Construction method, device and equipment of digital twin model based on industrial internet

Abstract

The application provides a construction method, device and equipment of a digital twin model based on an industrial internet, and relates to the technical field of the industrial internet. The method comprises the following steps: distributing global digital twin model parameters to multiple edge devices, obtaining sample data quantity and data mapping deviation of a local digital twin model obtained after each edge device constructs the local digital twin model based on the global digital twin model parameters and completes a round of training; based on the obtained data, determining an edge device selection strategy, a resource allocation strategy and a transmission power allocation strategy for the current round of training from the multiple edge devices, with the optimization target being to maximize the quality of the global digital twin model; and based on the strategy determined at this time, obtaining the local digital twin model parameters of the edge device corresponding to the edge device selection strategy, so as to update the global digital twin model parameters. The method realizes dynamic strategy selection in the model construction process, reduces the negative impact caused by data deviation, and improves the construction quality of the model.

Description

Technical Field

[0001] This application relates to the field of digital twin model technology, and in particular to a method, apparatus and equipment for constructing a digital twin model based on the Industrial Internet. Background Technology

[0002] In the context of the Industrial Internet, digital twin technology integrates real-time industrial data with artificial intelligence algorithms to construct a dynamic digital mapping of the physical production system, playing a key role in areas such as predictive maintenance of equipment, optimization of production processes, and quality monitoring.

[0003] Currently, traditional centralized digital twin model construction methods rely on industrial field equipment to directly transmit sensor data to a central server. However, due to the dynamic complexity of the industrial environment and the limitations of the sensing capabilities of field equipment, the collected data often contains noise and errors, leading to physical-virtual mapping deviation (referring to the phenomenon that the virtual representation and the behavior of the physical entity are inconsistent due to model simplification, data errors, or dynamic mismatch during the interaction between the physical and virtual systems), which negatively impacts the construction quality of the digital twin model. Summary of the Invention

[0004] This application provides a method, apparatus, and device for constructing a digital twin model based on the Industrial Internet, which addresses the technical problem that the physical-virtual mapping deviation in the data obtained in traditional digital twin model construction methods negatively impacts the construction quality of the digital twin model.

[0005] In a first aspect, embodiments of this application provide a method for constructing a digital twin model based on the Industrial Internet, comprising:

[0006] Distribute global digital twin model parameters to multiple edge devices;

[0007] The sample data volume and data mapping deviation of the local digital twin model obtained by each edge device after constructing and completing one round of training based on the parameters of the global digital twin model are obtained;

[0008] Based on the sample data volume and data mapping deviation of the local digital twin model, with the optimization objective of maximizing the quality of the global digital twin model, the edge device selection strategy, resource allocation strategy, and transmit power allocation strategy for this round of training are determined from the multiple edge devices.

[0009] According to the resource allocation strategy and the transmit power allocation strategy, the local digital twin model parameters of the edge device corresponding to the edge device selection strategy are obtained, and the global digital twin model parameters are updated based on the local digital twin model parameters.

[0010] In one possible implementation, the step of determining the edge device selection strategy, resource allocation strategy, and transmit power allocation strategy for this round of training from the plurality of edge devices, with the optimization objective of maximizing the quality of the global digital twin model based on the sample data volume and data mapping deviation of the local digital twin model, includes:

[0011] Based on the sample data volume, data mapping deviation, edge device selection strategy, resource allocation strategy, and transmit power allocation strategy of the local digital twin model, the error function of the global digital twin model is constructed. Minimizing the error function is the optimization objective. From the multiple edge devices, the edge device selection strategy, resource allocation strategy, and transmit power allocation strategy for this round of training are determined. The quality of the global digital twin model is negatively correlated with the error of the global digital twin model.

[0012] In one possible implementation, the error function for constructing the global digital twin model based on the sample data volume, data mapping deviation, edge device selection strategy, resource allocation strategy, and transmit power allocation strategy of the local digital twin model includes:

[0013] Based on the resource allocation strategy and the transmit power allocation strategy, the error rate of the edge device is determined;

[0014] Based on the sample data volume, data mapping deviation, edge device selection strategy, and error rate of the edge device in the local digital twin model, an error function for the global digital twin model is constructed.

[0015] In one possible implementation, the method for constructing a digital twin model based on the Industrial Internet further includes: defining the following as constraints for optimization objectives:

[0016] The edge device selection strategy corresponds to the resource allocation for each edge device in the resource allocation strategy;

[0017] The edge device selection strategy corresponds to a total latency of each edge device that is less than or equal to a preset latency threshold.

[0018] The edge device selection strategy corresponds to the total energy consumption of each edge device being less than or equal to a preset energy consumption threshold.

[0019] The sample data volume of the edge device is greater than or equal to the preset minimum data volume;

[0020] The transmit power of the edge device is less than or equal to the maximum transmit power.

[0021] In one possible implementation, determining the error rate of the edge device based on the resource allocation strategy and the transmit power allocation strategy includes:

[0022] The error rate of the edge device is determined based on the resource allocation strategy, the transmit power allocation strategy, the total number of resource blocks, the waterfall threshold for the impact of received signal quality on the transmission error rate, the bandwidth of the resource block, the channel gain expectation, the distance of the edge device, the Rayleigh fading parameter, the noise power spectral density, and the interference generated by other edge devices using the resource block.

[0023] In one possible implementation, the step of determining the edge device selection strategy, resource allocation strategy, and transmit power allocation strategy for this round of training from the plurality of edge devices, with minimizing the error function as the optimization objective, includes:

[0024] The reward function is determined based on the error function;

[0025] Based on the reward function, a reinforcement learning algorithm is used to determine the edge device selection strategy, resource allocation strategy, and transmit power allocation strategy for this round of training.

[0026] In one possible implementation, the state space in the reinforcement learning algorithm includes the edge device selection strategy, resource allocation strategy, and transmit power allocation strategy of the previous training round, as well as the data mapping bias and the cumulative error of the global digital twin model;

[0027] The action space in the reinforcement learning algorithm includes the edge device selection strategy, resource allocation strategy, and transmit power allocation strategy for this round of training.

[0028] Secondly, embodiments of this application provide an apparatus for constructing a digital twin model based on the Industrial Internet, comprising:

[0029] The processing module is used to distribute global digital twin model parameters to multiple edge devices.

[0030] The acquisition module is used to acquire the sample data volume and data mapping deviation of the local digital twin model obtained by each edge device after constructing and completing one round of training based on the parameters of the global digital twin model.

[0031] The processing module is further configured to determine the edge device selection strategy, resource allocation strategy, and transmit power allocation strategy for this round of training from the multiple edge devices, based on the sample data volume and data mapping deviation of the local digital twin model and with the optimization objective of maximizing the quality of the global digital twin model.

[0032] The acquisition module is further configured to acquire local digital twin model parameters of the edge device corresponding to the edge device selection strategy, according to the resource allocation strategy and the transmit power allocation strategy.

[0033] The processing module is also used to update the global digital twin model parameters based on the local digital twin model parameters.

[0034] Optionally, the processing module is further configured to construct an error function of the global digital twin model based on the sample data volume, data mapping deviation, edge device selection strategy, resource allocation strategy, and transmit power allocation strategy of the local digital twin model, with minimizing the error function as the optimization objective, and determine the edge device selection strategy, resource allocation strategy, and transmit power allocation strategy for this round of training from the multiple edge devices. The quality of the global digital twin model is negatively correlated with the error of the global digital twin model.

[0035] Optionally, the processing module is further configured to determine the error rate of the edge device based on the resource allocation strategy and the transmit power allocation strategy.

[0036] The processing module is further configured to construct an error function for the global digital twin model based on the sample data volume of the local digital twin model, the data mapping deviation, the edge device selection strategy, and the error rate of the edge device.

[0037] Optionally, the processing module is further configured to determine the error rate of the edge device based on the resource allocation strategy, the transmit power allocation strategy, the total number of resource blocks, the waterfall threshold for the impact of received signal quality on the transmission error rate, the bandwidth of the resource block, the channel gain expectation, the distance of the edge device, the Rayleigh fading parameter, the noise power spectral density, and the interference generated by other edge devices using the resource block.

[0038] Optionally, the processing module is further configured to determine a reward function based on the error function.

[0039] The processing module is further configured to determine the edge device selection strategy, resource allocation strategy, and transmit power allocation strategy for this round of training based on the reward function and using a reinforcement learning algorithm.

[0040] Thirdly, embodiments of this application provide an electronic device, including: a memory and a processor;

[0041] The memory stores computer-executed instructions;

[0042] The processor executes computer execution instructions stored in the memory, causing the processor to perform the first aspect and / or various possible implementations of the first aspect as described above.

[0043] Fourthly, embodiments of this application provide a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, are used to implement the first aspect and / or various possible implementations of the first aspect.

[0044] Fifthly, embodiments of this application provide a computer program product, including a computer program that, when executed by a processor, implements the first aspect and / or various possible implementations of the first aspect.

[0045] The method, apparatus, and device for constructing a digital twin model based on the Industrial Internet provided in this application embodiment distribute global digital twin model parameters to multiple edge devices. Then, it acquires the sample data volume and data mapping deviation of the local digital twin models obtained by each edge device after completing one round of training based on the global digital twin model parameters. Based on the sample data volume and data mapping deviation of the local digital twin models, and with the optimization objective of maximizing the quality of the global digital twin model, it determines the edge device selection strategy, resource allocation strategy, and transmission power allocation strategy for this round of training from multiple edge devices. Then, according to the determined resource allocation strategy and transmission power allocation strategy, it acquires the local digital twin model parameters of the edge devices corresponding to the edge device selection strategy, and updates the global digital twin model parameters based on the local digital twin model parameters. This method, applied in the Industrial Internet, reduces the impact of abnormal industrial data through distributed training, reduces data collection and transmission costs, and simultaneously realizes dynamic device selection and corresponding resource allocation during the digital twin model construction process. This reduces the negative impact caused by physical-virtual mapping deviations and improves the performance of digital twin models in the Industrial Internet. Attached Figure Description

[0046] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0047] Figure 1 This is a schematic diagram illustrating a scenario of the method for constructing a digital twin model based on the Industrial Internet provided in this application;

[0048] Figure 2 This is a flowchart illustrating the method for constructing a digital twin model based on the Industrial Internet provided in this application. Figure 1 ;

[0049] Figure 3 This is a flowchart illustrating the method for constructing a digital twin model based on the Industrial Internet provided in this application. Figure 2 ;

[0050] Figure 4This is a schematic diagram of the structure of the device for constructing a digital twin model based on the Industrial Internet provided in this application;

[0051] Figure 5 This is a schematic diagram of the structure of the electronic device provided in this application;

[0052] Figure label:

[0053] 101 - Central Server;

[0054] 102 - Intelligent processing equipment;

[0055] 103 - Intelligent Mobile Terminal;

[0056] 104 - Intelligent camera equipment.

[0057] The accompanying drawings illustrate specific embodiments of this application, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the concept in any way, but rather to illustrate the concept of this application to those skilled in the art through reference to particular embodiments. Detailed Implementation

[0058] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0059] The terms "first," "second," "third," "fourth," etc. (if present) in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that embodiments of the invention described herein can be implemented, for example, in orders other than those illustrated or described herein.

[0060] In this application, the terms "exemplary" or "for example" are used to indicate examples, illustrations, or descriptions. Any embodiment or design described as "exemplary" or "for example" in this application should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of terms such as "exemplary" or "for example" is intended to present the relevant concepts in a specific manner.

[0061] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0062] First, the terms used in this application will be explained:

[0063] Digital twin model: a technology that creates a virtual copy of a physical entity in a digital way, which can reflect the state of the physical entity in real time and assist in analysis, prediction and optimization decision-making.

[0064] Cyclic Redundancy Check (CRC) is an efficient algorithm for detecting errors during data transmission or storage. Based on polynomial division, it adds redundant check bits (CRC checksums) to the end of the data, ensuring that the entire data (including the original data and the checksum) is divisible by a predetermined generator polynomial. At the receiving end, the same generator polynomial is used to divide the received data. If the remainder is zero, the data transmission is considered error-free; otherwise, an error has occurred during transmission.

[0065] Reinforcement learning is an important branch of machine learning that studies how an agent takes a series of actions in an environment to maximize cumulative reward. Unlike supervised and unsupervised learning, reinforcement learning emphasizes that the agent learns the optimal policy through interaction with the environment. Reinforcement learning involves three basic elements: agent, environment, and reward. The agent takes actions in the environment, and the environment returns a new state and reward signal based on the agent's actions. The agent's goal is to learn a policy that maximizes the cumulative reward obtained in the long run.

[0066] Reward function: This is a core concept in reinforcement learning. In the reinforcement learning framework, the agent learns the optimal behavioral strategy by interacting with the environment. The reward function is used to evaluate the quality of the agent's actions in a specific state, providing feedback signals to guide the agent to learn toward the desired goal.

[0067] Proximal Policy Optimization (PPO) is a reinforcement learning algorithm based on policy gradients. It aims to solve problems such as excessive policy updates, unstable training, and low sample efficiency that occur in traditional policy gradient algorithms during training.

[0068] The actor-critic approach is a reinforcement learning framework that combines policy gradients and value functions, aiming to address the problems of slow convergence, high variance, or unstable learning in traditional methods.

[0069] Markov Decision Processes (MDPs) are a mathematical framework for modeling dynamic systems with stochasticity and decision-making characteristics. They consist of five core elements: state, action, transition probability, reward function, and discount factor. MDPs aim to help agents make optimal decisions in uncertain environments to maximize long-term cumulative rewards.

[0070] In the context of the Industrial Internet, digital twin technology integrates real-time industrial data with artificial intelligence algorithms to construct a dynamic digital mapping of the physical production system, playing a key role in areas such as predictive maintenance of equipment, optimization of production processes, and quality monitoring.

[0071] Currently, traditional centralized digital twin model building methods rely on industrial field equipment directly transmitting sensor data to a central server. The central server, as the "brain" of data processing and model building, receives data from various industrial field equipment and carries out subsequent digital twin model building work on this basis to achieve accurate mapping and dynamic simulation of physical entities in virtual space.

[0072] However, due to the dynamic complexity of the industrial environment and the limitations of the sensing capabilities of on-site equipment, the collected data often contains noise and errors, leading to physical-virtual mapping deviations. The dynamic unpredictability of physical-virtual mapping deviations in industrial scenarios further complicates equipment selection, thus negatively impacting the quality of digital twin model construction. In addition, communication bottlenecks, edge device energy limitations, and dynamic resource fluctuations (such as channel status and computing resource availability) mean that only a portion of industrial equipment can be selected for construction, and static resource allocation strategies exacerbate model performance degradation and resource waste.

[0073] To address the aforementioned issues, this application provides a method for constructing a digital twin model based on the Industrial Internet.

[0074] Figure 1 This is a schematic diagram illustrating a scenario of the method for constructing a digital twin model based on the Industrial Internet provided in this application. For example... Figure 1As shown, the central server 101 is communicatively connected to the intelligent processing equipment 102, the intelligent mobile terminal 103, and the intelligent camera device 104, respectively. The central server 101, intelligent processing equipment 102, intelligent mobile terminal 103, and intelligent camera device 104 are all located within the same industrial internet, and are considered edge devices within this industrial internet. During the construction of the global digital twin model, the intelligent processing equipment 102, intelligent mobile terminal 103, and intelligent camera device 104 acquire relevant edge data, process the data, and send the processed results to the central server 101. The central server 101 determines which device is participating in the construction of the global digital twin model and controls the corresponding device to transmit data with the central server 101 again, thereby realizing the construction of the global digital twin model. This application does not impose any special restrictions on the location of the central server 101.

[0075] Edge devices in this industrial internet include, for example, machine tool A, machine tool B, mobile intelligent device A, mobile intelligent device B, intelligent camera device A, and intelligent camera device B.

[0076] For example, the intelligent processing equipment 102 can be a machine tool in a processing workshop, the intelligent mobile terminal 103 can be a mobile intelligent device, and the intelligent camera device 104 can be a camera. During the construction of the global digital twin model, the central server 101 set in the base station conducts multiple data exchanges with the machine tool in the processing workshop, the mobile intelligent device, and the camera. After the first data transmission, the central server determines the edge device currently participating in the model construction, and conducts data transmission again according to the edge device determined at this time. The currently determined edge devices (e.g., the machine tool in the workshop or the mobile intelligent device) exchange relevant model data, and the central server 101 constructs the global digital twin model. This application does not impose any special restrictions on the type of edge device.

[0077] This application provides a method for constructing a digital twin model based on the Industrial Internet. Based on a reinforcement learning algorithm, it jointly optimizes the edge device selection strategy and corresponding bandwidth and transmit power allocation strategies involved in the construction of the federated digital twin model. This maximizes the quality of the global industrial digital twin model under strict latency and energy constraints. The method reduces the impact of abnormal industrial data through distributed training, lowers data collection and transmission costs, and enables dynamic device selection and corresponding resource allocation during the digital twin model construction process. This reduces the negative impact of physical-virtual mapping deviations and improves the performance of digital twin models in the Industrial Internet.

[0078] The technical solution of this application and how the technical solution of this application solves the above-mentioned technical problems are described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments. The embodiments of this application will now be described with reference to the accompanying drawings.

[0079] Figure 2 A flowchart illustrating the method for constructing a digital twin model based on the Industrial Internet provided in this application. Figure 1 The executing entity in this embodiment can be, for example, a central server located in an industrial internet. Figure 2 As shown in this embodiment, the method for constructing a digital twin model based on the Industrial Internet includes:

[0080] S201: Distribute global digital twin model parameters to multiple edge devices.

[0081] Among them, the global digital twin model parameters are used to accurately map and describe entities (such as devices, systems, processes, etc.) and their operating environments in the physical world in the digital space, while edge devices are used to indicate devices located at the edge of the industrial internet or on edge computing nodes.

[0082] In this embodiment, the Industrial Internet combines Internet technology with industrial production. By connecting various devices, machines, personnel, and systems, it achieves comprehensive interconnection, allowing industrial data to flow freely, thereby improving production efficiency, reducing costs, and enhancing product quality. The Industrial Internet includes multiple edge devices and a central server. The edge devices can collect their own operating status, production data, environmental data, etc. in real time, and perform preprocessing such as cleaning, filtering, and compression to provide corresponding preprocessed data to the central server.

[0083] In the Industrial Internet, the establishment of digital twin models aims to achieve accurate simulation, monitoring, analysis, and optimization of physical entities, helping enterprises or organizations better manage physical entities, improve production efficiency, reduce costs, and enhance product quality. The establishment of digital twin models relies on data collected from physical entities, including sources such as sensors and IoT devices. This data is the foundation for building and operating digital twin models, reflecting information such as the state and behavior of the physical entities. When the Industrial Internet is launched, the central server and edge devices need to work collaboratively. The same initial parameters ensure that the digital twin models are in the same initial state at startup. Therefore, multiple edge devices need to use model parameters uniformly provided by the central server to build their corresponding digital twin models.

[0084] During the operation of the Industrial Internet, the central server calls upon the global digital twin model parameters stored within it and distributes these global digital twin model parameters to multiple edge devices in the current Industrial Internet.

[0085] For example, the currently operating Industrial Internet is the Industrial Internet of any parts processing production line. This production line is equipped with multiple edge devices, such as sensors installed on various machine tools (for monitoring parameters such as rotation speed, temperature, and vibration), controllers for material handling robots, etc. The central server is responsible for coordinating and managing the operation of the entire production line and interacting with the edge devices through the Industrial Internet. The global digital twin model parameters include: the overall layout of the production line, the performance indicators of each device, the logical relationship of the production process, and the thresholds of key quality indicators. When the Industrial Internet is operating, the central server calls the model parameters "the overall layout of the production line, the performance indicators of each device, the logical relationship of the production process, and the thresholds of key quality indicators" and distributes the model parameters to all edge devices in the current Industrial Internet. The edge devices in the Industrial Internet include, for example, machine tool A, machine tool B, mobile intelligent device A, mobile intelligent device B, intelligent camera device A, and intelligent camera device B.

[0086] S202: Obtain the sample data volume and data mapping deviation of the local digital twin model obtained by each edge device after constructing and completing one round of training based on the global digital twin model parameters.

[0087] Among them, the local digital twin model is used to indicate the digital twin model built on the edge device, the sample data volume is used to indicate the actual amount of data acquired by the edge device during the construction of the digital twin model, and the data mapping deviation is used to describe the ratio between the accurate amount of data acquired by the edge device and the actual amount of data acquired.

[0088] It is understandable that during the data acquisition process of edge devices, due to the combined effects of various factors such as data source, acquisition environment, device performance, and data processing, there will be both usable and unusable data in the actual data acquired. Usable data may be accurate data acquired by the edge device, while unusable data may be invalid or erroneous data acquired by the edge device.

[0089] In this embodiment of the application, each round of training is based on the global digital twin model parameters sent from the central server to the edge device each time; the edge device constructs a digital twin model based on the global digital twin model parameters obtained in the current round and the data information obtained by the edge device itself, and the constructed digital twin model becomes the local digital twin model of the corresponding edge device.

[0090] In each training round, multiple edge devices in the industrial internet create local digital twin models for their respective edge devices based on the global digital twin model parameters acquired in that round. For each edge device, the data information acquired in that round is analyzed and processed to obtain the sample data volume and data mapping deviation corresponding to the local digital twin model.

[0091] For example, the training epochs of a local digital twin model are: The current industrial internet has a number of [number missing] [units missing]. Edge devices, local digital twin models are denoted as In other words, local digital twin model In a set of quantities Created on edge devices, and local digital twin models Each edge device is obtained by utilizing feature data collected from edge devices. The datasets that can be collected are ,in, It is an edge device The number of samples collected, and This represents a data sample; that is, each edge device. The actual amount of data acquired in the current training round is And the amount of accurate data obtained in this instance is less than Because the data collected by edge devices has mapping bias, only accurate data from all collected data is used to create the corresponding local digital twin model to prevent a significant reduction in model quality; all data collected by the edge devices is denoted as... ,and The actual data in the data can be represented as:

[0092]

[0093] in, Edge devices In the The exact amount of data collected in each training round can be written as: .

[0094] Therefore, the difference between the data in the local digital twin model and the actual data in the physical world is denoted as . That is, the data mapping deviation is , specifically That is, edge devices Data mapping deviation for computing edge devices In the The ratio of accurate data collected during training rounds to real data.

[0095] S203: Based on the sample data volume and data mapping deviation of the local digital twin model, with the optimization objective of maximizing the quality of the global digital twin model, determine the edge device selection strategy, resource allocation strategy, and transmit power allocation strategy for this round of training from multiple edge devices.

[0096] Among them, the edge device selection strategy is used to determine the edge devices participating in the construction of the global digital twin model in the current training round; the resource allocation strategy is used to determine the specific resource allocation required for different edge devices participating in the construction of the global digital twin model; the transmit power allocation strategy is used to determine the transmit power used by each edge device when multiple edge devices participating in the construction of the global digital twin model transmit data with the central server; and maximizing the quality representation of the global digital twin model minimizes the training error of the global digital twin model on the dataset.

[0097] Understandably, digital twin models built on different edge devices differ, while the digital twin model built on the central server is based on the digital twin models corresponding to multiple edge devices. In other words, the global digital twin model is composed of multiple local digital twin models. Therefore, the quality of multiple local digital twin models affects the quality of the global digital twin model. Considering the impact of resource constraints and mapping bias on the quality of local digital twin models, before building the global digital twin model corresponding to the current industrial internet, the central server needs to select the edge devices participating in the global digital twin model for that training round. The selected edge devices upload the local digital twin model values ​​obtained during training to the central server. The central server then globally aggregates the model parameters of the local digital twin models corresponding to the selected edge devices participating in the construction, thereby obtaining the global digital twin model corresponding to the current training round.

[0098] In this embodiment, the resource allocation strategy determines the computing and storage resources that the edge device can obtain. When computing resources are insufficient, the edge device may not be able to fully process and analyze the collected data, leading to errors in the data processing process. The transmission power allocation strategy directly affects the communication quality between the edge device and the central server or other devices. When the transmission power is too low and the signal strength is insufficient, data is easily lost or bit errors occur during transmission. Therefore, the resource allocation strategy and the transmission power allocation strategy will affect the construction quality of the global digital twin model.

[0099] Because of differences in data transmission capabilities and data acquisition quality among multiple edge devices, different edge device selection strategies, resource allocation strategies, and transmit power allocation strategies affect the quality of the global digital twin model constructed by the central server. In other words, when determining the edge device selection strategy, resource allocation strategy, and transmit power allocation strategy for this round of training, it is necessary to ensure that the quality of the global digital twin model in the current round is optimal. For example, the... The model quality of the global digital twin model in the wheel is , specifically It can be represented as:

[0100]

[0101] in, Indicates the first The model quality of the global digital twin model, This indicates the total amount of data retrieved by the central server. Indicates the first Wheel part digital twin model The quality of the model.

[0102] In some embodiments, to better determine the model quality of the digital twin model, ideally, each edge device... Both can collect complete and accurate datasets. ,set up Indicates the first Wheel part digital twin model The goal of each part of the digital twin model is to find the optimal model parameters. To maximize model quality, which is equivalent to minimizing the dataset. Error on; assume Representing data If the training error is taken into account, then the quality of a partial digital twin model can be expressed using the corresponding training error. Therefore, the quality of a partial digital twin model can be defined as:

[0103]

[0104] Similarly, using Indicates the first Given the parameters of the global digital twin model of the wheel, the quality function of the global digital twin model can be calculated as follows:

[0105]

[0106] For example, based on the currently acquired local digital twin models corresponding to multiple edge devices... and To maximize the quality of the global digital twin model To optimize the objective, the edge device selection strategy, resource allocation strategy, and transmit power allocation strategy are determined from multiple edge devices for this round of training; at this point, the quality of the global digital twin model is maximized. That is, to determine the training error of the minimized global digital twin model; and to calculate the results under different conditions for different edge device selection strategies, resource allocation strategies, and transmit power allocation strategies. And filter out those that meet the requirements The edge device selection strategy, resource allocation strategy, and transmit power allocation strategy corresponding to the minimum case, that is, satisfying... The largest edge device selection strategy Resource allocation strategy and transmit power allocation strategy .

[0107] S204: According to the resource allocation strategy and transmit power allocation strategy, obtain the local digital twin model parameters of the edge device corresponding to the edge device selection strategy, and update the global digital twin model parameters based on the local digital twin model parameters.

[0108] For example, according to the resource allocation strategy determined at that time. and transmit power allocation strategy Acquisition and edge device selection strategy Corresponding local digital twin model parameters of edge devices And based on multiple local digital twin model parameters Update global digital twin model parameters Specifically, integrating portions of the digital twin model from edge devices into a central server ultimately constructs a global digital twin model. This process can be defined as follows:

[0109]

[0110] in, This represents the set of all partial digital twin models within the global digital twin model. This represents a specific part of the digital twin model. This represents the aggregation of all parts of the digital twin model.

[0111] In some embodiments, to capture performance increments during the training of the digital twin model, gradient descent is used to update the model parameters of each part of the digital twin model. For example, the update of the model parameters of each part of the digital twin model can be expressed as:

[0112]

[0113] in, Represents the gradient operator, This represents the learning rate.

[0114] The update of the global digital twin model can be represented as:

[0115]

[0116] in, Indicates the first Should edge devices be selected during round training? Participate in the construction of the global digital twin model, based on the edge device selection strategy. Confirmed, satisfied .

[0117] In this embodiment, the global digital twin model parameters stored in the central server are updated in real time. When the global digital twin model is updated once, it indicates that the central server has completed the construction of a global digital twin model. Correspondingly, all edge devices in the industrial internet also construct a local digital twin model. During the construction of a new global digital twin model, the central server reissues the global digital twin model parameters as the basis for all edge devices in the industrial internet to construct a new round of local digital twin models. In other words, the global digital twin model parameters obtained by the edge devices in each round are the model parameters updated in the previous round.

[0118] The method for constructing a digital twin model based on the Industrial Internet provided in this application involves distributing global digital twin model parameters to multiple edge devices, obtaining the sample data volume and data mapping deviation of the local digital twin models obtained by each edge device after completing one round of training based on the global digital twin model parameters, and then, based on the sample data volume and data mapping deviation of the local digital twin models, determining the edge device selection strategy, resource allocation strategy, and transmit power allocation strategy for this round of training from multiple edge devices with the optimization objective of maximizing the quality of the global digital twin model. Following the resource allocation strategy and transmit power allocation strategy, the local digital twin model parameters of the edge devices corresponding to the edge device selection strategy are obtained, and the global digital twin model parameters are updated based on the local digital twin model parameters. This method achieves dynamic device selection and corresponding resource allocation during the construction of the digital twin model, reduces the negative impact caused by physical-virtual mapping deviations, and improves the performance of digital twin models in the Industrial Internet.

[0119] Figure 3 A flowchart illustrating the method for constructing a digital twin model based on the Industrial Internet provided in this application. Figure 2 ,like Figure 3 As shown, in this embodiment... Figure 2Based on the examples, the method for constructing a digital twin model based on the Industrial Internet is described in detail. The method includes:

[0120] S301: Distribute global digital twin model parameters to multiple edge devices.

[0121] S302: Obtain the sample data volume and data mapping deviation of the local digital twin model obtained by each edge device after constructing and completing one round of training based on the global digital twin model parameters.

[0122] Steps S301-S302 are similar to steps S201-S202 above, and will not be repeated here.

[0123] S303: Based on the sample data volume, data mapping deviation, edge device selection strategy, resource allocation strategy, and transmit power allocation strategy of the local digital twin model, construct the error function of the global digital twin model. With minimizing the error function as the optimization objective, determine the edge device selection strategy, resource allocation strategy, and transmit power allocation strategy for this round of training from multiple edge devices.

[0124] The error function is used to reflect the gap between the quality of the digital twin model and its theoretical optimal value.

[0125] For example, based on the first The amount of sample data acquired during the training rounds of the local digital twin model Data mapping bias Edge device selection strategy Resource allocation strategy and transmit power allocation strategy The error function for constructing the global digital twin model can be defined as follows:

[0126]

[0127] Among them, L, , All are constants, representing the number of training rounds. For the first Edge devices during round training Error rate.

[0128] At this point, with minimizing the error function as the optimization objective, the edge device selection strategy for this round of training is determined from multiple edge devices. Resource allocation strategy and transmit power allocation strategy That is, minimizing the gap between the quality of the digital twin model and its theoretical optimal value, as shown in the following formula:

[0129]

[0130] In some embodiments, the following are defined as constraints for minimizing the gap between the quality of the digital twin model and its theoretical optimum:

[0131] The edge device selection strategy corresponds to the resource allocation of each edge device in the resource allocation strategy; the total latency of each edge device is less than or equal to the preset latency threshold; the total energy consumption of each edge device is less than or equal to the preset energy consumption threshold; the sample data volume of the edge device is greater than or equal to the preset minimum data volume; and the transmission power of the edge device is less than or equal to the maximum transmission power.

[0132] For example, the edge device selection strategy corresponds to the resource allocation for each edge device in the resource allocation strategy, including:

[0133] ;

[0134] ;

[0135] ;

[0136] The edge device selection strategy requires that the total latency of all edge devices be less than or equal to a preset latency threshold. ,in, For the first The upper limit of the total latency of all edge devices used in the global digital twin model construction process;

[0137] The edge device selection strategy requires that the total energy consumption of all edge devices be less than or equal to a preset energy consumption threshold. ,in, For the first The upper limit of the total energy consumption of all edge devices used in the construction of the global digital twin model;

[0138] The sample data size of the edge device is greater than or equal to the preset minimum data size: ,in, Minimum limits on the amount of data that edge devices can acquire;

[0139] The transmit power of the edge device is less than or equal to the maximum transmit power when:

[0140] .

[0141] Among them, the The total latency and total energy consumption of all edge devices used in the joint digital twin model construction can be calculated as follows:

[0142]

[0143]

[0144] Among them, total latency Includes: perceived latency (That is, the latency in edge devices acquiring data information), and the energy consumption of distributing the global digital twin model. The delay in creating a partial digital twin model Uploading some digital twin models has a delay. , Represents the maximum latency between all edge devices; total power consumption. Includes perceived energy consumption (That is, the energy consumption of edge devices acquiring data information), energy consumption for creating partial digital twin models. Upload some digital twin models of energy consumption Specifically, the latency and energy consumption of the sensed data can be calculated as follows:

[0145]

[0146]

[0147] in, , Indicates edge device The latency and energy consumption required to sense a single data sample. The latency and energy consumption of creating a partial digital twin model can be expressed as:

[0148]

[0149]

[0150] in, It is the floating-point number for each data sample. This represents the total number of floating-point operations used to create a partial digital twin model. It is the number of floating-point operations per CPU cycle, and This is the available CPU frequency. The energy consumption per unit of computing resources is... ,in It is an effective switched capacitor. After creating a partial digital twin model, the edge device needs to upload the corresponding model, and the upload rate is defined as:

[0151]

[0152] in, Indicate whether to... Resource blocks are allocated to edge devices , This indicates that each edge device can occupy at most one resource block; This is the bandwidth of each resource block; Indicates to Expectations; Denotes the channel gain, where, It is the distance from the device to the base station. It is the Rayleigh fading parameter; Indicates the first Wheel edge device The transmission power; It is the noise power spectral density. Indicates the use of the first Interference from other devices in the resource block.

[0153] The latency and energy consumption of the corresponding upload portion of the digital twin model can be expressed as follows:

[0154]

[0155]

[0156] in, Representing a partial digital twin model The number of bits in the parameters. After the global digital twin model aggregation is completed, the base station sends new digital twin model parameters to the participating edge devices. The corresponding downlink rate can be expressed as:

[0157]

[0158] in, It's the base station's bandwidth. It is the base station's transmission power. The interference is caused by other base stations; therefore, the delay in the base station transmitting the digital twin model parameters is:

[0159]

[0160] in, This represents the number of bits in the global digital twin model parameters.

[0161] In some embodiments, the construction of the error function includes: determining the error rate of the edge device based on the resource allocation strategy and the transmit power allocation strategy; and constructing the error function of the global digital twin model based on the sample data volume of the local digital twin model, the data mapping deviation, the edge device selection strategy, and the error rate of the edge device.

[0162] The error rate reflects the probability of errors occurring during data transmission between edge devices and the central server.

[0163] Understandably, the error rate of edge devices directly affects the accuracy of their data acquisition and transmission. A high error rate in edge devices indicates that the global digital twin model built based on the corresponding erroneous data will inevitably have significant errors. Therefore, by determining the error rate of edge devices based on resource allocation and transmit power allocation strategies, and comprehensively considering the sample data volume of the local digital twin model, data mapping deviation, edge device selection strategy, and edge device error rate to construct the error function of the global digital twin model, the quality of the global model can be comprehensively and accurately evaluated. This helps to optimize edge device selection, resource allocation, and transmit power allocation strategies in practical applications, improve the accuracy and reliability of the global digital twin model, and thus better support the monitoring, prediction, and optimization of physical entities.

[0164] In some embodiments, the error rate is determined based on resource allocation strategy, transmit power allocation strategy, total number of resource blocks, waterfall threshold of the impact of received signal quality on transmission error rate, bandwidth of resource blocks, channel gain expectation, distance of the edge device, Rayleigh fading parameter, noise power spectral density, and interference generated by other edge devices using resource blocks.

[0165] For example, edge devices error rate for:

[0166]

[0167] in, It is a waterfall threshold used to simulate the impact of received signal quality on transmission error rate. This represents the total number of resource blocks.

[0168] In other embodiments, wireless communication errors are unavoidable during the construction of the federated digital twin model; and transmission errors degrade the quality of the global digital twin model. Therefore, a CRC mechanism is used at the base station receiver to verify the uploaded parameters.

[0169] In some embodiments, during the process of determining the edge device selection strategy, resource allocation strategy, and transmit power allocation strategy for the current training round from multiple edge devices with the goal of minimizing the error function, a corresponding reward function is determined based on the error function, and a reinforcement learning algorithm is used to determine the edge device selection strategy, resource allocation strategy, and transmit power allocation strategy for the current training round based on the reward function.

[0170] For example, agent optimization is performed based on a soft proximal policy optimization algorithm and an actor-critic algorithm. The agent makes real-time dynamic decisions, specifically including device selection strategies. Communication resource block allocation strategy Transmit power allocation strategy The optimization objective is to minimize the gap between the quality of the digital twin model and its theoretical optimal value. This is achieved by training an agent in a networked digital twin model building environment, continuously learning the impact of decisions on the system state, and then adaptively optimizing the selection of edge devices and resource allocation in each round. This reduces the negative impact of physical-virtual mapping deviations and communication uncertainties, thus achieving high-quality digital twin model construction. Specifically, the... The reward function during the training round is defined as follows: :

[0171]

[0172] In some embodiments, the state space in the reinforcement learning algorithm includes the edge device selection strategy, resource allocation strategy, and transmit power allocation strategy of the previous training round, as well as the data mapping bias and the cumulative error of the global digital twin model; the action space in the reinforcement learning algorithm includes the edge device selection strategy, resource allocation strategy, and transmit power allocation strategy of the current training round.

[0173] For example, since the device selection, resource block allocation, and transmit power allocation strategies in each round of the digital twin model building system depend only on the decisions of the previous round and the resulting system state, this means that the state transitions satisfy the MDP; therefore, the agent in the first round... The state of a training round can be defined as:

[0174]

[0175] in, The first Edge device selection, resource block allocation, and power allocation strategies for the wheel. For the first Dynamic mapping deviation of the wheel, To reach the first The upper bound of the cumulative convergence error of the wheel;

[0176] The agent in the first The motion of a wheel can be defined as:

[0177]

[0178] in, The first Edge device selection, resource block allocation, and power allocation strategies for the wheel.

[0179] For example, the agent in the first... The transition probability of a training round can be defined as: Essentially, it represents the probability distribution of the agent's decision based on the observed state; the decision is based on the Soft Proximal Policy Optimization (SPPO) algorithm; whereby the SPPO algorithm extends the PPO algorithm by introducing maximum entropy reinforcement learning, transforming the optimization objective into maximizing both reward and policy entropy, thus avoiding getting trapped in local optima; furthermore, by introducing entropy regularization into the optimization objective, the policy is prompted to randomly select from multiple near-optimal behaviors, making the agent more adaptable to environmental changes; therefore, the final policy is more robust to unpredictable changes or noise, thus better able to handle complex dynamic environments. The optimal policy objective of this algorithm can be mathematically expressed as:

[0180]

[0181] in, It is the entropy regularization coefficient. It is a measurement strategy The entropy of the randomness level.

[0182] This method utilizes an actor-critic framework to help the agent find the optimal strategy. The complete training process is as follows:

[0183] In the In the training round, the actor network representing the old strategy It will copy the actor network from the previous update. The parameters, Used for interacting with the environment and sampling actions, This is used to update policy parameters; actor network Observe the current state Execute actions Receive the next state and calculate the action The resulting rewards Subsequently, the replay buffer stores the current state. ,action ,award and the next state Once the replay buffer reaches its maximum capacity... The buffer will be cleared, and the length selected from the buffer will be used. The trajectory is used to train the network of actors and critics. The agent's future cumulative reward, i.e., the discounted reward, can be calculated as follows:

[0184]

[0185] So, the first The advantage function in a round can be calculated as follows:

[0186]

[0187] in, Indicates the state The following is based on the current value network (parameters are) The estimated state value function is the expected cumulative reward that an agent can obtain by starting from this state and acting according to the current policy π.

[0188] This advantage function Indicates the state given Take action below Compared to the average behavior of the strategy, this function considers the rewards and state values ​​over the entire cycle from the current round to the last round, evaluating the action. Advantages relative to other possible actions under the same conditions.

[0189] Finally, based on the loss function and Update the parameters of the policy network and parameters of the value network To improve the strategy and thus achieve a higher expected cumulative reward, namely:

[0190]

[0191] in, It is the ratio between the old and new strategies, which can be defined as:

[0192]

[0193] The objective of the value part is to minimize the squared error loss of the value estimate:

[0194]

[0195] S304: According to the resource allocation strategy and transmit power allocation strategy, obtain the local digital twin model parameters of the edge device corresponding to the edge device selection strategy, and update the global digital twin model parameters based on the local digital twin model parameters.

[0196] Step S304 is similar to step S204 above, and will not be described again here.

[0197] The method for constructing a digital twin model based on the Industrial Internet provided in this application involves distributing global digital twin model parameters to multiple edge devices, obtaining the sample data volume and data mapping deviation of the local digital twin models obtained by each edge device after completing one round of training based on the global digital twin model parameters, constructing an error function for the global digital twin model based on the sample data volume, data mapping deviation, edge device selection strategy, resource allocation strategy, and transmit power allocation strategy of the local digital twin model, minimizing the error function as the optimization objective, and determining the edge device selection strategy, resource allocation strategy, and transmit power allocation strategy for this round of training from multiple edge devices, obtaining the local digital twin model parameters of the edge devices corresponding to the edge device selection strategy according to the determined resource allocation strategy and transmit power allocation strategy, and updating the global digital twin model parameters based on the local digital twin model parameters. This method, applied in the Industrial Internet, reduces the impact of abnormal industrial data through distributed training, reduces data collection and transmission costs, and simultaneously realizes dynamic device selection and corresponding resource allocation during the construction of the digital twin model, reducing the negative impact caused by physical-virtual mapping deviation and improving the performance of digital twin models in the Industrial Internet.

[0198] Figure 4 A schematic diagram of the structure of the device for constructing a digital twin model based on the Industrial Internet provided in this application is shown below. Figure 4 As shown, the digital twin model construction device 40 based on the Industrial Internet provided in this embodiment includes:

[0199] The processing module 401 is used to send global digital twin model parameters to multiple edge devices.

[0200] The acquisition module 402 is used to acquire the sample data volume and data mapping deviation of the local digital twin model obtained by each edge device after constructing and completing one round of training based on the parameters of the global digital twin model.

[0201] The processing module 401 is further configured to determine the edge device selection strategy, resource allocation strategy, and transmit power allocation strategy for this round of training from the multiple edge devices, based on the sample data volume and data mapping deviation of the local digital twin model and with the optimization objective of maximizing the quality of the global digital twin model.

[0202] The acquisition module 402 is further configured to acquire local digital twin model parameters of the edge device corresponding to the edge device selection strategy, according to the resource allocation strategy and the transmit power allocation strategy.

[0203] The processing module 401 is further configured to update the global digital twin model parameters based on the local digital twin model parameters.

[0204] Optionally, the processing module 401 is further configured to construct an error function of the global digital twin model based on the sample data volume, data mapping deviation, edge device selection strategy, resource allocation strategy, and transmit power allocation strategy of the local digital twin model, with minimizing the error function as the optimization objective, and determine the edge device selection strategy, resource allocation strategy, and transmit power allocation strategy for this round of training from the multiple edge devices, wherein the quality of the global digital twin model is negatively correlated with the error of the global digital twin model.

[0205] Optionally, the processing module 401 is further configured to determine the error rate of the edge device based on the resource allocation strategy and the transmit power allocation strategy.

[0206] The processing module 401 is further configured to construct an error function for the global digital twin model based on the sample data volume of the local digital twin model, the data mapping deviation, the edge device selection strategy, and the error rate of the edge device.

[0207] Optionally, the processing module 401 is further configured to determine the error rate of the edge device based on the resource allocation strategy, the transmit power allocation strategy, the total number of resource blocks, the waterfall threshold for the impact of received signal quality on the transmission error rate, the bandwidth of the resource block, the channel gain expectation, the distance of the edge device, the Rayleigh fading parameter, the noise power spectral density, and the interference generated by other edge devices using the resource block.

[0208] Optionally, the processing module 401 is further configured to determine a reward function based on the error function.

[0209] The processing module 401 is further configured to determine the edge device selection strategy, resource allocation strategy, and transmit power allocation strategy for this round of training based on the reward function and using a reinforcement learning algorithm.

[0210] The device for constructing a digital twin model based on the Industrial Internet provided in this embodiment can execute the method provided in the above method embodiment. Its implementation principle and technical effect are similar, and will not be described in detail here.

[0211] Figure 5 A schematic diagram of the structure of the electronic device provided in this application. Figure 5 As shown, the electronic device 50 provided in this embodiment includes at least one processor 501 and a memory 502. Optionally, the device 50 further includes a communication component 503. The processor 501, memory 502, and communication component 503 are connected via a bus 504.

[0212] In a specific implementation, at least one processor 501 executes computer execution instructions stored in memory 502, causing at least one processor 501 to perform the above-described method.

[0213] The specific implementation process of processor 501 can be found in the above method embodiments, and its implementation principle and technical effect are similar. It will not be repeated here.

[0214] In the above embodiments, it should be understood that the processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), etc. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in this invention can be directly implemented by a hardware processor, or implemented by a combination of hardware and software modules within the processor.

[0215] The memory may include random access memory (RAM) and may also include non-volatile memory (NVM), such as at least one disk storage device.

[0216] The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc. For ease of illustration, the buses shown in the accompanying drawings are not limited to a single bus or a single type of bus.

[0217] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the above-described method.

[0218] This application also provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the above-described method.

[0219] The aforementioned readable storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. The readable storage medium can be any available medium accessible to a general-purpose or special-purpose computer.

[0220] An exemplary readable storage medium is coupled to a processor, enabling the processor to read information from and write information to the readable storage medium. Of course, the readable storage medium can also be a component of the processor. The processor and the readable storage medium can reside in an Application Specific Integrated Circuit (ASIC). Alternatively, the processor and the readable storage medium can exist as discrete components in the device.

[0221] The division of units is merely a logical functional division; in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be indirect coupling or communication connection through some interfaces, devices, or units, and may be electrical, mechanical, or other forms.

[0222] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0223] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0224] If a function is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0225] Those skilled in the art will understand that all or part of the steps of the above-described method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When executed, the program performs the steps of the above-described method embodiments; and the aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disks, or optical disks.

[0226] Finally, it should be noted that other embodiments of the invention will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This invention is intended to cover any variations, uses, or adaptations of the invention that follow the general principles of the invention and include common knowledge or customary techniques in the art not disclosed herein, and is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of the invention is limited only by the appended claims.

Claims

1. A method for constructing a digital twin model based on an industrial internet, characterized in that, The method comprises: downloading a global digital twin model parameter to a plurality of edge devices; obtaining sample data quantity and data mapping deviation of a local digital twin model of each edge device after the edge device constructs and completes a round of training based on the global digital twin model parameter, the data mapping deviation being used to describe a ratio between accurate data quantity obtained by the edge device and actual obtained data quantity; determining, from the plurality of edge devices, an edge device selection strategy, a resource allocation strategy and a transmission power allocation strategy for the current round of training according to the sample data quantity and the data mapping deviation of the local digital twin model, with the optimization objective being to maximize the quality of the global digital twin model; obtaining the local digital twin model parameter of the edge device corresponding to the edge device selection strategy according to the resource allocation strategy and the transmission power allocation strategy, and updating the global digital twin model parameter based on the local digital twin model parameter; the determining, from the plurality of edge devices, the edge device selection strategy, the resource allocation strategy and the transmission power allocation strategy for the current round of training according to the sample data quantity and the data mapping deviation of the local digital twin model, with the optimization objective being to maximize the quality of the global digital twin model, comprises: constructing an error function of the global digital twin model based on the sample data quantity, the data mapping deviation, the edge device selection strategy, the resource allocation strategy and the transmission power allocation strategy of the local digital twin model, with the optimization objective being to minimize the error function, and determining, from the plurality of edge devices, the edge device selection strategy, the resource allocation strategy and the transmission power allocation strategy for the current round of training, wherein the quality of the global digital twin model is negatively correlated with the error of the global digital twin model.

2. The method of claim 1, wherein, the constructing the error function of the global digital twin model based on the sample data quantity, the data mapping deviation, the edge device selection strategy, the resource allocation strategy and the transmission power allocation strategy of the local digital twin model, comprises: determining an error rate of the edge device based on the resource allocation strategy and the transmission power allocation strategy; constructing the error function of the global digital twin model based on the sample data quantity, the data mapping deviation, the edge device selection strategy and the error rate of the edge device of the local digital twin model.

3. The method of claim 1, wherein, Further comprising: determining the following as optimization objectives: the edge device selection strategy corresponds to respective resource allocation of each edge device in the resource allocation strategy; the total time delay of each edge device corresponding to the edge device selection strategy is less than or equal to a preset time delay threshold; the total energy consumption of each edge device corresponding to the edge device selection strategy is less than or equal to a preset energy consumption threshold; the sample data quantity of the edge device is greater than or equal to a preset minimum data quantity; the transmission power of the edge device is less than or equal to a maximum transmission power.

4. The method of claim 2, wherein, the determining the error rate of the edge device based on the resource allocation strategy and the transmission power allocation strategy, comprises: determine an error rate of the edge device based on the resource allocation strategy, the transmit power allocation strategy, a total number of resource blocks, a waterfall threshold of a received signal quality to a transmission error rate, a bandwidth of the resource blocks, a channel gain expectation, a distance of the edge device, a Rayleigh fading parameter, a noise power spectral density, and interference generated by other edge devices using the resource blocks.

5. The method of claim 1, wherein, The optimization objective of minimizing the error function is used to determine the edge device selection strategy, the resource allocation strategy, and the transmit power allocation strategy in the current round of training from the plurality of edge devices, including: determine a reward function based on the error function; determine the edge device selection strategy, the resource allocation strategy, and the transmit power allocation strategy in the current round of training based on the reward function using a reinforcement learning algorithm.

6. The method of claim 5, wherein, The state space in the reinforcement learning algorithm includes the edge device selection strategy, the resource allocation strategy, and the transmit power allocation strategy in the previous round of training, and the accumulated error of the data mapping deviation and the global digital twin model. The action space in the reinforcement learning algorithm includes the edge device selection strategy, the resource allocation strategy, and the transmit power allocation strategy in the current round of training. 7.A device for constructing an industrial internet-based digital twin model, characterized in that, including: a processing module configured to distribute global digital twin model parameters to a plurality of edge devices; an acquisition module configured to acquire sample data volume and data mapping deviation of a local digital twin model of each edge device after the edge device constructs and completes a round of training based on the global digital twin model parameters, the data mapping deviation being used to describe a ratio between accurate data volume and actual acquired data volume of the edge device; The processing module is further configured to determine the edge device selection strategy, the resource allocation strategy, and the transmit power allocation strategy in the current round of training from the plurality of edge devices according to the sample data volume and the data mapping deviation of the local digital twin model, with the optimization objective of maximizing the quality of the global digital twin model; The acquisition module is further configured to acquire local digital twin model parameters of the edge device corresponding to the edge device selection strategy according to the resource allocation strategy and the transmit power allocation strategy; The processing module is further configured to update the global digital twin model parameters based on the local digital twin model parameters; The processing module is further configured to determine the edge device selection strategy, the resource allocation strategy, and the transmit power allocation strategy in the current round of training from the plurality of edge devices according to the sample data volume and the data mapping deviation of the local digital twin model, with the optimization objective of maximizing the quality of the global digital twin model. The processing module is further configured to construct an error function of the global digital twin model based on the sample data volume, the data mapping deviation, the edge device selection strategy, the resource allocation strategy, and the transmit power allocation strategy of the local digital twin model, with the optimization objective of minimizing the error function, to determine the edge device selection strategy, the resource allocation strategy, and the transmit power allocation strategy in the current round of training from the plurality of edge devices, and the quality of the global digital twin model is negatively correlated with the error of the global digital twin model.

8. An electronic device, comprising: including: a memory; a processor; The memory stores computer-executable instructions; The processor executes the computer-executable instructions stored in the memory, so that the processor executes the method according to any one of claims 1-6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions, and the computer-executable instructions are executed by the processor to implement the method according to any one of claims 1-6.

Images