Generation method and generation equipment of digital twin network, storage medium and product

Abstract

The invention discloses a digital twin network generation method and device, a storage medium and a product. The method comprises the following steps: acquiring actual sample data and random noise extracted for a target sub-network digital twin model in a network twin environment; inputting actual sample data and random noise into the diffusion model, completing Gaussian diffusion, and obtaining noise-added data; randomly sampling a noise from the noise-added data, performing denoising and sampling operation on the noise to recover data distribution from the noise, and ending denoising until the recovered data distribution meets a preset distribution condition; wherein the preset distribution condition comprises Gaussian distribution; and dynamically correcting the target sub-network digital twinning model according to the data after noise addition and the sensing data when de-noising is ended so as to update the network twinning environment.

Description

Technical Field

[0001] This application relates to the field of computer technology, and in particular to a method, device, storage medium, and product for generating digital twin networks. Background Technology

[0002] A communication system is an extremely complex and vast system. The entire end-to-end network contains various network elements, each with multiple different job roles and focuses on different aspects.

[0003] Current implementations of network digital twins involve creating digital twins for each component of the network individually (such as the core network and network topology). Due to the complexity of the overall network, it is difficult to create a single digital twin for the entire network. Summary of the Invention

[0004] This application provides a method, device, storage medium, and product for generating a digital twin network.

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

[0006] This application provides a method for generating a digital twin network, the method comprising:

[0007] Obtain actual sample data and random noise extracted from the digital twin model of the target subnetwork in the network twin environment;

[0008] The actual sample data and the random noise are input into the diffusion model to complete Gaussian diffusion and obtain the data with added noise.

[0009] A noise sample is randomly selected from the noisy data. The noise is then denoised and sampled to recover the data distribution. The denoising process ends when the recovered data distribution meets a preset distribution condition. The preset distribution condition includes a Gaussian distribution.

[0010] Based on the data after noise addition and the perceived data at the end of denoising, the target sub-network digital twin model is dynamically corrected to update the network twin environment.

[0011] In the above scheme, the network twin environment is generated by multiple sub-network digital twin models, and the actual sample data belongs to the behavioral observation data of the multiple sub-network digital twin models on their respective network tasks.

[0012] In the above scheme, the step of inputting the actual sample data and the random noise into the diffusion model to complete Gaussian diffusion and obtain the noisy data includes:

[0013] The actual sample data and the random noise are input into the diffusion model. The actual sample data is randomly noise-added through a preset Markov chain until the actual sample data tends to a Gaussian distribution, at which point the diffusion ends, and the noisy data is obtained.

[0014] In the above scheme, the data distribution obtained by the target sub-network digital twin model at each time step is related to the result distribution of the previous sub-network digital twin model in the network twin environment, as well as the diffusion kernel of the target sub-network digital twin model.

[0015] The method in the above scheme further includes:

[0016] The sensing data is obtained from the target sub-network digital twin model based on the result distribution and the diffusion kernel;

[0017] The error is calculated based on the noisy data and the perceived data until the target sub-network digital twin model converges.

[0018] The method in the above scheme further includes:

[0019] Preset parameters are introduced into the location space of different sub-network digital twin models in the network twin environment. The preset parameters are used to control the amount of noise injected into the location space of the different sub-network digital twin models.

[0020] By continuously adding noise to disrupt the training data, randomly sampled noise is fed into the different sub-network digital twin models. The deviation between the output of each sub-network digital twin model and the actual collected results is modeled to obtain the diffusion model.

[0021] In the above scheme, the deviation includes a combination of deterministic and random components.

[0022] A device for generating a digital twin network includes: a communication interface and a processor; wherein the processor is configured to: acquire actual sample data and random noise extracted for a target sub-network digital twin model in a network twin environment;

[0023] The actual sample data and the random noise are input into the diffusion model to complete Gaussian diffusion and obtain the data with added noise.

[0024] A noise sample is randomly selected from the noise-added data. The noise is then denoised and sampled to recover the data distribution from the noise. The denoising process ends when the recovered data distribution meets a preset distribution condition. The preset distribution condition includes a Gaussian distribution.

[0025] This application also provides a storage medium storing a computer program thereon, characterized in that the computer program, when executed by a processor, implements the steps of any of the above methods.

[0026] This application also provides a computer product, including a computer program, which, when executed by a processor, implements the steps of any of the above methods.

[0027] This application provides a method, device, storage medium, and product for generating a digital twin network. The method includes: acquiring actual sample data and random noise extracted from a target sub-network digital twin model in a network twin environment; inputting the actual sample data and random noise into a diffusion model to perform Gaussian diffusion, obtaining noisy data; randomly sampling a noise from the noisy data, and recovering the data distribution from the noise through denoising and sampling operations, until the recovered data distribution meets a preset distribution condition, at which point denoising ends; wherein the preset distribution condition includes a Gaussian distribution; and dynamically correcting the target sub-network digital twin model based on the noisy data and the perceived data at the end of denoising, thereby updating the network twin environment. This application introduces a diffusion model to perform self-correction for each network digital twin model in the network twin environment, promoting cross-communication between network digital twin models and gradually generating an overall optimized network twin environment; it solves the problem in related technologies that it is difficult to establish a single digital twin for the entire network. Attached Figure Description

[0028] Figure 1 This is a schematic diagram illustrating how a network digital twin is obtained by digitally twinning each component of the network in related technologies.

[0029] Figure 2 This is a flowchart illustrating a method for generating a digital twin network according to an embodiment of this application;

[0030] Figure 3 This is a schematic diagram illustrating a process for implementing a digital twin model based on a diffusion model, according to an embodiment of this application.

[0031] Figure 4 This is a schematic diagram of the structure of a digital twin network generation device according to an embodiment of this application;

[0032] Figure 5 This is a schematic diagram of the structure of the digital twin network generation device according to an embodiment of this application. Detailed Implementation

[0033] The present application will now be described in further detail with reference to the accompanying drawings and embodiments.

[0034] Communication systems are extremely complex and vast systems. The entire end-to-end network contains various network elements, each with multiple different roles and focuses. Since current communication networks are built on dedicated hardware with very sophisticated software systems containing detailed and real-time interactively updated data, it is possible to simplify and select appropriate components to build a virtual end-to-end fully virtualized twin system on a general-purpose server.

[0035] In real-world scenarios, firstly, each base station in the network has a precise model within the network management system. These models include not only engineering parameters such as site location, latitude and longitude, antenna height, azimuth, and elevation angles, but also numerous functional characteristics, thousands of wireless parameters used to improve network performance, and ground parameters for collaborative operation with other network elements. Base stations are also extremely busy; besides providing services to mobile phones, they aggregate all collected information into performance indicators, also known as Key Performance Indicators (KPIs), to evaluate user experience quality. Therefore, the entire system has a comprehensive understanding of the content and flow of information, and this data is sufficient to construct a digital model of the base station. Secondly, the core network stores all data for each user, including their location, service plan, preferred applications, mobile phone, and call duration. Thirdly, terminals constantly interact with the network. Besides the actual data transmitted, this includes signaling for system control, detecting wireless signals in various frequency bands, and reporting the signal quality they measure.

[0036] However, due to the complexity of the overall network, it is usually difficult to create a single digital twin for the entire network.

[0037] See Figure 1As shown, current network digital twin implementations involve digitally twinning each component of the network (e.g., mobile phones, wireless channels, base stations, transmission, core network, and applications). These components encompass the core network, network topology, Radio Access Network (RAN), and user behavior. The implementation method for each component's digital twin is customized based on the architecture and function of its respective network element. For example, the core network is primarily based on communication protocols and utilizes conventional programming development from open-source projects for network digital twinning. Network topology typically employs graph neural networks such as Principal Component Analysis (PCA) for network digital twinning. The RAN side utilizes simulation software such as Matlab for network digital twinning. User behavior is achieved using Bayesian models and Recurrent Neural Networks (RNNs) for network digital twinning.

[0038] However, the current network digital twin solutions mentioned above have the following two drawbacks:

[0039] (1) Coordinating and configuring the unified process between the twins composed of various networks is difficult. (2) Due to the differences in the corresponding functions in different scenarios, it is difficult to maintain the complete overall function after expanding and customizing the new functions of each type of network.

[0040] In the field of network digital twins, current traditional network digital twin models are often limited by their inherent understanding capabilities, lacking insight into external or other components. Therefore, coordinating and configuring unified processes among the various network-component twins is difficult. This application proposes a new network digital twin network generation method that introduces a novel diffusion-model-based network digital twin fusion architecture. This promotes cross-communication between network digital twin models, enhancing collective understanding during the twinning process. Through this architecture, the digital twin models of each network component can coordinate the reasoning of other models, thereby better coordinating and improving their own solutions.

[0041] This application provides a method for generating a digital twin network, applied to a digital twin network generation device. This generation device can be a core diffusion model control network element device, such as... Figure 2 As shown, the method includes:

[0042] Step 201: Obtain actual sample data and random noise extracted from the digital twin model of the target subnetwork in the network twin environment.

[0043] In practical applications, the target sub-network digital twin model is any twin model in the network twin environment. The extracted actual sample data is represented as (NetStatus, xij); the random noise is represented as λi, where λi is randomly generated noise, and i and j are positive integers.

[0044] Step 202: Input the actual sample data and random noise into the diffusion model to complete Gaussian diffusion and obtain the data with added noise.

[0045] In practical applications, actual sample data and random noise are input into the diffusion model. The data is then corrupted by continuously adding noise, completing a T-step Gaussian diffusion process, transforming it into a pure noise structure, thus obtaining the noisy data. This application achieves a complete digital twin of the network based on the diffusion model.

[0046] Step 203: Randomly sample a noise from the noise-added data, and perform denoising and sampling operations on the noise to recover the data distribution from the noise until the recovered data distribution meets the preset distribution conditions and the denoising ends; wherein the preset distribution conditions include Gaussian distribution.

[0047] In practical applications, data is recovered through a learning inversion process. This process is achieved by learning through an extended model based on a diffusion network, thereby generating a digital twin environment. This application randomly samples noise from the noisy data and performs inverse diffusion. By denoising and sampling the noise, the data distribution is recovered from the noise. Denoising ends when the recovered data distribution meets preset distribution conditions, thus completing the reconstruction of the network digital twin's behavioral observation data on various network tasks. The preset distribution conditions include a Gaussian distribution.

[0048] Step 204: Based on the data after adding noise and the perceived data at the end of denoising, dynamically correct the target sub-network digital twin model to update the network twin environment.

[0049] This method decomposes a holistic digital process into many subdivided diffusion denoising steps through sampling. Each network part's digital twin step is self-corrected. Based on the data after denoising and the perceived data at the end of denoising, the target sub-network's digital twin model is dynamically corrected until the network converges, gradually generating a complete and optimized overall digital twin network to update the network twin environment. This approach, which decomposes a holistic digital process into many subdivided diffusion denoising steps through sampling, with each network part's digital twin step being self-corrected, gradually generates a complete and optimized overall digital twin network. This solves the problem of difficult unified coordination and configuration of processes between network twins, realizes the updating of the overall network twin environment, and effectively improves the performance and efficiency of the model network digital application in practice.

[0050] This application provides a method for generating a digital twin network. The method involves acquiring actual sample data and random noise extracted from a target sub-network digital twin model within a network twin environment; inputting the actual sample data and random noise into a diffusion model to perform Gaussian diffusion, resulting in noisy data; randomly sampling a noise source from the noisy data; and performing denoising and sampling operations on the noise source to recover the data distribution from the noise, continuing until the recovered data distribution meets preset distribution conditions, including a Gaussian distribution; and dynamically correcting the target sub-network digital twin model based on the noisy data and the perceived data at the end of denoising to update the network twin environment. This application introduces a diffusion model to enable self-correction for each network digital twin model in the network twin environment, promoting cross-communication between network digital twin models and gradually generating an overall optimized network twin environment; thus solving the problem in related technologies where it is difficult to establish a single digital twin for the entire network.

[0051] In some embodiments, the network twin environment is generated by multiple sub-network digital twin models, and the actual sample data belongs to the behavioral observation data of multiple sub-network digital twin models on their respective network tasks.

[0052] In practical applications, the overall network twin environment is generated from multiple sub-network digital twin models. The current network digital twin application is digitally twinned for each component of the network, such as the core network, network topology, RAN, and user behavior.

[0053] In practical applications, the network twin environment is generated from multiple sub-network digital twin models. The selection and determination of the network twin environment to be implemented, i.e., the overall network digital twin tasks and paradigms, includes end-to-end quality assurance, network optimization, algorithm simulation, and network autonomy. Simultaneously, behavioral observation data of the overall network digital twin on each network task is collected. This behavioral observation data includes network traffic information, network traffic patterns, abnormal traffic information, communication data, communication rules, resources (e.g., spectrum resources), and calculated resource efficiency, data rate, and reliability. Based on this, the network twin state NetStatus at time j in the i-th network twin task paradigm is denoted as (NetStatus, xij), where i and j are positive integers.

[0054] In some embodiments, step 202 inputs the actual sample data and random noise into the diffusion model to complete Gaussian diffusion, obtaining the noisy data, including:

[0055] The actual sample data and random noise are input into the diffusion model. The actual sample data is randomly noise-added through a preset Markov chain until the actual sample data tends to a Gaussian distribution, at which point the diffusion ends and the noisy data is obtained.

[0056] In practical applications, a Markov chain C(T1,T1.1,T1.n,…T1.z) is defined as a fixed process for the diffusion steps between twins composed of multiple networks. The forward diffusion process in this application is: random noise is added through the defined Markov chain.

[0057] In practical applications, based on the Markov property, given the present and past states of a stochastic process, the data distribution of its future states depends only on the current state. By adding a bias component to the initial twin environment, the twin environment is transformed into a pure noise result TZ(1,…,N) as the bias component is continuously added. Since the parameters of the twin environment are known, the process of adding the bias component in the forward propagation of the diffusion model is controllable. This application inputs actual sample data and random noise into the diffusion model, and randomly adds noise to the actual sample data through a pre-defined Markov chain until the actual sample data tends to a Gaussian distribution, ending the diffusion process and obtaining the noisy data. This achieves controllable addition of a bias component to the data during the forward propagation process.

[0058] In practical applications, data is recovered by learning the reversed debiasing component process. Unlike the forward process where bias components are added to the data, the reverse process is a process of debiasing components from a pure noise twin environment. This requires the diffusion model to learn from the diffusion-based network to generate the twin environment.

[0059] In some embodiments, the data distribution obtained by the target sub-network digital twin model at each time step is related to the result distribution of the previous sub-network digital twin model in the network twin environment, as well as the diffusion kernel of the target sub-network digital twin model.

[0060] In practical applications, the (NetStatus, xij) sample is used as input, and the rules of the twinning process of the sub-digital twin network are used as the diffusion kernel. Therefore, the distribution at each time step is only related to the distribution of the previous sub-digital twin network result and the diffusion kernel of the sub-digital twin network, forming the final joint distribution result and completing T-step Gaussian diffusion.

[0061] In some embodiments, the above method further includes:

[0062] The perceived data is obtained from the target subnetwork digital twin model based on the result distribution and diffusion kernel;

[0063] The error is calculated based on the noisy data and the perceived data until the target sub-network digital twin model converges.

[0064] In practical applications, the noise output by the sub-digital twin network in the T-step and the noise prediction based on the distribution of the previous sub-digital twin network result are used to calculate the error for training until the network converges, thereby realizing data perception and dynamic correction.

[0065] In some embodiments, the above method further includes:

[0066] Preset parameters are introduced into the location space of different sub-network digital twin models in the network twin environment. These preset parameters are used to control the amount of noise injected into the location space of different sub-network digital twin models.

[0067] By continuously adding noise to disrupt the training data, randomly sampled noise is fed into different sub-network digital twin models. The deviation between the output of each sub-network digital twin model and the actual collected results is modeled to obtain a diffusion model.

[0068] In some embodiments, the bias includes a combination of deterministic and random components.

[0069] In practical applications, training the diffusion model involves: given a real network digital twin environment sample T, a parameter λ is introduced into the spatial updates of different network components to control the amount of noise nz injected in the location space of different network components. The training data is corrupted by continuously adding noise. Randomly sampled noise is passed into each part of the network digital twin model. The deviation between the result of each part of the digital twin and the actual result obtained is modeled. The diffusion model defines the deviation as a deterministic component F(T,t) and a random component R(T,t). The combination of the two components is as follows: The generation process is based on the latent variable model to obtain the parameterized distribution of the training data sample, and then the distribution of the generated data is made similar to the distribution of the training data to generate data similar to the training data.

[0070] In practical applications, updated deviation data is generated based on the combination of extrapolated deviation data and data obtained from the twin. The updated deviation data is then used to update the model. Real-time diffusion data and twin perception data need to be received for real-time dynamic correction, thereby improving the digital twin simulation model.

[0071] As described above, the digital twin network generation method provided in this application, and the basic principle of the digital twin architecture for the communication network, consists of a core diffusion model control element and various sub-network digital twin models. The core diffusion model control element is used to determine which sub-model should handle the input sample based on the diffusion Markov chain, and to set the diffusion noise parameter λ for each sub-network. The various sub-models are a set of relatively independent sub-twin models, each responsible for processing a specific twin subspace within the network's sub-domain. See also... Figure 3 As shown, generating a digital twin network can be achieved through the following steps:

[0072] Step 301: Select and determine the tasks and paradigms of the overall network digital twin to be implemented, while collecting behavioral observation data of the overall network digital twin on each network task.

[0073] The tasks and paradigms include end-to-end quality assurance, network optimization, algorithm simulation, and network autonomy. The collected behavioral observation data includes network traffic information, network traffic patterns, abnormal traffic information, communication data, communication rules, resources (e.g., spectrum resources), and calculated resource efficiency, data rate, and reliability. The network twin state NetStatus at time j in the i-th network twin task paradigm is denoted as (NetStatus, xij).

[0074] Step 302: The core diffusion model continuously extracts (NetStatus, xij) samples and randomly generates noise λi for each sub-digital twin network.

[0075] Step 303: Using the extracted (NetStatus, xij) samples as input, the rules of the twinning process of this sub-digital twin network are used as the diffusion kernel. The distribution at each time step is only related to the distribution of the previous sub-digital twin network result and the diffusion kernel of this sub-digital twin network, forming the final joint distribution result and completing T-step Gaussian diffusion.

[0076] Step 304: Start back-diffusion from the sub-digital twin network of step T, determine whether it has converged, and gradually reconstruct the behavioral observation data of the original data network digital twin on each network task.

[0077] Step 305: Use the noise output by the sub-digital twin network in step T and the noise prediction error based on the result distribution of the previous sub-digital twin network to train the network until it converges, thereby realizing data perception and dynamic correction.

[0078] Step 306: Sample potential noise variables from a Gaussian distribution. Since the amount of noise added in each step is different, the mean and variance of the added noise are controlled by multiple diffusion accuracy thresholds ai. The mean and variance of the trained sub-digital twin network are sampled to achieve dynamic correction of the digital twin network.

[0079] Repeat steps 304 and 305, using the updated deviation data for further updates. This requires receiving real-time diffusion data and twin-based sensing data for real-time dynamic correction. The final digital twin simulation model is then optimized and improved. This application controls diffusion accuracy by flexibly setting multiple diffusion accuracy thresholds, reducing the amount of twin-based data sensing data collected, avoiding the collection of excessive redundant data, and achieving dynamic correction of the network digital twin.

[0080] This application also provides an apparatus for generating a digital twin network, such as... Figure 4 As shown, the device includes:

[0081] The acquisition unit 401 is used to acquire actual sample data and random noise extracted from the digital twin model of the target sub-network in the network twin environment;

[0082] Processing unit 402 is used to input actual sample data and random noise into the diffusion model, complete Gaussian diffusion, and obtain data with added noise.

[0083] The processing unit 402 is used to randomly sample a noise from the noisy data, and to recover the data distribution from the noise by performing denoising and sampling operations on the noise, until the recovered data distribution meets the preset distribution conditions and the denoising ends; wherein the preset distribution conditions include Gaussian distribution;

[0084] The processing unit 402 is used to dynamically correct the target sub-network digital twin model based on the data after adding noise and the perceived data at the end of denoising, so as to update the network twin environment.

[0085] In some embodiments, the network twin environment is generated by multiple sub-network digital twin models, and the actual sample data belongs to the behavioral observation data of multiple sub-network digital twin models on their respective network tasks.

[0086] In some embodiments, the processing unit 402 is used to input actual sample data and random noise into a diffusion model, randomly add noise to the actual sample data through a preset Markov chain, and end the diffusion when the actual sample data tends to a Gaussian distribution, thereby obtaining the noisy data.

[0087] In some embodiments, the data distribution obtained by the target sub-network digital twin model at each time step is related to the result distribution of the previous sub-network digital twin model in the network twin environment, as well as the diffusion kernel of the target sub-network digital twin model.

[0088] In some embodiments, the acquisition unit 401 is used to acquire the sensing data obtained by the target sub-network digital twin model based on the result distribution and diffusion kernel; calculate the error based on the noisy data and the sensing data until the target sub-network digital twin model converges.

[0089] In some embodiments, the processing unit 402 is used to introduce preset parameters into the location space of different sub-network digital twin models in the network twin environment. The preset parameters are used to control the amount of noise injected into the location space of different sub-network digital twin models.

[0090] By continuously adding noise to disrupt the training data, randomly sampled noise is fed into different sub-network digital twin models. The deviation between the output of each sub-network digital twin model and the actual collected results is modeled to obtain a diffusion model.

[0091] In some embodiments, the bias includes a combination of deterministic and random components.

[0092] This application also provides a device for generating a digital twin network, such as... Figure 5 As shown, the digital twin network generation device 500 includes: a communication interface 501 and a processor 502; wherein,

[0093] The communication interface 501 enables information exchange with network devices;

[0094] The processor 502, connected to the communication interface 501, is used to execute the methods provided by one or more of the above-mentioned terminal-side technical solutions when running a computer program;

[0095] Memory 503 stores computer programs that can run on processor 502.

[0096] The processor 502 is used to acquire actual sample data and random noise extracted from the digital twin model of the target sub-network in the network twin environment;

[0097] The actual sample data and random noise are input into the diffusion model to complete Gaussian diffusion and obtain the data with added noise.

[0098] A noise sample is randomly selected from the noise-added data. The noise is then denoised and sampled to recover the data distribution from the noise. The denoising process ends when the recovered data distribution meets the preset distribution conditions. The preset distribution conditions include a Gaussian distribution.

[0099] In some embodiments, the network twin environment is generated by multiple sub-network digital twin models, and the actual sample data belongs to the behavioral observation data of multiple sub-network digital twin models on their respective network tasks.

[0100] In some embodiments, the processor 502 is used to input actual sample data and random noise into a diffusion model, randomly add noise to the actual sample data through a preset Markov chain, and end the diffusion when the actual sample data tends to a Gaussian distribution, thereby obtaining the noisy data.

[0101] In some embodiments, the data distribution obtained by the target sub-network digital twin model at each time step is related to the result distribution of the previous sub-network digital twin model in the network twin environment, as well as the diffusion kernel of the target sub-network digital twin model.

[0102] In some embodiments, the processor 502 is used to acquire the sensing data obtained by the target subnetwork digital twin model based on the result distribution and diffusion kernel; calculate the error based on the noisy data and the sensing data until the target subnetwork digital twin model converges.

[0103] In some embodiments, the processor 502 is used to introduce preset parameters into the location space of different sub-network digital twin models in the network twin environment. The preset parameters are used to control the amount of noise injected into the location space of different sub-network digital twin models. By continuously adding noise to destroy the training data, randomly sampled noise is introduced into different sub-network digital twin models. The deviation between the output of each sub-network digital twin model and the actual results obtained is modeled to obtain a diffusion model.

[0104] In some embodiments, the bias includes a combination of deterministic and random components.

[0105] It should be noted that the specific processing procedures of communication interface 501 and processor 502 can be understood by referring to the above method, and will not be repeated here.

[0106] Of course, in practical applications, the various components in the digital twin network generation device 500 are coupled together via a bus system 504. It can be understood that the bus system 504 is used to implement communication between these components. In addition to a data bus, the bus system 504 also includes a power bus, a control bus, and a status signal bus. However, for clarity, in... Figure 5 The general designated all buses as Bus System 504.

[0107] The memory 503 in this embodiment is used to store various types of data to support the operation of the digital twin network generation device 500. Examples of such data include any computer program used to operate on the digital twin network generation device 500.

[0108] The methods disclosed in the embodiments of this application can be applied to or implemented by processor 502. Processor 502 may be an integrated circuit chip with signal processing capabilities. During implementation, each step of the above method can be completed by integrated logic circuits in the hardware of processor 502 or by instructions in software form. Processor 502 may be a general-purpose processor, a digital signal processor (DSP), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Processor 502 can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of this application. A general-purpose processor may be a microprocessor or any conventional processor. The steps of the methods disclosed in the embodiments of this application can be directly manifested as execution by a hardware decoding processor, or execution by a combination of hardware and software modules in the decoding processor. The software modules may be located in a storage medium, specifically memory 503. Processor 502 reads information from memory 503 and, in conjunction with its hardware, completes the steps of the aforementioned method.

[0109] In an exemplary embodiment, the digital twin network generation device 500 may be implemented by one or more application-specific integrated circuits (ASICs), DSPs, programmable logic devices (PLDs), complex programmable logic devices (CPLDs), field-programmable gate arrays (FPGAs), general-purpose processors, controllers, microcontrollers (MCUs), microprocessors, or other electronic components to perform the aforementioned method.

[0110] In an exemplary embodiment, the network device may be implemented by one or more application-specific integrated circuits (ASICs), DSPs, programmable logic devices (PLDs), complex programmable logic devices (CPLDs), field-programmable gate arrays (FPGAs), general-purpose processors, controllers, microcontrollers (MCUs), microprocessors, or other electronic components to perform the aforementioned method.

[0111] In an exemplary embodiment, the user terminal may be implemented by one or more ASICs, DSPs, PLDs, CPLDs, FPGAs, general-purpose processors, controllers, MCUs, microprocessors, or other electronic components to perform the aforementioned method.

[0112] It is understood that the memory in the embodiments of this application can be volatile memory or non-volatile memory, or both. Specifically, non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), ferromagnetic random access memory (FRAM), flash memory, magnetic surface memory, optical disc, or compact disc read-only memory (CD-ROM); magnetic surface memory can be disk storage or magnetic tape storage. Volatile memory can be random access memory (RAM), which is used as an external cache. By way of example, but not limitation, many forms of RAM are available, such as Static Random Access Memory (SRAM), Synchronous Static Random Access Memory (SSRAM), Dynamic Random Access Memory (DRAM), Synchronous Dynamic Random Access Memory (SDRAM), Double Data Rate Synchronous Dynamic Random Access Memory (DDRSDRAM), Enhanced Synchronous Dynamic Random Access Memory (ESDRAM), Sync Link Dynamic Random Access Memory (SLDRAM), and Direct Rambus Random Access Memory (DRRAM).The memories described in the embodiments of this application are intended to include, but are not limited to, these and any other suitable types of memories.

[0113] In an exemplary embodiment, this application also provides a storage medium, namely a computer storage medium, specifically a computer-readable storage medium, such as a memory 503 storing a computer program. This computer program can be executed by the processor 502 of the digital twin network generation device 500 to complete the steps of the aforementioned method on the terminal side. The computer-readable storage medium can be a memory such as FRAM, ROM, PROM, EPROM, EEPROM, Flash Memory, magnetic surface memory, optical disc, or CD-ROM.

[0114] In an exemplary embodiment, this application also provides a computer program product. When executed by one or more computing devices, the computer program product enables the one or more computing devices to perform any of the methods described in the aforementioned methods for generating a digital twin network. The computer program product can be a software installation package. When any of the aforementioned methods for generating a digital twin network is required, the computer program product can be downloaded and executed on a computer.

[0115] It should be noted that terms such as "first" and "second" are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence.

[0116] Furthermore, the technical solutions described in the embodiments of this application can be combined arbitrarily without conflict.

[0117] The above description is merely a preferred embodiment of this application and is not intended to limit the scope of protection of this application.

Claims

1. A method for generating a digital twin network, characterized in that, include: Obtain actual sample data and random noise extracted from the digital twin model of the target subnetwork in the network twin environment; The actual sample data and the random noise are input into the diffusion model to complete Gaussian diffusion and obtain the data with added noise. A noise sample is randomly selected from the noisy data. The noise is then denoised and sampled to recover the data distribution from the noise. The denoising process ends when the recovered data distribution meets a preset distribution condition. The preset distribution condition includes a Gaussian distribution. Based on the data after noise addition and the perceived data at the end of denoising, the target sub-network digital twin model is dynamically corrected to update the network twin environment.

2. The method according to claim 1, characterized in that, The network twin environment is generated by multiple sub-network digital twin models, and the actual sample data belongs to the behavioral observation data of the multiple sub-network digital twin models on their respective network tasks.

3. The method according to claim 1, characterized in that, The step of inputting the actual sample data and the random noise into the diffusion model to complete Gaussian diffusion and obtain the noisy data includes: The actual sample data and the random noise are input into the diffusion model. The actual sample data is randomly noise-added through a preset Markov chain until the actual sample data tends to a Gaussian distribution, at which point the diffusion ends, and the noisy data is obtained.

4. The method according to claim 3, characterized in that, The data distribution obtained by the target sub-network digital twin model at each time step is related to the result distribution of the previous sub-network digital twin model in the network twin environment, as well as the diffusion kernel of the target sub-network digital twin model.

5. The method according to claim 4, characterized in that, The method further includes: The sensing data is obtained from the target sub-network digital twin model based on the result distribution and the diffusion kernel; The error is calculated based on the noisy data and the perceived data until the target sub-network digital twin model converges.

6. The method according to claim 1, characterized in that, The method further includes: Preset parameters are introduced into the location space of different sub-network digital twin models in the network twin environment. The preset parameters are used to control the amount of noise injected into the location space of the different sub-network digital twin models. By continuously adding noise to disrupt the training data, randomly sampled noise is fed into the different sub-network digital twin models. The deviation between the output of each sub-network digital twin model and the actual collected results is modeled to obtain the diffusion model.

7. The method according to claim 6, characterized in that, The deviation includes a combination of deterministic and random components.

8. A device for generating a digital twin network, characterized in that, include: Communication interface and processor; among which, The processor is configured to: acquire actual sample data and random noise extracted from the digital twin model of the target sub-network in the network twin environment; The actual sample data and the random noise are input into the diffusion model to complete Gaussian diffusion and obtain the data with added noise. A noise sample is randomly selected from the noise-added data. The noise is then denoised and sampled to recover the data distribution from the noise. The denoising process ends when the recovered data distribution meets a preset distribution condition. The preset distribution condition includes a Gaussian distribution.

9. A storage medium having a computer program stored thereon, characterized in that, When a computer program is executed by a processor, it implements the steps of any one of claims 1 to 7.

10. A computer product comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method of any one of claims 1 to 7.

Images