Industrial edge collaborative scheduling optimization method and system based on federal digital twinning

Abstract

The invention relates to the technical field of industrial edge scheduling optimization, in particular to an industrial edge collaborative scheduling optimization method and system based on federal digital twinning. The method comprises the following steps: extracting a compressed feature vector by using local convolution and principal component analysis; mapping the feature vectors into a digital twinborn model of the edge nodes; calculating a scheduling weight score of each edge node based on a digital twin model; a global model based on a neural network is constructed, and global model parameters and a learning rate are initialized based on scheduling weight scores; combining a hyperbolic tangent function and a collaborative credible score to aggregate the updated parameter gradient of each edge node, and calculating to obtain a task migration score; and performing migration scheduling according to the task migration score.

Description

Technical Field

[0001] This invention relates to the field of industrial edge scheduling optimization technology, and in particular to an industrial edge collaborative scheduling optimization method and system based on federated digital twins. Background Technology

[0002] With the rapid development of the Industrial Internet and intelligent manufacturing, more and more industrial enterprises are deploying edge computing nodes and digital twin models to improve data acquisition efficiency, process visualization capabilities, and distributed intelligent decision-making. Digital twin technology, through real-time modeling and mapping of physical objects, enables prediction, optimization, and control in virtual space, serving as a crucial infrastructure for industrial intelligence. Meanwhile, edge computing reduces transmission latency and enhances system responsiveness by pushing computing tasks closer to the data source at edge nodes. In recent years, to address the issues of model synchronization and privacy protection among multiple edge nodes, federated learning has been gradually introduced into industrial edge scenarios for collaborative training of shared models without directly exchanging raw data.

[0003] However, the existing industrial edge collaborative computing methods still have several problems. First, in large-scale distributed environments, digital twin nodes face enormous pressure to synchronize spatiotemporal data. Traditional compression algorithms struggle to balance data fidelity and transmission efficiency, leading to lag and distortion in the twin model. Second, edge nodes exhibit heterogeneous computing power and frequently changing tasks. Current federated training algorithms often employ static weighted averaging, which fails to adequately reflect differences in node capabilities and state changes, severely impacting model convergence quality. Furthermore, existing computational flow scheduling strategies are mostly based on fixed rules or single-factor scoring, lacking a comprehensive understanding of network bandwidth fluctuations, task urgency, and node stability, easily leading to uneven allocation of computing resources and system load imbalance.

[0004] Therefore, there is an urgent need for an innovative collaborative computing system that integrates efficient compressed modeling, adaptive federated training, and intelligent task migration mechanisms to meet the pressing requirements for low latency, high throughput, and strong robustness in large-scale industrial edge scenarios. Summary of the Invention

[0005] To address the aforementioned problems, this invention provides an industrial edge collaborative scheduling optimization method and system based on federated digital twins. This solves the technical problems of low spatiotemporal data synchronization efficiency, rigid edge collaborative optimization mechanisms, and lack of intelligence in computing task migration in multi-node industrial scenarios.

[0006] Firstly, the present invention provides an industrial edge collaborative scheduling optimization method based on federated digital twins, which adopts the following technical solution: An industrial edge collaborative scheduling optimization method based on federated digital twins includes: Acquire industrial time-series data from edge nodes; Compressed feature vectors are extracted using local convolution and principal component analysis. Map feature vectors to the digital twin model of edge nodes; The scheduling weight score for each edge node is calculated based on the digital twin model; A global model based on a neural network is constructed, wherein the global model parameters and learning rate are initialized based on the scheduling weight score; the updated parameter gradients of each edge node are aggregated by combining the hyperbolic tangent function and the collaborative credibility score, and the task transfer score is calculated. Migration scheduling is performed based on task migration scores.

[0007] Furthermore, the extraction of compressed feature vectors using local convolution and principal component analysis includes first applying local convolution operations to the original data to extract local spatial features; then applying principal component analysis (PCA) to the extracted local spatial features, retaining only some principal components, and finally generating compressed feature vectors for the corresponding edge nodes. , in, Indicates the edge node i at time i The compressed feature vector; The original data represents edge node i at time i. Industrial time series data; It refers to the size of the window during local convolution operations; Indicates the number of principal components retained; This indicates that for edge node i at time... raw data Perform convolution processing. Indicates the size of the convolution kernel; This indicates that principal component analysis is performed on the local spatial features extracted by convolution, and the most representative features are retained. One principal component.

[0008] Furthermore, the mapping of feature vectors to the digital twin model of edge nodes includes preprocessing the real-time sensor data and device status information data collected by the edge nodes, including noise removal, data normalization, and feature extraction. The preprocessed data is then combined with compressed feature vectors and resource capability vectors of the edge nodes and input into the digital twin model. The real-time sensor data and device status information are directly mapped and written directly into the corresponding parameters of the existing digital twin model through an API interface. The compressed feature vectors and resource capability vectors of the edge nodes are mapped using a rule engine, which converts them into executable instructions for the digital twin model through preset rules, thereby forming the digital twin model of each edge node. The digital twin model of each edge node represents the current state, resource capabilities, and task execution status of the edge node.

[0009] Furthermore, the calculation of the scheduling weight score for each edge node based on the digital twin model includes calculating the scheduling weight score for each edge node at the current time based on the compressed feature vector and resource capability vector in the digital twin model of each edge node. First, the L2 norm of the compressed feature vector in the digital twin model of the edge node is calculated to measure the importance of the compressed feature vector in its digital twin model. Then, the L2 norm of the resource capability vector is calculated, representing the overall computing power level of the edge node. Finally, the feature strength is divided by the L2 norm of the resource capability vector in the digital twin model of each edge node plus a positive number φ, thereby obtaining the scheduling weight score of edge node i at time φ in the node feature model. Scheduling weight score : , in, To represent a positive number, ensuring that the denominator is not zero in calculations, it is a constant. Represents compressed feature vectors The second norm; The L2 norm represents the resource capability vector.

[0010] Furthermore, the initialization of global model parameters and learning rate based on scheduling weight scoring includes defining a neural network-based global model, initializing the current global model parameters, including the weight matrix, bias vector, and learning rate, using the hyperbolic tangent function. The updated parameter gradients uploaded by each edge node are aggregated with the collaborative trust score, and then federated optimization iterations are performed: , in, Indicates the current number Global model parameters of the wheel; This indicates the global model parameters after the next update. Indicates the learning rate. This represents the total number of edge nodes participating in federated learning; Indicates the first The index of each edge node. Represents edge nodes The updated parameter gradients uploaded to the global model; Represents the hyperbolic tangent function. Represents edge nodes In the Collaborative reliability scoring in round training.

[0011] Furthermore, the aggregation of the updated parameter gradients for each edge node by combining the hyperbolic tangent function and the collaborative credibility score includes first taking the edge node... Scheduling weight scoring in this round As an edge node This provides an overall reflection of resource and data quality, followed by the calculation of the current global model parameters for edge nodes. Gradient of compressed feature vectors in local digital twin models This indicates the current global model's view of edge nodes. The fitting direction of the compressed feature vector in the local digital twin model is related to the parameter update direction of the global model between the previous and current iterations. Perform cosine similarity calculation to measure edge nodes To determine the fit between the current global model's changing trend and the final result, the scheduling weight score is multiplied by the cosine value to obtain the edge node. The collaborative reliability score in this round of training reflects the edge nodes The effectiveness and credibility of collaboration in this round; , in, This indicates the direction and magnitude of global parameter updates between the previous and current rounds, representing the trend of global model changes between the two training rounds.

[0012] Furthermore, the aggregation of the updated parameter gradients for each edge node using the hyperbolic tangent function and collaborative credibility scoring also includes assessing whether the computational task needs to be migrated to other nodes based on the node status and resource capability information in the global digital twin model; firstly, using the edge nodes... Collaborative Trust Scoring in this Round of Training As a fundamental factor, a task urgency function is introduced. The task urgency function is combined with the resource intensity required for the task calculated on edge node i. The task has been waiting time. By taking the logarithm of the product of adding one to each of the two, we ensure that urgent tasks are prioritized during the migration process; finally, we consider the target node. Current bandwidth status Network volatility over the past time window ,in The volatility term represents the variance or standard deviation of bandwidth over a fixed time window in the past, expressed by the bandwidth volatility sensitivity factor. Adjust the weights, and finally use the power exponent. The overall calculation above is enhanced nonlinearly to generate a transfer score. The higher the transfer score, the more suitable the target section is as a transfer target for the current computation flow. The formula is as follows: , in, This represents the calculation of task transfer score, at the [number]th [time]. Round-robin computation tasks start from nodes Assigned to nodes Overall compatibility; This represents the resource intensity requirement of task i.

[0013] Secondly, an industrial edge collaborative scheduling optimization system based on federated digital twins includes: The data acquisition module is configured to acquire industrial time-series data from edge nodes; The feature module is configured to extract compressed feature vectors using local convolution and principal component analysis; The mapping module is configured to map feature vectors to the digital twin model of the edge nodes; The scheduling module is configured to calculate the scheduling weight score for each edge node based on the digital twin model; The global module is configured to build a global model based on a neural network, wherein the global model parameters and learning rate are initialized based on the scheduling weight score; the updated parameter gradients of each edge node are aggregated by combining the hyperbolic tangent function and the collaborative credibility score, and the task transfer score is calculated. The migration module is configured to schedule migrations based on task migration scores.

[0014] Thirdly, the present invention provides a computer-readable storage medium storing a plurality of instructions adapted for loading and execution by a processor of a terminal device of the aforementioned industrial edge collaborative scheduling optimization method based on federated digital twins.

[0015] Fourthly, the present invention provides a terminal device, including a processor and a computer-readable storage medium, wherein the processor is used to implement various instructions; the computer-readable storage medium is used to store multiple instructions, the instructions being adapted to be loaded and executed by the processor to provide the aforementioned industrial edge collaborative scheduling optimization method based on federated digital twins.

[0016] In summary, the present invention has the following beneficial technical effects: 1. Enhance dynamic scheduling and load balancing capabilities of computing flows: By combining federated learning and edge computing, task allocation and computing flow migration can be adjusted in real time to ensure load balancing of each edge node and improve computing efficiency.

[0017] 2. Enhance the system's adaptability: By utilizing the digital twin model of edge nodes, the migration of computing flow can be automatically evaluated and optimized based on factors such as node resource status, the urgency of computing tasks, and bandwidth status, ensuring that tasks are executed on appropriate nodes.

[0018] 3. Improve the execution efficiency of computing tasks: Through computing task migration scoring and collaborative trust scoring, the optimal target node can be selected for computing flow migration, reducing task execution latency and improving overall performance.

[0019] 4. Optimize resource utilization: By accurately assessing the resource capabilities and computing task requirements of each edge node, efficient resource allocation can be achieved, avoiding node overload or resource waste and improving the overall operating efficiency of the system.

[0020] 5. Supports efficient operation of large-scale distributed systems: Suitable for scenarios where multiple edge nodes work together, it can achieve efficient scheduling of computing resources and task management without increasing the burden on the central server, and adapt to complex distributed computing tasks. Attached Figure Description

[0021] Figure 1 This is a structural diagram of an industrial edge collaborative scheduling optimization method based on federated digital twins as described in this invention. Detailed Implementation

[0022] The present invention will be further described in detail below with reference to the accompanying drawings.

[0023] Example 1 Reference Figure 1 This embodiment of an industrial edge collaborative scheduling optimization method based on federated digital twins includes: Acquire industrial time-series data from edge nodes; Compressed feature vectors are extracted using local convolution and principal component analysis. Map feature vectors to the digital twin model of edge nodes; The scheduling weight score for each edge node is calculated based on the digital twin model; A global model based on a neural network is constructed, wherein the global model parameters and learning rate are initialized based on the scheduling weight score; the updated parameter gradients of each edge node are aggregated by combining the hyperbolic tangent function and the collaborative credibility score, and the task transfer score is calculated. Migration scheduling is performed based on task migration scores.

[0024] Specifically: S1. Data Acquisition At each edge node, data is collected through data acquisition devices such as sensors to obtain industrial time series data, including motor current, temperature data, vibration data, speed / frequency, load power, etc.

[0025] S2. Multi-node twin modeling and capability assessment: Collect and compress industrial time-series data of edge nodes, extract compressed feature vectors using local convolution and principal component analysis, and map the three-dimensional resource capability vector, the compressed feature vector of each edge node, the real-time sensor data and equipment status information collected by the edge nodes into the digital twin model of each edge node; calculate the scheduling weight score of the edge nodes to assess their comprehensive computing power level. Feature extraction includes extracting compressed feature vectors using local convolution and principal component analysis, as expressed by the following formula: , The constructed 3D resource capability vector includes: for each edge node Define a three-dimensional resource capability vector. These represent the current computing power of the edge node (such as CPU frequency or number of cores), memory capacity (such as available memory space), and network bandwidth status (i.e., data uplink / downlink rate).

[0026] The 3D resource capability vector, the compressed feature vector of each edge node, and the real-time sensor data and device status information collected by the edge nodes are mapped onto the digital twin model of each edge node. This will be discussed later. The preprocessed data, combined with compressed feature vectors and edge node resource capability vectors, are then input into the existing digital twin model. Real-time sensor data and device status information are directly mapped, meaning they are written directly into the corresponding parameters of the existing digital twin model via API interfaces or database links. The compressed feature vectors and edge node resource capability vectors, however, are mapped using a rule engine. This involves converting the data into executable instructions for the digital twin model through preset rules (such as formulas or logical judgments), thus forming the digital twin model for each edge node.

[0027] S3. Feature Extraction At each edge node, industrial time-series data of the current edge node at the current moment is periodically collected as raw data. To improve the synchronization efficiency of the raw data and reduce the transmission load, the raw data undergoes compressed feature extraction. Specifically, firstly, a local convolution operation is applied to the raw data to extract local spatial features; next, principal component analysis (PCA) is applied to the extracted local spatial features, retaining only the most representative principal components to achieve data dimensionality compression. Finally, a compressed feature vector for the corresponding edge node is generated, as shown in the following formula: , in, Indicates the edge node i at time i The compressed feature vector; The original data represents edge node i at time i. Industrial time series data; It is the size of the window in the local convolution operation. Its value is derived from the actual task requirements and the range is [1, 10]. Indicates the number of principal components retained; This indicates that for edge node i at time... raw data Perform convolution processing. Indicates the size of the convolution kernel; This indicates that principal component analysis (PCA) is performed on the local spatial features extracted by convolution, and the most representative features are retained. One principal component.

[0028] S4. Digital Twin Model After obtaining the compressed feature vector of edge node i, it is also necessary to quantize and represent the computational resources of the edge node. For each edge node... Define a three-dimensional resource capability vector. These represent the current computing power of the edge node (such as CPU frequency or number of cores), memory capacity (such as available memory space), and network bandwidth status (i.e., data uplink / downlink rate).

[0029] Real-time sensor data (such as temperature, humidity, and pressure) and device status information (such as power consumption and operating status) collected from edge nodes are preprocessed, including noise removal, data normalization (ensuring data is within the same scale range), and feature extraction (extracting key features) to ensure data quality and availability. The preprocessed data, combined with compressed feature vectors and edge node resource capability vectors, are then input into the existing digital twin model. Real-time sensor data and device status information are directly mapped, i.e., written directly to the corresponding parameters of the existing digital twin model via API interfaces or database links; for example, temperature data is directly written to the ambient temperature parameter in the model, and device power consumption is directly updated to the energy consumption parameter in the model. The compressed feature vectors and edge node resource capability vectors are mapped using a rule engine, i.e., by pre-setting logic according to specific application scenario requirements, the compressed feature vectors are converted into instructions executable by the digital twin model. For example, when the bandwidth element in the resource capability vector is lower than a preset value, the model automatically updates its network availability parameter; thus forming a digital twin model for each edge node.

[0030] Each edge node's digital twin model represents its current state, resource capabilities, and task execution status. Because the state and resource capabilities of edge nodes are constantly changing (e.g., industrial time series data vary over time), the digital twin models of edge nodes need to be updated periodically, incorporating the updated parameters (…). Gradients (etc.) are used for subsequent federated training.

[0031] S5. Scheduling Weight Scoring Next, based on the compressed feature vector and resource capability vector in the digital twin model of each edge node, the scheduling weight score of each edge node at the current time is calculated: First, the L2 norm (i.e., feature strength) of the compressed feature vector in the digital twin model of the edge node is calculated to measure the importance of the compressed feature vector in its digital twin model; then, the L2 norm of the resource capability vector is calculated, representing the overall computing power level of the edge node. Finally, the feature strength is divided by the L2 norm of the resource capability vector in the digital twin model of each edge node plus a positive number φ, thus obtaining the scheduling weight score of edge node i at time φ in the node feature model. Scheduling weight score : , in, To represent a positive number, ensure that the denominator is not zero in the calculation; it should be a constant (e.g., ...). This improves the numerical stability of the formula; Represents compressed feature vectors The second norm; The L2 norm represents the resource capability vector.

[0032] S6. Federated Training and Computational Flow Migration Co-execution: Initialize global model parameters and learning rate; aggregate updated parameter gradients uploaded by each edge node by combining hyperbolic tangent function and collaborative credibility score; calculate task migration score; finally select the target node with the largest migration score for computational flow migration to ensure load balancing and dynamic scheduling.

[0033] S6.1 defines a global model based on a neural network for updating the parameter gradients of digital twin models of edge nodes in a federated learning environment. It supports task scheduling and load balancing in edge computing and includes global model parameters such as weight matrix and bias vector.

[0034] S6.2 Federated Training and Computation Flow Migration Cooperative Execution: The hyperbolic tangent function is used to perform nonlinear transformation smoothing on the parameter gradient uploaded by each edge node. The parameter gradient is then updated by combining the smoothed parameter gradient with the cooperative trust score calculated based on the scheduling weight score. At the same time, the task migration score is calculated based on the cooperative trust score. Finally, the target node with the largest migration score is selected for computation flow migration, thereby achieving load balancing and dynamic scheduling.

[0035] The collaborative reliability of edge nodes is evaluated based on the scheduling weight scoring of edge nodes. Specifically, the edge nodes are first selected... The scheduling weight score at the current moment As an edge node An overall reflection of resource and data quality. Furthermore, computing edge nodes... Gradient of compressed eigenvectors , indicating the current edge node The fitting direction of the compressed feature vector is related to the difference in parameter gradient updates between the previous and current time steps. Cosine similarity is calculated. Finally, the scheduling weight score is multiplied by the cosine value to obtain the edge node. The collaborative trust score at the current moment reflects the edge node's... The effectiveness and credibility of collaboration at the current moment.

[0036] , in, This represents the difference in parameter gradient updates between the previous time step and the current time step, i.e., the trend of parameter gradient changes; The cosine similarity function is used to measure the angular similarity between two vectors, with a value range of [-1, 1]. The collaborative credibility score is used to adjust the contribution of each edge node in the parameter gradient update process. The higher the collaborative credibility score, the more consistent the gradient direction of the current edge node is with the global update direction, i.e., the greater the contribution of the current edge node.

[0037] S6.3 updates the parameter gradients of the digital twin model of edge nodes in a federated learning environment. Specifically, firstly, to improve the stability of the parameter gradient update process, a hyperbolic tangent function is used ( The parameter gradients uploaded by each edge node are smoothed using a nonlinear transformation to mitigate interference from extreme gradients. Next, the smoothed parameter gradients of each edge node are weighted and summed based on the collaborative credibility score of each edge node to form the aggregated parameter gradient. Furthermore, the updated parameter gradient θ(t+1) is obtained. , in, Indicates the current The parameter gradient at time step; Indicates the next moment (the...) The updated parameter gradient at time (time); The learning rate is used to control the step size of each parameter gradient update. Its value is derived from expert experience or adaptive learning rate algorithm, and is set according to the actual situation. The value range is [0.001, 0.01]. This represents the total number of edge nodes participating in federated learning; Indicates the first The index of each edge node. ; Let be the loss function, representing the parameter gradient. At the edge node compressed feature vectors The prediction error used in the prediction process; Indicates the current time at the edge node Update direction; This represents the hyperbolic tangent function, which has the effect of smoothing and limiting extreme values, preventing the training process from becoming unstable due to excessive gradients. Represents edge nodes In the Time-based collaborative credibility scoring; The updated parameter gradient θ(t+1) is fed back to each edge node until the change in the parameter gradient is less than the threshold preset by the expert experience method. The edge nodes continue to perform local training and updates based on the updated parameter gradient.

[0038] S6.4 Based on the node status and resource capacity information of the edge nodes, assess whether it is necessary to migrate the computing tasks to other nodes. Specifically, this includes: using edge nodes... Collaborative credibility scoring As a fundamental factor, a task urgency function is introduced. This is used to characterize the urgency of the computational tasks performed on edge node i. The task urgency function combines the resource intensity required by the computational tasks on edge node i. The task has been waiting time. By taking the logarithm of the product of adding one to each of the two, we ensure that urgent tasks are prioritized during the migration process. Finally, we consider the target node. Current bandwidth status Network volatility over the past time window ,in This represents the variance or standard deviation of bandwidth over a fixed time window, reflecting the degree of change and stability of the target node's network bandwidth. The volatility term is expressed through a bandwidth volatility sensitivity factor. Adjust the weights, and finally use the power exponent. The overall calculation above is enhanced nonlinearly to generate a transfer score. A higher transfer score indicates that the target section is more suitable as a transfer target for the current computational flow. The formula for calculating the task transfer score is: , in, This represents the calculation of task transfer score at time [time]. Computational tasks from edge nodes Assigned to nodes The overall fit is indicated by a higher value, which means a higher degree of matching and a more suitable migration. The resource intensity requirements (such as CPU / GPU time usage, memory requirements, etc.) of the task on edge node i are normalized; This represents the waiting time from submission to the current state of the task on edge node i; The task urgency function, which represents the urgency of a task's execution, is defined as follows: ; Represents the target node The bandwidth load status indicates the target node. At any moment Network bandwidth utilization; This represents the bandwidth fluctuation of target node j, and the variance of the bandwidth of target node j over the past time window, reflecting network stability. This represents the bandwidth fluctuation sensitivity factor, which controls the bandwidth fluctuation. The strength of the impact on migration scores; This represents the adjustment coefficient, which can adjust the response strength to the urgency of the computing task, bandwidth status, and bandwidth fluctuation.

[0039] S6.5 Final Selection of Migration Scoring Largest node As a migration target, soon nodes The value of is used as the migration target, and the node Migrate computational tasks from one node to another. The process is performed at the point of origin to achieve computational flow migration.

[0040] After obtaining the migration scores of all candidate target nodes Then, all reachable nodes are scored and sorted, and the target node with the highest score is selected. As the optimal migration target for the current computing task, the migration score comprehensively considers the source node's collaborative trust score, the urgency of the current task, and the target node's bandwidth load status and bandwidth fluctuation, reflecting the optimal adaptability of the task migration in the current environment.

[0041] S6.6 After the global model completes this round of training, the process of entering the next round of global model training will continue. Each iteration will feed back the updated global model θ(t+1) to each edge node. The edge nodes will continue to train and update locally based on the new global model until the change of global model parameters is less than the threshold preset by the expert experience method, which indicates that the global model has converged and the training is terminated.

[0042] Example 2 This embodiment provides an industrial edge collaborative scheduling optimization system based on federated digital twins, including: The data acquisition module is configured as follows: A computer-readable storage medium storing a plurality of instructions adapted for loading and execution by a processor of a terminal device of the aforementioned industrial edge collaborative scheduling optimization method based on federated digital twins.

[0043] A terminal device includes a processor and a computer-readable storage medium, the processor being used to implement various instructions; the computer-readable storage medium being used to store multiple instructions adapted for loading and execution by the processor of the aforementioned industrial edge collaborative scheduling optimization method based on federated digital twins.

[0044] The above are all preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Therefore, all equivalent changes made in accordance with the structure, shape and principle of the present invention should be covered within the scope of protection of the present invention.

Claims

1. An industrial edge collaborative scheduling optimization method based on federated digital twins, characterized in that, include: Acquire industrial time-series data from edge nodes; Compressed feature vectors are extracted using local convolution and principal component analysis. Map feature vectors to the digital twin model of edge nodes; The scheduling weight score for each edge node is calculated based on the digital twin model; A global model based on a neural network is constructed, wherein the global model parameters and learning rate are initialized based on the scheduling weight score; the updated parameter gradients of each edge node are aggregated by combining the hyperbolic tangent function and the collaborative credibility score, and the task transfer score is calculated. Migration scheduling is performed based on task migration scores.

2. The industrial edge collaborative scheduling optimization method based on federated digital twins according to claim 1, characterized in that, The method of extracting compressed feature vectors using local convolution and principal component analysis includes first applying local convolution operations to the original data to extract local spatial features; then applying principal component analysis (PCA) to the extracted local spatial features, retaining only some principal components, and finally generating compressed feature vectors for the corresponding edge nodes. ， in, This indicates that the edge node 𝑖 is at time 1 / 2. The compressed feature vector; The original data represents the edge node 𝑖 at time 1. Industrial time series data; It refers to the size of the window during local convolution operations; Indicates the number of principal components retained; This indicates that for edge node 𝑖 at time 1 raw data Perform convolution processing. Indicates the size of the convolution kernel; This indicates that principal component analysis is performed on the local spatial features extracted by convolution, and the most representative features are retained. One principal component.

3. The industrial edge collaborative scheduling optimization method based on federated digital twins according to claim 2, characterized in that, The process of mapping feature vectors to the digital twin model of edge nodes includes preprocessing real-time sensor data and device status information collected by edge nodes, including noise removal, data normalization, and feature extraction. The preprocessed data is then combined with compressed feature vectors and resource capability vectors of edge nodes and input into the digital twin model. The real-time sensor data and device status information are directly mapped and written into the corresponding parameters of the existing digital twin model via an API interface. The compressed feature vectors and resource capability vectors of edge nodes are mapped using a rule engine, converting them into executable instructions for the digital twin model through preset rules, thus forming the digital twin model for each edge node. Each edge node's digital twin model represents its current state, resource capabilities, and task execution status.

4. The industrial edge collaborative scheduling optimization method based on federated digital twins according to claim 3, characterized in that, The method for calculating the scheduling weight score of each edge node based on the digital twin model includes calculating the scheduling weight score of each edge node at the current time based on the compressed feature vector and resource capability vector in the digital twin model of each edge node. First, the L2 norm of the compressed feature vector in the digital twin model of the edge node is calculated to measure the importance of the compressed feature vector in its digital twin model. Then, the L2 norm of the resource capability vector is calculated, representing the overall computing power level of the edge node. Finally, the feature strength is divided by the L2 norm of the resource capability vector in the digital twin model of each edge node plus a positive number *k*, thus obtaining the scheduling weight score of the edge node *k* at time *k* in the node feature model. Scheduling weight score : ， in, To represent a positive number, ensuring that the denominator is not zero in calculations, it is a constant. Represents compressed feature vectors The second norm; The L2 norm represents the resource capability vector.

5. The industrial edge collaborative scheduling optimization method based on federated digital twins according to claim 4, characterized in that, The initialization of global model parameters and learning rate based on scheduling weight scoring includes defining a global model based on a neural network, initializing the current global model parameters, including the weight matrix, bias vector, and learning rate, using the hyperbolic tangent function. The updated parameter gradients uploaded by each edge node are aggregated with the collaborative trust score, and then federated optimization iterations are performed: ， in, Indicates the current number Global model parameters of the wheel; This indicates the global model parameters after the next update. Indicates the learning rate. This represents the total number of edge nodes participating in federated learning; Indicates the first The index of each edge node, Represents edge nodes The updated parameter gradients uploaded to the global model; Represents the hyperbolic tangent function. Represents edge nodes In the Collaborative reliability scoring in round training.

6. The industrial edge collaborative scheduling optimization method based on federated digital twins according to claim 5, characterized in that, The method of combining the hyperbolic tangent function and collaborative credibility scoring to aggregate the updated parameter gradients of each edge node includes first taking the edge node... Scheduling weight scoring in this round As an edge node This provides an overall reflection of resource and data quality, followed by the calculation of the current global model parameters for edge nodes. Gradient of compressed feature vectors in local digital twin models This indicates the current global model's view of edge nodes. The fitting direction of the compressed feature vector in the local digital twin model is related to the parameter update direction of the global model between the previous and current iterations. Perform cosine similarity calculation to measure edge nodes To determine the fit between the current global model's changing trend and the final result, the scheduling weight score is multiplied by the cosine value to obtain the edge node. The collaborative reliability score in this round of training reflects the edge nodes The effectiveness and credibility of collaboration in this round; ， in, This indicates the direction and magnitude of global parameter updates between the previous and current rounds, representing the trend of global model changes between the two training rounds.

7. The industrial edge collaborative scheduling optimization method based on federated digital twins according to claim 6, characterized in that, The method of aggregating the updated parameter gradients of each edge node by combining the hyperbolic tangent function and collaborative credibility scoring also includes assessing whether the computational task needs to be migrated to other nodes based on the node status and resource capability information in the global digital twin model; firstly, using the edge nodes... Collaborative Trust Scoring in this Round of Training As a fundamental factor, a task urgency function is introduced. The task urgency function is combined with the resource intensity required for the task calculated on the edge node 𝑖. The task has been waiting time. By taking the logarithm of the product of adding one to each of the two, we ensure that urgent tasks are prioritized during the migration process; finally, we consider the target node. Current bandwidth status Network volatility over the past time window ,in The volatility term represents the variance or standard deviation of bandwidth over a fixed time window in the past, expressed by the bandwidth volatility sensitivity factor. Adjust the weights, and finally use the power exponent. The overall calculation above is enhanced nonlinearly to generate a transfer score. The higher the transfer score, the more suitable the target section is as a transfer target for the current computation flow. The formula is as follows: ， in, This represents the calculation of task transfer score, at the [number]th [time]. Round-robin computation tasks start from nodes Assigned to nodes Overall compatibility; This indicates the resource intensity requirements of task 𝑖.

8. An industrial edge collaborative scheduling optimization system based on federated digital twins, characterized in that, include: The data acquisition module is configured to acquire industrial time-series data from edge nodes; The feature module is configured to extract compressed feature vectors using local convolution and principal component analysis. The mapping module is configured to map feature vectors to the digital twin model of the edge nodes; The scheduling module is configured to calculate the scheduling weight score for each edge node based on the digital twin model; The global module is configured to build a global model based on a neural network, wherein the global model parameters and learning rate are initialized based on the scheduling weight score; the updated parameter gradients of each edge node are aggregated by combining the hyperbolic tangent function and the collaborative credibility score, and the task transfer score is calculated. The migration module is configured to schedule migrations based on task migration scores.

9. A computer-readable storage medium storing a plurality of instructions, characterized in that, The instructions are adapted to be loaded and executed by the processor of the terminal device as described in claim 1, which is an industrial edge collaborative scheduling optimization method based on federated digital twins.

10. A terminal device, comprising a processor and a computer-readable storage medium, wherein the processor is configured to implement instructions; and the computer-readable storage medium is configured to store multiple instructions, characterized in that, The instructions are adapted to be loaded and executed by a processor as described in claim 1, which is an industrial edge collaborative scheduling optimization method based on federated digital twins.

Images