Cloud side-end architecture digital twinning real-time updating method and system under industrial Internet of Things scene
Through cloud edge architecture and incremental learning, real-time update of digital twin models in industrial Internet of Things scenarios is achieved, solving the problem of low data collection and processing efficiency, improving the adaptability and resource utilization efficiency of the model, and adapting to changing data Distribute and support personalized model adjustments.
Patent Information
- Application Number
- CN202410870029.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-01
- Publication Date
- 2024-11-19
AI Technical Summary
In the industrial Internet of Things scenario, there are challenges in data collection and processing, and existing digital twin technologies are difficult to achieve real-time model updates and adapt to changing data distribution, resulting in low model update efficiency and waste of resources.
Adopt the cloud edge architecture, first generate a digital twin model locally, upload it to the edge server for preliminary updates, and then form a digital twin network model in the cloud for convergence and updates. Finally, the updated model is sent to the edge side, and incremental learning is used Only update parameters related to new data, reduce communication overhead and improve model adaptability.
Real-time updates and adaptability are achieved, the accuracy of the model and resource utilization efficiency are improved, the dependence on large-scale data sets is reduced, the data distribution is adapted to changing, and personalized model adjustment is supported.
Smart Images

Figure CN118981937A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to a cloud-edge-end architecture digital twin real-time update method and system in an industrial Internet of Things scenario, belonging to the technical field of digital twin real-time update. Background Art
[0002] With the continuous development of new generation information technology, the number of devices in industrial Internet of Things scenarios continues to increase. With the emergence of various sensors and intelligent devices on the terminal, data collection and processing as well as unified deployment decisions have been greatly challenged. Digital twins, a key technology of new generation information technology, can map the physical world into the virtual world and make predictions, guide decisions, and optimize production based on the large amount of data collected. However, due to the large number of devices in industrial Internet of Things scenarios, there are also great challenges in data collection. Digital twins can enable digital twins in the network to interact by forming a digital twin network in the cloud. Digital twins in the network can connect to each other to transmit model information, thereby avoiding the need to process a large amount of data on the user side, and using incremental learning to achieve real-time updates of user-side models. Summary of the invention
[0003] The purpose of the present invention is to provide a real-time update method for digital twins of a cloud-edge-end architecture in an industrial Internet of Things scenario to solve the problems raised in the above-mentioned background technology.
[0004] In the present invention, first, the model is trained locally, and its digital twin model is generated on the edge side; then, the edge server uploads the received digital twin model to the cloud, and the digital twin models in the cloud form a model network, which can enable the models in the cloud network to interact with each other, share information, and perform fusion updates to make the information elements of the digital twin model more comprehensive. The model updated by the network is transmitted from the cloud to the edge server again, and waits for the fusion update of the model as a teacher model. The end-side user will perform local updates for the new data locally, instead of using the method of retraining the local model using all the data. During the update process, only the parameters related to the new data are updated, while other parameters remain unchanged. Therefore, the communication overhead can be reduced, and by adopting incremental learning, the local model can adapt to the new data faster, because only the parameters related to the new data have changed, which allows the model to adapt to the changing data distribution more flexibly.
[0005] The present invention also provides a cloud-edge-end architecture digital twin real-time update system in an industrial Internet of Things scenario.
[0006] In order to achieve the above object, the present invention provides the following technical solutions:
[0007] A real-time update method for digital twins of a cloud-edge-end architecture in an industrial Internet of Things scenario, wherein the industrial Internet of Things includes a user end, an edge side, and a cloud side; including:
[0008] Generate a digital twin model; the user performs model training locally and generates an initialized digital twin model on the edge server;
[0009] The digital twin model is uploaded to the cloud, where it forms a digital twin network model;
[0010] Digital twin network model fusion update;
[0011] The digital twin model is sent to the edge side;
[0012] Parameter update of local model;
[0013] Digital twin model integration update;
[0014] Feedback and iteration; the iteration termination condition is that the accuracy of the digital twin model reaches 95%.
[0015] Preferably, according to the present invention, the user terminal performs model training locally and generates an initialized digital twin model on the edge server; including:
[0016] The user uses the original local data to perform initial local model training locally. After the initial local model training is completed, the initial local model is uploaded to the server on the edge side to generate an initialized digital twin model, and the generated initialized digital twin model is preliminarily evaluated.
[0017] Further preferably, the calculation formula for generating the initial local model is as follows:
[0018]
[0019] The formula is expressed as, the participants perform gradient descent updates locally;
[0020] is the t+1th iteration of user i, w t is the model produced by the tth iteration, η is the learning rate, g is the gradient, t is the number of iterations, and i is the user.
[0021] Further preferably, when the accuracy requirement is met, that is, when the change in the parameter is less than the minimum value, an initialized digital twin model is generated on the server on the edge side, and the initialized digital twin model is as follows:
[0022]
[0023] M i To initialize the digital twin model, is the model of the t+1th iteration, and i is the user.
[0024] Further preferably, the initialized digital twin model is evaluated by measuring the mean square error, as shown below:
[0025]
[0026] Among them, y represents the actual value, It is represented as the predicted value of the initialized digital twin model, MSE is the mean square error, and n is the number of samples.
[0027] Preferably, according to the present invention, the digital twin model is uploaded to the cloud, and a digital twin network model is formed in the cloud; including:
[0028] After generating a digital twin model that meets the accuracy requirements, the server on the edge side uploads the digital twin model to the cloud and forms a digital twin network model, as shown below:
[0029] DT N ={M1,M2,…,M i}.
[0030] DT N It is a digital twin network model composed of multiple digital twin models.
[0031] Preferably, according to the present invention, the digital twin network model fusion update includes:
[0032] In the cloud, various digital twin models form a digital twin network model to exchange information between digital twin models.
[0033] Preferably, according to the present invention, the parameter updating of the local model includes:
[0034] In the process of updating parameters during local model training, the predicted values of new and old data are calculated using the nearest mean classification algorithm, and the predicted values are brought into the loss function for optimization. The process of updating parameters can be divided into the following steps:
[0035] Step 1: Calculate the category mean
[0036] First, for each category, calculate its feature mean. Assume that there are K categories in the data set, and each category k contains N k samples, the feature vector is x j . Then the mean vector m of category k is k It can be expressed as:
[0037]
[0038] Step 2: Predict the category of new samples
[0039] For each new sample x new , and predict its category by calculating the Euclidean distance between it and the mean of each category. Predicted category It can be expressed as:
[0040]
[0041] Step 3: Calculate the loss function
[0042] The loss function usually uses the cross entropy loss function. For a sample (x j ,y j ), where y j is the true label, and the cross entropy loss can be expressed as:
[0043]
[0044] Among them, y j,c is the true probability that sample i belongs to category c, is the predicted probability.
[0045] Preferably, according to the present invention, the digital twin model fusion update includes:
[0046] The edge server receives the initial model from the client and the digital twin model from the cloud, and receives the digital twin model from the cloud as the teacher model v k To update the model, the teacher model will be used to update the digital twin model on the server on the edge side; as shown below:
[0047]
[0048] is the updated model of user i under the guidance of the teacher model, is the t-th iteration update model for user i, and Фt is the loss function.
[0049] A computer device includes a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the steps of a real-time update method of a cloud-edge-end architecture digital twin in an industrial Internet of Things scenario are implemented.
[0050] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of a real-time update method for a cloud-edge-end architecture digital twin in an industrial Internet of Things scenario.
[0051] A cloud-edge-end architecture digital twin real-time update system in an industrial Internet of Things scenario, comprising:
[0052] The digital twin model generation module is configured to: generate a digital twin model; the user terminal performs model training locally and generates an initial digital twin model on the edge server;
[0053] The digital twin model uploading module is configured to: upload the digital twin model to the cloud and form a digital twin network model in the cloud;
[0054] The digital twin network model fusion update module is configured as follows: digital twin network model fusion update;
[0055] The digital twin model delivery module is configured to: deliver the digital twin model to the edge side;
[0056] The parameter updating module of the local model is configured to: update the parameters of the local model;
[0057] The digital twin model fusion update module is configured as: digital twin model fusion update.
[0058] The beneficial effects of the present invention are:
[0059] 1. Real-time update and adaptability: Incremental learning allows the local model to be updated in time when new data arrives, thereby maintaining the real-time nature of the model. This enables the digital twin model to more accurately reflect the current state of the actual system and increases the adaptability of the model.
[0060] 2. Resource efficiency: Incremental learning is usually more efficient than global retraining because it only requires a small amount of updates using new data. This helps save computing resources and reduces dependence on large-scale datasets.
[0061] 3. Model personalization: The edge server allows different models to be integrated and updated through the teacher model, and the model is adjusted according to personalized data. This can be used in the digital twin model to better reflect the heterogeneity of the system and the differences in different environments. BRIEF DESCRIPTION OF THE DRAWINGS
[0062] Figure 1 A flowchart of a real-time update method for a digital twin of a cloud-edge-end architecture in an industrial Internet of Things scenario provided by the present invention;
[0063] Figure 2 A block diagram of the implementation system of the real-time update method of the cloud-edge-end architecture digital twin in the industrial Internet of Things scenario provided by the present invention. DETAILED DESCRIPTION
[0064] The present invention will be further defined below in conjunction with the accompanying drawings and embodiments, but is not limited thereto.
[0065] Example 1
[0066] A real-time update method for digital twins of cloud-edge-end architecture in industrial Internet of Things scenarios. In the industrial Internet of Things (IIoT) scenario, the real-time update of digital twins of cloud-edge-end architecture involves multiple levels of collaborative work, such as Figure 2 As shown in Figure 1, the Industrial Internet of Things includes the user side, edge side, and cloud side; this architecture is designed to achieve efficient data collection, processing, and feedback so that the digital twin model can reflect the status and behavior of physical entities in real time. Figure 1 As shown, including:
[0067] Generate a digital twin model; the user performs model training locally and generates an initialized digital twin model on the edge server;
[0068] The digital twin model is uploaded to the cloud, where it forms a digital twin network model;
[0069] Digital twin network model fusion update;
[0070] The digital twin model is sent to the edge side;
[0071] Parameter update of local model;
[0072] Digital twin model fusion update;
[0073] Feedback and iteration; the iteration termination condition is that the accuracy of the digital twin model reaches 95%. According to the performance of the digital twin model, feedback can be provided to each participant to improve the quality of the local model. This process can be an iterative process to continuously improve the performance of the digital twin model.
[0074] Example 2
[0075] The difference between the real-time update method of cloud-edge-device architecture digital twin in an industrial Internet of Things scenario described in Example 1 is that:
[0076] The user performs model training locally and generates an initialized digital twin model on the edge server; including:
[0077] The user uses the original local data to perform initial local model training locally. After the initial local model training is completed, the initial local model is uploaded to the server on the edge side to generate an initialized digital twin model, and the generated initialized digital twin model is preliminarily evaluated.
[0078] The calculation formula for generating the initial local model is as follows:
[0079]
[0080] The formula is expressed as, the participants perform gradient descent updates locally;
[0081] is the t+1th iteration of user i, w t is the model produced by the tth iteration, η is the learning rate, g is the gradient, t is the number of iterations, and i is the user.
[0082] When the accuracy requirement is met, that is, when the parameter change is less than the minimum value, an initialized digital twin model is generated on the edge server. The initialized digital twin model is as follows:
[0083]
[0084] M i To initialize the digital twin model, is the model of the t+1th iteration, and i is the user.
[0085] The initialization digital twin model is evaluated by measuring the mean square error, as shown below:
[0086]
[0087] Among them, y represents the actual value, It is represented as the predicted value of the initialized digital twin model, MSE is the mean square error, and n is the number of samples.
[0088] The digital twin model is uploaded to the cloud, where it forms a digital twin network model; including:
[0089] After generating a digital twin model that meets the accuracy requirements, the server on the edge side uploads the digital twin model to the cloud and forms a digital twin network model, as shown below:
[0090] DT N ={M1,M2,…,M i}.
[0091] DT N It is a digital twin network model composed of multiple digital twin models.
[0092] Digital twin network model fusion update; including:
[0093] In the cloud, each digital twin model forms a digital twin network model to exchange information between digital twin models. That is, by exchanging information between models in the digital twin network, each digital twin model can quickly learn the information of other models, making the information elements of the digital twin model more comprehensive.
[0094] Parameter update of local model; including:
[0095] In the process of updating parameters during local model training, the predicted values of new and old data are calculated using the nearest mean classification algorithm, and the predicted values are brought into the loss function for optimization. The process of updating parameters can be divided into the following steps:
[0096] Step 1: Calculate the category mean
[0097] First, for each category, calculate its feature mean. Assume that there are K categories in the data set, and each category k contains N k samples, the feature vector is x j . Then the mean vector m of category k is k It can be expressed as:
[0098]
[0099] Step 2: Predict the category of new samples
[0100] For each new sample x new , and predict its category by calculating the Euclidean distance between it and the mean of each category. Predicted category It can be expressed as:
[0101]
[0102] Step 3: Calculate the loss function
[0103] The loss function usually uses the cross entropy loss function. For a sample (x j ,y j ), where y j is the true label, and the cross entropy loss can be expressed as:
[0104]
[0105] Among them, y j,c is the true probability that sample i belongs to category c, is the predicted probability.
[0106] Digital Twin Fusion Updates; including:
[0107] The edge server receives the initial model from the client and the digital twin model from the cloud, and receives the digital twin model from the cloud as the teacher model v k To update the model, the teacher model will be used to update the digital twin model on the server on the edge side; as shown below: In this way, frequent communication between the end side and the cloud can be avoided, reducing communication latency.
[0108]
[0109] is the updated model of user i under the guidance of the teacher model, is the t-th iteration update model for user i, and Фt is the loss function.
[0110] Example 3
[0111] The difference between the real-time update method of cloud-edge-device architecture digital twin in an industrial Internet of Things scenario described in Example 2 is that:
[0112] This embodiment is a wind turbine digital twin;
[0113] Scenario: Wind farms require real-time monitoring and maintenance of wind turbines.
[0114] The terminal equipment layer includes:
[0115] Sensors: Sensors installed on each wind turbine collect data (such as wind speed, wind direction, temperature, vibration, rotation speed).
[0116] Edge devices: Edge computing nodes (such as industrial PCs or embedded devices) are deployed near wind turbines for preliminary data processing and real-time monitoring.
[0117] The edge layer includes:
[0118] Data processing: Edge devices run data processing algorithms to clean, filter, and aggregate sensor data.
[0119] Real-time analytics: Run lightweight machine learning models (such as anomaly detection) to identify potential failures.
[0120] Local storage and transmission: The processed data and analysis results are sent to the cloud via the MQTT protocol and saved locally to prevent data loss in the event of network interruption.
[0121] Cloud layer:
[0122] Big data storage and management: Use cloud storage (such as AWS S3, Azure Blob Storage) to store large amounts of sensor data and historical data.
[0123] Advanced analytics and prediction: Run complex machine learning models in the cloud for predictive maintenance and performance optimization analysis.
[0124] Digital twin update: Update the digital twin model with the latest data and analysis results, and provide a real-time visualization interface for operator monitoring.
[0125] Example 4
[0126] The difference between the real-time update method of cloud-edge-device architecture digital twin in an industrial Internet of Things scenario described in Example 2 is that:
[0127] This embodiment is for manufacturing workshop equipment monitoring;
[0128] Scenario: Manufacturing workshops need to monitor and maintain various production equipment, such as CNC machine tools, injection molding machines, etc.
[0129] The terminal equipment layer includes:
[0130] Sensors: Various sensors (such as temperature, pressure, vibration, and current) are installed on the equipment.
[0131] Control system: The equipment control system (such as PLC) collects sensor data and communicates with the edge computing node.
[0132] The edge layer includes:
[0133] Edge computing devices: Deploy edge computing devices (such as industrial computers) in the workshop to perform data processing and real-time analysis.
[0134] Real-time monitoring: Edge devices run real-time monitoring applications to monitor device status and performance and issue alerts when anomalies are detected.
[0135] The cloud layer includes:
[0136] Data storage and management: Use cloud storage services (such as Google Cloud Storage) to save historical data and analysis results.
[0137] Machine Learning and Optimization: Perform deep learning and optimization analysis in the cloud to identify potential equipment failure modes and optimize production parameters.
[0138] In the industrial Internet of Things scenario, the real-time update method of digital twins based on the cloud-edge-end architecture can update the digital twin model in real time by collaboratively learning the data of multiple devices, thereby more accurately reflecting the status of the actual industrial process. On the factory production line, real-time data of the equipment is obtained through sensors, and the data is transmitted to the local end for real-time analysis. Using incremental learning, the digital twin model can be continuously updated based on the latest data to better reflect the status of the actual equipment. This helps to improve production efficiency, predict equipment failures, and optimize production plans. Based on the collaboratively updated digital twin model, the system can perform anomaly detection in real time. When a device or sensor observes an anomaly, the information is propagated to the global model, and other devices can benefit from this information to better understand the overall condition of the system.
[0139] Example 5
[0140] A computer device includes a memory and a processor, the memory stores a computer program, and when the processor executes the computer program, it implements the steps of the real-time update method of the cloud-edge-end architecture digital twin in the industrial Internet of Things scenario described in any of Examples 1-4.
[0141] Example 6
[0142] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of a method for real-time updating of a cloud-edge-end architecture digital twin in an industrial Internet of Things scenario as described in any one of Examples 1-4.
[0143] Example 7
[0144] A cloud-edge-end architecture digital twin real-time update system in an industrial Internet of Things scenario, comprising:
[0145] The digital twin model generation module is configured to: generate a digital twin model; the user terminal performs model training locally and generates an initial digital twin model on the edge server;
[0146] The digital twin model uploading module is configured to: upload the digital twin model to the cloud and form a digital twin network model in the cloud;
[0147] The digital twin network model fusion update module is configured as follows: digital twin network model fusion update;
[0148] The digital twin model delivery module is configured to: deliver the digital twin model to the edge side;
[0149] The parameter updating module of the local model is configured to: update the parameters of the local model;
[0150] The digital twin model fusion update module is configured as: digital twin model fusion update.
Claims
1. A real-time update method for digital twins of cloud-edge-device architecture in industrial Internet of Things scenarios. It is characterized in that The Industrial Internet of Things includes the user side, edge side and cloud side; including: Generate a digital twin model; the user performs model training locally and generates an initialized digital twin model on the edge server; The digital twin model is uploaded to the cloud, where it forms a digital twin network model; Digital twin network model fusion update; The digital twin model is sent to the edge side; Parameter update of local model; Digital twin model fusion update; Feedback and iteration; the iteration termination condition is that the accuracy of the digital twin model reaches 95%.
2. According to the method for real-time updating of digital twins of cloud-edge-end architecture in an industrial Internet of Things scenario described in claim 1, It is characterized in that The user performs model training locally and generates an initialized digital twin model on the edge server; including: The user uses the original local data to perform initial local model training locally. After the initial local model training is completed, the initial local model is uploaded to the server on the edge side to generate an initialized digital twin model, and the generated initialized digital twin model is preliminarily evaluated. Further preferably, the calculation formula for generating the initial local model is as follows: The formula is expressed as, the participants perform gradient descent updates locally; is the t+1th iteration of user i, w t is the model produced by the tth iteration, η is the learning rate, g is the gradient, t is the number of iterations, and i is the user.
3. According to the method for real-time updating of digital twins of cloud-edge-end architecture in an industrial Internet of Things scenario described in claim 1, It is characterized in that When the accuracy requirement is met, that is, when the parameter change is less than the minimum value, an initialized digital twin model is generated on the edge server. The initialized digital twin model is as follows: M i To initialize the digital twin model, is the model of the t+1th iteration, and i is the user.
4. According to the method for real-time updating of digital twins of cloud-edge-end architecture in an industrial Internet of Things scenario described in claim 1, It is characterized in that The initialization digital twin model is evaluated by measuring the mean square error, as shown below: Among them, y represents the actual value, It is represented as the predicted value of the initialized digital twin model, MSE is the mean square error, and n is the number of samples.
5. According to the method for real-time updating of digital twins of cloud-edge-end architecture in an industrial Internet of Things scenario described in claim 1, It is characterized in that The digital twin model is uploaded to the cloud, where it forms a digital twin network model; including: After generating a digital twin model that meets the accuracy requirements, the server on the edge side uploads the digital twin model to the cloud and forms a digital twin network model, as shown below: DT N ={M 1 ,M 2 ,…,M i }。 DT N It is a digital twin network model composed of multiple digital twin models; Further preferably, the digital twin network model fusion update includes: In the cloud, various digital twin models form a digital twin network model to exchange information between digital twin models.
6. According to the method for real-time updating of digital twins of cloud-edge-end architecture in an industrial Internet of Things scenario described in claim 1, It is characterized in that Parameter updates of the local model; including: Step 1: Calculate the category mean First, for each category, calculate its feature mean. Assume that there are K categories in the data set, and each category k contains N k samples, the feature vector is x j . Then the mean vector m of category k is k It can be expressed as: Step 2: Predict the category of new samples For each new sample x new , and predict its category by calculating the Euclidean distance between it and the mean of each category. Predicted category It can be expressed as: Step 3: Calculate the loss function The loss function usually uses the cross entropy loss function. For a sample (x j ,y j ), where y j is the true label, and the cross entropy loss can be expressed as: Among them, y j,c is the true probability that sample i belongs to category c, is the predicted probability.
7. A real-time update method for digital twins of a cloud-edge-end architecture in an industrial Internet of Things scenario according to any one of claims 1-6, It is characterized in that Digital Twin Fusion Updates; including: The edge server receives the initial model from the client and the digital twin model from the cloud, and receives the digital twin model from the cloud as the teacher model v k To update the model, the teacher model will be used to update the digital twin model on the server on the edge side; as shown below: is the updated model of user i under the guidance of the teacher model, is the t-th iteration update model for user i, and Фt is the loss function.
8. A computer device comprising a memory and a processor, wherein the memory stores a computer program. It is characterized in that When the processor executes the computer program, it implements the steps of the real-time update method of the cloud-edge-end architecture digital twin in the industrial Internet of Things scenario described in any one of claims 1-7.
9. A computer-readable storage medium having a computer program stored thereon, It is characterized in that When the computer program is executed by a processor, the steps of the real-time update method of the cloud-edge-end architecture digital twin in the industrial Internet of Things scenario described in any one of claims 1-7 are implemented.
10. A cloud-edge-end architecture digital twin real-time update system in the industrial Internet of Things scenario, It is characterized in that include: The digital twin model generation module is configured to: generate a digital twin model; the user terminal performs model training locally and generates an initial digital twin model on the edge server; The digital twin model uploading module is configured to: upload the digital twin model to the cloud and form a digital twin network model in the cloud; The digital twin network model fusion update module is configured as follows: digital twin network model fusion update; The digital twin model delivery module is configured to: deliver the digital twin model to the edge side; The parameter updating module of the local model is configured to: update the parameters of the local model; The digital twin model fusion update module is configured as: digital twin model fusion update.
