening out nodes with high centrality from the second sub-nodes as first sub-nodes through the model flow diagram; when the second sub-node B has data change, the data is sent to the first sub-node A to which the second sub-node B belongs, the first sub-node A sends the data information of the second sub-node B to all the first sub-nodes, and each first sub-node distributes the data information of the second sub-node B to all the second sub-nodes in the first sub-node.

In addition, the first sub-node can also be updated, specifically, when each node is started, the degree centrality of the first sub-node is calculated, and the first sub-node is searched by broadcasting in a local area; any node monitors the broadcast message and returns the ranking of the node in the network centrality ranking list; if the self is not in the first position, the self is determined to belong to a second child node, and the list is stored; if the self-centering is highest, the self-centering is determined to belong to a new first child node, information index synchronization and resource acquisition routing are carried out with the original first child node, and the latest centering ranking list is broadcasted.

For each child node, when the network environment changes, corresponding processing is performed according to the changing condition, and the specific steps are as follows: when the first sub-node part is damaged, the second sub-node part is damaged or newly added, the rest second sub-nodes reselect the first sub-node according to the updating step of the first sub-node; when the first sub-node is not damaged and the second sub-node is partially damaged, the first sub-node updates a second sub-node table according to heartbeat feedback of the second sub-node; when the first child node is not damaged and the second child node x is newly added, x firstly broadcasts own information, and if no feedback is given to any first child node, the first child node is used as the first child node to construct a region; when all first child nodes receive x information, calculating the logic distance between the node and the x node, and calculating the distance d determined by the minimum time for guaranteeing the data transmission of the model, if the logic distance is smaller than d and x exists in the original list, moving x to the tail of the queue, and indicating that the node x is updated recently; if the logic distance is smaller than d and the node x is not recorded in the original queue, judging the communication condition of the node x and the first sub-node, if the node x has a corresponding relationship, adding the node x into the domain corresponding to the first sub-node, otherwise, neglecting the node x.

Three new functional modules proposed by the embodiment of the present invention are described in detail above, and the basic components (data pool, probability model, simulation model) in this embodiment are described in detail below.

In the preferred embodiment, the data pool is used to store digital twin data, including the following three types of data: 1. the performance data directly reported from the equipment unit; 2. service, configuration, alarm and fault data uploaded from the management and control unit; 3. the digital twin model operation data includes intermediate quantities and results of the model. Such as shown in fig. 3, an exemplary diagram of data stored in a digital twin data pool. The performance data of the model operation related to the OA (amplifier) comprises the number of OAs, the model number of the OAs and single wave power of the first OA. The data related to topology in the service data comprises a network element ID, a network element name, a network element abscissa and a network element ordinate. The intermediate quantity and the result of the monitoring model in the model operation data comprise OA single wave power, optical fiber single wave gain and OA noise index.

In the preferred embodiment, the probabilistic model/simulation model is shown in FIG. 4, and each probabilistic model/simulation model is composed of a plurality of atomic level models below or at the level, each atomic level model corresponding to a model input and output.

Thus, the probability, simulation model can be described in a unified form as:

：/>

wherein,,

indicate->

Probability/simulation model.

The operation steps of the digital twin device provided in this embodiment are as shown in fig. 5: the device uploads performance data to a management and control (network management and control system) and a digital twin platform, a digital twin data pool is used for managing and controlling synchronous service data, a digital twin model is used for taking out data from the data pool, the digital twin model is used for updating model parameters according to acquired data, the digital twin model enters an operation state according to task requirements as required, and after the digital twin model is called, the output parameters are stored in the data pool.

For example, at the current moment, the device collects data, and when the user issues a task of monitoring the link performance, the implementation steps should be as follows:

1. the device uploads the data required by the performance monitoring model: the first OA single wave power and the last OA single wave power are equal to the network management and control system and the digital twin platform.

2. The digital twin data pool synchronizes service data from a network management and control system, and comprises a network element ID, a network element abscissa and a network element ordinate.

3. The digital twin performance monitoring model takes out data from the data pool, takes out the performance data uploaded by the equipment just and the model intermediate quantity and result storage data at the previous moment, and comprises the performance data: first OA single wave power, last OA single wave power and the like, and performance monitoring model operation data: OA single wave power, optical fiber single wave gain, OA noise figure, etc.

4. The digital twin performance monitoring model updates performance monitoring model parameters according to the acquired data.

5. And selecting a digital twin performance monitoring model to enter an operating state according to the monitoring model requirement of a user.

6. The digital twin performance monitoring model is called, and output parameters such as OA single wave power, optical fiber single wave gain and OA noise index are stored in the data pool.

In summary, the digital twin framework provided by the invention is additionally provided with three functional modules: the system comprises a resource calling module, a data model interaction module and a model updating module. And finally, reasonably distributing the model according to storage, network and computing resources, and finally establishing a reward and punishment mechanism for the model by verifying the performance of the model under the resource distribution scheme.

According to the model updating module provided by the invention, a user/demand model operation flow chart is constructed according to the input-output relation of the twin model, a weight relation is established between the model and the model, which model is operated automatically and which model is not operated in a legal way, and the minimum model is enabled to support more application scenes under the condition of limited resources.

The newly added data model interaction module comprehensively considers heterogeneous source data and stores the distributed double-chain confidentiality into a data pool. The data storage efficiency is improved, and meanwhile, the data storage and use safety are greatly enhanced.

Example 2:

based on the digital twin device provided in embodiment 1, embodiment 2 further provides a digital twin model evaluation system, where the digital twin model evaluation system defines three attributes of the hierarchical property, the instance property and the credit of the digital twin model, that is, may include three modules of the hierarchical property module, the instance property module and the credit module.

The hierarchy characteristic module is used for defining the hierarchy characteristic of the digital twin model so as to establish the physical attribute of the digital twin model. Referring to fig. 6, a hierarchical characteristic of the digital twin model is shown as a hierarchical schematic diagram, where the hierarchical characteristic divides the digital twin model into six layers, and from large to small, the digital twin model can be divided into six layers including a network, a network element, a single disk, a module, a device, and a material, and each layer corresponds to one type of model, that is, can correspond to a network model layer, a network element model layer, a single disk model layer, a module model layer, a device model layer, and a material model layer. Each level may be subdivided into multiple functional models according to different functions, for example, a single disk level may be subdivided into functions: master control disk model, business disk model, computing disk model. The minimum particle model of each layer of functional model is an atomic level model, for example, on a single disk side, an OA optical amplifying disk atomic level model, an OGC computing disk atomic level model and the like are divided according to functions, and for example, a network element side performance monitoring model is formed by combining a plurality of single disk atomic level models, including an OA optical amplifying disk atomic level model, an optical fiber atomic level model and the like. Each application/demand model requires multiple levels, multiple functional model combinations to complete.

In this embodiment, the example feature module divides the digital twin model into a probability model and a simulation model according to different digital twin construction mechanisms, and the simulation model is formed by translating the internal mechanism through a mathematical mode (for example, fourier integration method, etc.) after knowing the internal mechanism of the physical space communication device in detail; the probability model is obtained by modeling and fitting technology for the characterization data, such as the Monte Carlo technology, according to the input and output data of the communication equipment measured in a laboratory when the internal mechanism of the physical space communication equipment cannot be known. Multiple probabilistic models, simulation models, may compute the same physical characteristic parameter. For example, BER can simulate ten thousands times of code input and output conditions through Monte Carlo, and the ratio of the current network output error code to the output code is counted; and the BER at the current network moment can be calculated by analyzing the electric propagation, photoelectric change and optical propagation processes through Fourier change. The probability model has the advantages of multiple performances, quick calculation time, less occupied resources and lower accuracy. The simulation model has the advantages of long calculation time, more occupied resources and high accuracy. The vast majority of applications/requirements require the use of probabilistic models and simulation models in combination.

In this embodiment, the credit module is configured to determine, in an instantiation and combination process of the digital twin model, whether the model is good or bad in the corresponding model combination, if so, the credit is high, if so, the credit is low, and correct and combine model parameters according to feedback of the credit, so that when the former model is used in combination with the latter model, calculation errors of the two models due to theory/statistics are reduced. In particular, because of the mechanism theory of the digital twin model or the limitation of the analog curve, errors between the calculated value and the true value are often caused. For example, when multiple models are matched, the calculation deviation of the current model a is too large, and when part of the model B is matched with the previous model, two situations can be caused according to the mechanism theory or the simulation curve of the matched model B: (1) The calculation deviation is further larger, for example, when the nonlinear effect in the C96 wave system is calculated, if the four-wave mixing calculation deviation of the previous section OMS is too large, the four-wave mixing deviation of the system is multiplied when the GN model is adopted by the next section OMS model. (2) The calculation deviation is small or unchanged, for example, when the nonlinear effect is calculated as well, if the calculation of the front OMS Duan Si wave mixing is large, the four wave mixing deviation is small or unchanged when the EGN model is adopted by the OMS model of the next section. Then in the first case the fit B model credit is low and in the second case the fit B model credit is high.

In summary, the digital twin model evaluation system provided by the invention realizes establishment of physical properties such as hierarchy, functions and the like of the digital twin model, enables a communication equipment manufacturer to carry out simulation modeling on physical entity equipment while developing and producing the physical entity equipment, and can give guiding comments considering which level model is established and whether probability or simulation model is given in developing and producing the digital twin model, and can also guide and verify development and development of the physical entity equipment by carrying out experimental and even destructive experiments on the virtual digital twin model. In addition, in the combined use of the twin models, model parameters are corrected and combined according to the use condition of each task call model, so that when the former model is combined with the latter model, calculation errors of the two models caused by theory/statistics are reduced, the combined use of the models becomes more efficient in the current environment, and the model precision is greatly improved.

Example 3:

based on the digital twin device provided in embodiment 1 and the digital twin model evaluation system provided in embodiment 2, embodiment 3 provides a digital twin model operation method, where the model operation method includes a demand model flow strategy, and referring to fig. 7, a flow strategy diagram of the demand model is shown, where the flow strategy is: resource call layer-simulation, probability model-model update layer-simulation, probability model-resource call layer. The resource calling layer selects different simulation models and probability model combination schemes according to different scenes of a user (corresponding to the steps 301-303), the model updating layer selects the combination schemes according to the resource calling layer, a directed graph is constructed (corresponding to the step 304), user task calculation is completed, and finally output results of the simulation models and the probability models are fed back to the resource calling layer. Specifically, referring to fig. 8, the flow of the digital twin model operation method in this embodiment specifically includes the following steps.

Step 301: the user/demand is evaluated, the minimum level of the model is found, and this is done in the resource call layer. For example, the user/demand calculates nonlinearity in an optical link, the minimum particle model is selected by evaluating the user demand precision, the network element level model is selected with the precision requirement of about 1.5dB, the single-disk level or module level model is selected with the precision requirement of about 1.0dB, and the device material level model is selected with the precision requirement of less than 0.5 dB.

Step 302: and dividing the hierarchical model relationship according to the user/requirement, and constructing a virtual model topology corresponding to the physical network, wherein the step is completed by an information acquisition module in a resource calling layer. For example, a single disk level model is selected in step 301. The optical link may be first divided into a plurality of OTS, and further divided into a plurality of OMS plus WSS disc models, each OMS further including a plurality of OCHs plus a composite wave-splitting disc model, each OCH corresponding to a plurality of optical fiber atomic models and an OA amplifying disc atomic model. And forming the single-disk-level atomic model into a physical entity network.

Step 303: according to the atomic level model, a model for calculating the same physical characteristic parameters is found, a model cluster is constructed, and the step is completed by a resource calling layer. For example, calculate a first OA disk output OSNR, including the following functional models: an OSNR monitoring function model, an OSNR prediction function model, an OA amplification mechanism function model, and the like. And forming a model cluster of the first OA by combining all the functional models. There is a coincidence function model between model clusters, e.g., an OSNR monitoring model can be contained in almost all OA clusters.

Step 304: and constructing a user/demand model flow chart according to the model input-output relation in the virtual model topology. The step is completed according to a directed graph construction module in the model updating layer. The minimum particle model found above, for example, the first OA output is the input of the first segment of fiber model, and each model in the first OA cluster is connected to each model in the first segment of fiber cluster by directional arrows. And finally, constructing a model operation flow chart by taking an atomic level model as a unit by depending on the virtual topology of the model. And finally, feeding back an output result of each model in the operation flow chart to a resource calling layer.

In summary, based on the twin model operation strategy of the user task and the current environment resource, the network topology is analyzed by matching with the user task requirement, the model cluster meeting the user requirement is selected, the models are combined according to different model input-output relations from the model cluster, and an operation flow diagram of the model is constructed. Therefore, efficient calculation of the model is realized, and the user calculation requirement is met as soon as possible under the condition that the current network condition is met.

Example 4:

based on the digital twin device provided in embodiment 1 and the digital twin model evaluation system provided in embodiment 2, embodiment 4 provides a digital twin model operation method, where the model operation method includes a model instance automatic optimization strategy, and referring to fig. 9, a strategy diagram is shown for automatically optimizing a model instance, where the strategy flow direction is as follows: model update layer-simulation, probabilistic model-data model interaction layer-simulation, probabilistic model-model update layer. The model updating layer calls simulation and probability model calculation through the established directed graph, the probability and simulation model sends a request for taking out the current network state data to the data model interaction layer, the data model interaction layer responds to the data required by the model request to return the model, the simulation and probability model calculates the user requirement based on the returned data, the model parameters are optimized, and finally the data such as the running state and the result are fed back to the model updating layer. Specifically, referring to fig. 10, the flow of the digital twin model operation method in this embodiment specifically includes the following steps.

Step 401: and (3) refining the task indexes of the user/demand, and finding out all model sets capable of meeting task demands by adopting a step-by-step time-sharing calculation mode, wherein the step is completed by a model updating layer. For example, in the performance monitoring model, the integration model is composed of an OSNR prediction model and a WSS atomic level model, wherein the OSNR prediction model belongs to a network element level probability model. The WSS atomic-level model belongs to a single-disk-level atomic simulation model. And the second set consists of an OSNR monitoring model, a combining and dividing wave disc model and a fault diagnosis model. The OSNR monitoring model and the fault diagnosis model belong to a network element level probability model. The composite wave-dividing disk model belongs to single-disk-level atomic simulation model.

Step 402: and analyzing the flow direction weight according to the flow direction diagram of the demand model, and finding the optimal set in all sets, wherein the step is completed by a model updating layer module. In the step, the selected simulation model is required to be ensured to have high calculation parameter accuracy and can support as many tasks as possible; it is also necessary to ensure that the selected probabilistic model uses the least resources to meet the user/task demands as soon as possible. For example, in step 401, the number of models supported by the WSS atomic level model is greater than the number of models supported by the composite spectral disk model in set two, and the OSNR prediction model is less than the OSNR monitoring model and the fault diagnosis model in set two, so that set one is the optimal set selected in step 401.

Step 403: according to the model running condition of the user/demand feedback in the current environment, the model weight of the user/demand model flow diagram is adjusted, and the steps are completed by a model updating layer and a data model interaction layer. In the step, when the combination of the two models is used, the error accumulation is overlarge, and the connection line between the models is deleted; when two model combinations are used, the error accumulation is smaller or the error of the previous model is reduced, the weight between model connecting lines is increased, and the model combinations are recorded into a data pool in the form of block chains. For example, in step 401, an OSNR prediction model and a WSS atomic level model are collected, and when the OSNR prediction model is combined with the WSS model and used together, both the OSNR prediction error and the WSS model error are reduced, the connection weight between the OSNR prediction model and the WSS model is increased, and the connection weight is stored in the data pool in the form of a blockchain. When the OSNR monitoring model and the fault diagnosis model in the second set are combined, OSNR monitoring errors are caused, and fault judgment errors are increased, so that a connecting line between the OSNR monitoring model and the fault diagnosis model is deleted.

In summary, the model instance automatic optimization strategy is used in cooperation with a model automatic operation mechanism, a model data weak centralized storage and the like, so that interaction between the model and the data is realized, and the problems of real mapping, data composition use, model parameter updating, confidentiality and storage in the real process of virtual mapping and real mapping can be solved, so that the digital twin model and the database are mutually matched, and the functions of comprehensive application, self-optimizing, self-healing, quick access and the like of the model data are realized.

Those of ordinary skill in the art will appreciate that all or a portion of the steps in the various methods of the embodiments may be implemented by a program that instructs associated hardware, the program may be stored on a computer readable storage medium, the storage medium may include: read Only Memory (ROM), random Access Memory (RAM), magnetic disk or optical disk.

The foregoing description of the preferred embodiments of the invention is not intended to be limiting, but rather is intended to cover all modifications, equivalents, and alternatives falling within the spirit and principles of the invention. What is not described in detail in this specification is prior art known to those skilled in the art.

Claims

1. The digital twin device is characterized by comprising a data model interaction module, a model updating module and a resource calling module, wherein:

2. The digital twinning apparatus of claim 1, wherein the data model interaction module comprises a data access layer, a weak centralization layer, and a blockchain layer, wherein:

3. The digital twinning apparatus of claim 2, wherein in the weakly centralized layer, a model consensus mechanism for digital twinning is commonly established by a root node, a first child node, and a second child node, in particular:

the digital twin owners and participants jointly specify root nodes;

4. The digital twin device according to claim 1, wherein the model updating module comprises a model evaluation, a directed graph construction, an automatic operation mechanism and a model parameter updating, the model evaluation evaluates four types of parameters required by a model, a demand model flow graph is constructed in the directed graph construction through a model input-output relationship, the automatic operation mechanism of the model is selected according to the flow graph, and finally the model parameters are comprehensively evaluated in a model parameter updating layer.

5. The digital twin device according to claim 4, wherein the constructing a demand model flow graph in the directed graph construction specifically comprises:

evaluating the requirements and finding the minimum level of the model;

6. The digital twin device according to claim 1, wherein the resource calling module includes information collection, credit calculation, allocation policy and penalty mechanism, specifically, the information collection evaluates the demand, the credit calculation calculates the credit of the model under the current demand, the allocation mechanism combines the models according to the demand to obtain an optimal set of models, and when the demand is over, a punishment is made to the model combination according to the feedback opinion, and the punishment is stored in the data pool.

7. The digital twin device as defined in claim 6, wherein the credit calculation of the credits for the current demand model specifically comprises:

8. The digital twinning apparatus of claim 5, wherein the data pool, the probabilistic model, and the simulation model form a base component, wherein:

9. A digital twin model evaluation system applied to the digital twin device according to any of claims 1-8, comprising a hierarchy property module, an instance property module, and a credit module, wherein:

10. A digital twin model operation method, applied to a digital twin device according to any of claims 1-8, comprising a demand model flow strategy, the demand model flow strategy being: the resource calling layer is used for selecting different simulation models and probability model combination schemes according to different scenes; and the model updating layer constructs a directed graph according to the combination scheme selected by the resource calling layer, completes user task calculation, and finally feeds back simulation model and probability model output results to the resource calling layer.

11. The digital twin model operation method according to claim 10, wherein the resource calling layer selects different simulation models and probability model combination schemes according to different scenes, and the method specifically comprises:

evaluating the requirements and finding the minimum level of the model;

12. The digital twin model operation method according to claim 11, wherein the model update layer constructs a directed graph according to a combination scheme selected by the resource call layer, specifically including: and constructing a demand model flow chart according to the model input-output relation in the virtual model topology.

13. A digital twin model operation method, characterized in that it is applied to a digital twin device according to any of claims 1-8, and comprises a model instance automatic optimization strategy, in particular:

14. The method according to claim 13, wherein the flow direction weights are analyzed according to the demand model flow chart, so that the selected simulation model can support as many tasks as possible while the calculation parameter accuracy is high in the process of finding the optimal set in all sets, and the selected probability model uses the least resources to meet the task demands as soon as possible.

15. The method for operating a digital twin model according to claim 14, wherein the adjusting the model weight in the flow chart of the demand model according to the model operating condition of the demand feedback in the current environment specifically comprises: