Industrial internet of things-oriented digital twin server adaptive deployment method and system

Abstract

The application discloses an industrial Internet of Things-oriented digital twin server adaptive deployment method and system, and relates to the technical field of digital twin. The method constructs a deployment architecture including a physical entity layer and a digital twin layer, and takes an intelligent gateway as a candidate deployment position; a single interaction delay model is constructed by distinguishing two situations of inside and outside the coverage range, a multi-objective constraint optimization problem is established by combining a load balancing model and an energy consumption model; a deployment decision and an associated decision are jointly encoded into a two-section chromosome, an improved non-dominated sorting genetic algorithm is used for iterative solution, four types of repair operations of failure association repair, resource over-limit migration, time delay over-limit addition and idle load abolition are used for adaptive adjustment of the deployment quantity and position according to the constraint satisfaction condition; and finally, a target deployment scheme is determined from a Pareto optimal solution set. The application significantly reduces the average interaction delay while taking into account the load balancing and total energy consumption, and improves the real-time performance and economy of the digital twin service.

Description

Technical Field

[0001] This application relates to the fields of industrial Internet of Things (IIoT) and digital twin technology, specifically to an adaptive deployment method and system for digital twin servers for industrial IoT. Background Technology

[0002] Under the human-centered, sustainable, and highly resilient intelligent manufacturing vision of Industry 5.0, digital twin technology provides crucial support for the status monitoring, fault prediction, and optimized control of intelligent manufacturing equipment by constructing a digital twin that maps in real time to the physical entity in the digital space. The digital twin needs to continuously receive status data uploaded by the intelligent manufacturing equipment and perform synchronous calculations, relying on digital twin servers deployed at the network edge. The deployment location and number of digital twin servers, as well as the relationship between the digital twin and the digital twin servers, directly determine the data transmission path, the response speed of synchronous calculations, and the resource utilization efficiency of the server cluster.

[0003] Existing digital twin server deployment solutions mainly include centralized deployment based on cloud centers, fixed-location deployment based on empirical rules, and heuristic deployment targeting a single latency metric. However, real-world industrial scenarios still face the following problems: First, centralized deployment requires all state data to be transmitted back to the cloud center over long distances, resulting in large interaction latency and making it difficult to meet the real-time requirements of millisecond-level synchronization for digital twins. Second, fixed-location, fixed-quantity deployment methods cannot perceive the spatial distribution differences and resource requirements of intelligent manufacturing equipment, easily leading to some servers being overloaded while others are idle, resulting in load imbalance and energy waste. Third, deployment methods targeting a single metric ignore the interrelationship between interaction latency, load balancing, and energy consumption, and most do not consider the additional latency introduced by gateway relay when equipment is outside the server's communication coverage area, resulting in a disconnect between the deployment scheme and actual network conditions. Fourth, the number of deployments usually needs to be predetermined manually, lacking the ability to adaptively increase or decrease the number of deployments based on constraint satisfaction. When the scale of equipment or latency requirements change, manual parameter tuning is required, resulting in poor scalability.

[0004] Therefore, there is a need for a digital twin server deployment method that can jointly optimize interaction latency, load balancing and energy consumption, distinguish between different transmission paths inside and outside the coverage area, and adaptively adjust the number and location of deployments according to the constraint satisfaction, so as to improve the real-time performance, stability and economy of industrial IoT digital twin services. Summary of the Invention

[0005] In order to solve the above-mentioned technical problems, this application proposes the following technical solution: In a first aspect, embodiments of this application provide an adaptive deployment method for digital twin servers for the Industrial Internet of Things, including: A digital twin deployment architecture is constructed, comprising a physical entity layer and a digital twin layer. The physical entity layer includes multiple smart manufacturing devices and multiple smart gateways. The digital twin server uses the smart gateways as candidate deployment locations. The digital twin layer includes digital twins maintained by the digital twin server that correspond one-to-one with each smart manufacturing device, and obtains the location information, status information, and resource information of each smart manufacturing device and each smart gateway. Based on the location information, the coverage relationship between the intelligent manufacturing equipment and the digital twin server is determined. For the two cases of being within the coverage area and being outside the coverage area, a single interaction latency model of the digital twin is constructed. A load balancing model and an energy consumption model of the digital twin server are also constructed. With average interaction latency, load balancing degree and total energy consumption as optimization objectives, and with association uniqueness, deployment mutual exclusion, resource capacity and interaction latency tolerance as constraints, a multi-objective constraint optimization problem for the deployment of the digital twin server is established. The deployment decision of the digital twin server and the association decision of the digital twin are jointly encoded into a two-segment chromosome. The two-segment chromosome includes a deployment gene segment that characterizes whether each smart gateway deploys a digital twin server, and an association gene segment that characterizes the digital twin server associated with each digital twin. The population is then initialized. An improved non-dominated sorting genetic algorithm is used to iteratively solve the multi-objective constrained optimization problem. During the iteration process, constraint repair operations are performed on chromosomes that do not meet the constraints. The constraint repair operations are carried out by redirecting failure associations, migrating overloaded loads, adding digital twin servers, and removing idle digital twin servers, so that the number and location of digital twin servers are adaptively adjusted according to the constraint satisfaction status. The target deployment scheme is determined from the Pareto optimal solution set obtained by iterative solution. The number of digital twin servers, their deployment locations, and the association between each digital twin and the digital twin server are determined according to the target deployment scheme. The deployment of the digital twin servers is then completed.

[0006] In one possible implementation, the construction of a digital twin deployment architecture comprising a physical entity layer and a digital twin layer includes: Let the set of intelligent manufacturing equipment in the physical entity layer be denoted as... Let the set of smart gateways be denoted as ,in, This indicates the number of intelligent manufacturing equipment. Indicates the number of smart gateways; Will be deployed on smart gateways The digital twin server on the server is denoted as Digital twin server Represented as a quadruple ,in, This indicates the location information of the digital twin server. Indicates that the digital twin server is in Status information at any given moment Indicates that the digital twin server is in Real-time resource information, Indicates that the digital twin server is in The set of digital twins maintained at any given moment; Intelligent manufacturing equipment The corresponding digital twin is denoted as Digital twin Represented as a quadruple ,in, This indicates the location information of the corresponding intelligent manufacturing equipment. Indicating digital twins in Real-time synchronized device status information, This indicates the resource information required for the operation of the digital twin. This represents the maximum tolerable interaction latency of the digital twin; Define related variables , Indicating intelligent manufacturing equipment Digital twins By smart gateway Maintenance of the digital twin server on the platform. This indicates that there is no maintenance relationship.

[0007] In one possible implementation, the step of constructing a single interaction latency model for the digital twin for both the case within and outside the coverage area includes: Computational intelligent manufacturing equipment With digital twin server Wireless transmission rate between: ; in: Indicates the wireless channel bandwidth. Indicating intelligent manufacturing equipment The transmission power, Indicating intelligent manufacturing equipment Channel gain between the smart gateway and the access gateway Indicates noise power; When intelligent manufacturing equipment With digital twin server Distance between No larger than the communication coverage radius of the digital twin server When the intelligent manufacturing equipment is within coverage area, its status data is directly uploaded to the digital twin server via a wireless link; when the distance... Greater than the communication coverage radius When it is determined that the intelligent manufacturing equipment is outside the coverage area, the status data is first uploaded to the intelligent gateway to which the equipment is connected via a wireless link. Then by the smart gateway Relayed to the digital twin server via wired backhaul link The uplink transmission delay is obtained as follows: ; in: Indicating intelligent manufacturing equipment The amount of status data uploaded in a single instance. Indicating intelligent manufacturing equipment Its connected smart gateway Wireless transmission rate between Indicates smart gateway With digital twin server The distance between the wired links, Represents the relay delay coefficient per unit distance of data volume on a wired backhaul link; Calculate the synchronization computation latency for the digital twin server to complete one digital twin state synchronization: ; in: This indicates the number of processing cycles required to synchronize a unit of data. Digital twin server Assigned to digital twin Computing resources; The latency of a single interaction is expressed as the sum of the uplink transmission latency and the synchronization calculation latency: ; For observation duration The time-averaged interaction latency is obtained by taking the average latency of each single interaction within the time frame. ; In one possible implementation, the load balancing model and energy consumption model for constructing the digital twin server include: Computational digital twin server exist Load at any moment: ; in: Digital twin server The computing resource capacity, Digital twin server Memory resource capacity, Digital twin server Assigned to digital twin memory resources, , These represent the weighting coefficients for computing resource utilization and memory resource utilization, respectively. Load balancing is measured by the standard deviation of the load on deployed digital twin servers: ; in: Indicates smart gateway Deployment variables for whether to deploy a digital twin server. This indicates the number of digital twin servers deployed. This indicates the average load and load balancing of the deployed digital twin servers. The smaller the value, the more balanced the load is among the digital twin servers; The power model is used to calculate the digital twin server based on a linear correlation with processor utilization. exist Operating power at any given time: ; in: This indicates the idle power of the digital twin server. This indicates the full-load power of the digital twin server. Digital twin server exist Processor utilization at any given moment.

[0008] In one possible implementation, the multi-objective constrained optimization problem for establishing the deployment of the digital twin server includes: With the optimization objectives of minimizing average interaction latency, load balancing, and total energy consumption, the following multi-objective constrained optimization problem is established: ; in: Indicates by deployment variables The deployment decision vector constituted Indicates the association variables The associated decision matrix; constraints This means that each digital twin is maintained by one and only one digital twin server, which is a constraint. This means that a digital twin can only be associated with a smart gateway that has a deployed digital twin server, and each smart gateway can deploy at most one digital twin server, constraining... This indicates that the computing and memory resource requirements of each digital twin server do not exceed its resource capacity, constraining... This indicates that the latency of a single interaction of each digital twin does not exceed its maximum tolerable interaction latency, constraining... This indicates the range of values ​​for deployment variables and associated variables.

[0009] In one possible implementation, the joint encoding of the deployment decision of the digital twin server and the association decision of the digital twin into a two-segment chromosome, and the initialization of the population, includes: The chromosome is divided into deployment gene segments and associated gene segments, wherein the deployment gene segments include The first deployment gene using 0-1 encoding, the... A deployment gene of 1 indicates that in the smart gateway A digital twin server is deployed on top of the system, with 0 indicating no deployment; the associated gene segment contains The associated genes encoded with integers, the first Related genes , Indicating intelligent manufacturing equipment The digital twin is a smart gateway The digital twin server is maintained, and the deployment variables and associated variables are arranged according to... Decoding, in which Indicates an indicator function; The optimal point set method is used to generate an initial population within the gene value space, constructing optimal points: ; in: This indicates taking the decimal part. The genetic dimension of a chromosome. Indicates satisfaction The smallest prime number; The first The individual Dimensional gene mapping to value range : ; in: , They represent the first The lower and upper bounds of the value range of the dimensional gene; Deployed gene segments are discretized according to a threshold; if the mapping value is not less than 0.5, the deployed gene is set to 1; otherwise, it is set to 0. Associated gene segments are discretized according to... Round up to obtain the associated gene value.

[0010] In one possible implementation, performing constraint repair operations on chromosomes that do not meet the constraints during the iteration process includes: Failure association repair: When the associated gene points to a smart gateway that has not deployed a digital twin server, traverse the deployed digital twin servers and redirect the associated gene to the digital twin server that meets the resource capacity constraints and has the smallest interaction latency increment. Resource Exceedance Migration and Repair: When the computing or memory resource requirements of a digital twin server exceed its resource capacity, the digital twins maintained by the digital twin server are migrated one by one to other deployed digital twin servers with sufficient remaining resources and the smallest interaction latency increment, in order of increasing interaction latency increment, until the resource capacity constraint is met. Latency Exceedance Adaptive Addition and Repair: When the maximum tolerable interaction latency cannot be met by any deployed digital twin server associated with a digital twin, a digital twin server is added on the smart gateway that is closest to the smart manufacturing equipment corresponding to the digital twin and has not yet deployed a digital twin server. The corresponding deployment gene is set to 1, and the associated gene of the digital twin is redirected to the added digital twin server, so that the number of deployments increases adaptively. Idle Deployment Repair: When the number of digital twins maintained by a certain deployed digital twin server is 0, the corresponding deployment gene is set to 0 to cancel the deployment, eliminate idle energy consumption, and allow the number of deployments to decrease adaptively.

[0011] In one possible implementation, the iterative solution of the multi-objective constrained optimization problem using an improved non-dominated sorting genetic algorithm includes: The parent population is merged with the offspring population after selection, crossover, mutation, and constraint repair. A fast non-dominated sort is performed on the merged population, selecting individuals to enter the next generation population in ascending order of non-dominated hierarchy. When not all individuals at a certain non-dominated hierarchy can enter the next generation population, the crowding degree of each individual at that hierarchy is calculated. ; in: , They represent according to the number After ranking the optimization objectives, and comparing them with the individual The objective function values ​​of two adjacent individuals , These represent the first in this level. The maximum and minimum values ​​of each optimization objective; Individuals are selected to fill the next generation of the population in descending order of crowding density in order to maintain the diversity of Pareto front solutions.

[0012] In one possible implementation, the step of iteratively solving the multi-objective constrained optimization problem using an improved non-dominated sorting genetic algorithm further includes: An adaptive linear ranking selection strategy is used to calculate the selection probability of individuals. After sorting the individuals in the population according to their non-dominance level and crowding degree, the ranking is determined. The probability that an individual in position 1 will be selected is: ; in: Indicates population size, , Let represent the expected selection coefficients of the best and worst individuals, respectively, and satisfy . , , Indicates the current iteration number. Indicates the maximum number of iterations. The adaptive exponent increases with the number of iterations, so that the selection pressure is small in the early stage of the iteration to maintain population diversity and increases in the later stage of the iteration to accelerate convergence. The calculated selection probability is normalized and then used for roulette wheel selection. A multi-point crossover strategy is adopted, in which the number of crossover points adaptively increases with the number of iterations. The number of intersections in the next iteration is: ; in: Indicates chromosome length; The gene is perturbed using a polynomial mutation strategy: ; ; in: , These represent the gene values ​​before and after the mutation, respectively. This represents a random number that follows a uniform distribution. Indicates the distribution index of variation. , The mutation results of the deployed gene segments are discretized into 0 or 1 according to a threshold, and the mutation results of the associated gene segments are rounded up and truncated to the nearest integer. .

[0013] In one possible implementation, determining the target deployment scheme from the Pareto optimal solution set obtained through iterative solving includes: Calculate the overall aggregate evaluation value of the individual in each iteration: ; in: Indicates the first The weighting coefficients of each optimization objective. Represents an individual The The objective function value is considered, and a higher comprehensive aggregate evaluation value indicates better overall performance of the individual; the comprehensive aggregate evaluation value should not be less than a threshold. Individuals are stored in an elite archive, the threshold. Take the first The comprehensive aggregated evaluation value of the worst individual in the next iteration population is used. After the iteration is completed, the elite archive set is merged with the final population and the Pareto optimal solution set is extracted. For each solution in the Pareto optimal solution set, calculate the weighted Euclidean distance between its normalized objective vector and the ideal point: ; The solution with the minimum weighted Euclidean distance is selected as the target deployment scheme.

[0014] Secondly, embodiments of this application provide an adaptive deployment system for a digital twin server for the Industrial Internet of Things, comprising: An architecture building module is used to build a digital twin deployment architecture that includes a physical entity layer and a digital twin layer. The physical entity layer includes multiple smart manufacturing devices and multiple smart gateways. The digital twin server uses the smart gateways as candidate deployment locations. The digital twin layer includes digital twins maintained by the digital twin server that correspond one-to-one with each smart manufacturing device, and obtains the location information, status information and resource information of each smart manufacturing device and each smart gateway. The optimization modeling module is used to determine the coverage relationship between the intelligent manufacturing equipment and the digital twin server based on the location information. It constructs a single interaction latency model for the digital twin for both cases where the equipment is within the coverage area and outside the coverage area. It also constructs a load balancing model and an energy consumption model for the digital twin server. With average interaction latency, load balancing degree and total energy consumption as optimization objectives, and with association uniqueness, deployment mutual exclusion, resource capacity and interaction latency tolerance as constraints, a multi-objective constraint optimization problem for the deployment of the digital twin server is established. The encoding module is used to jointly encode the deployment decision of the digital twin server and the association decision of the digital twin into a two-segment chromosome. The two-segment chromosome includes a deployment gene segment that characterizes whether each smart gateway deploys a digital twin server, and an association gene segment that characterizes the digital twin server associated with each digital twin, and performs initialization on the population. The iterative solution module is used to iteratively solve the multi-objective constrained optimization problem using an improved non-dominated sorting genetic algorithm. During the iteration process, constraint repair operations are performed on chromosomes that do not meet the constraints. The constraint repair operations are carried out by redirecting failure associations, migrating overloaded loads, adding digital twin servers, and removing idle digital twin servers, so that the number and location of digital twin servers are adaptively adjusted according to the constraint satisfaction status. The deployment decision module is used to determine the target deployment scheme from the Pareto optimal solution set obtained by iterative solution, determine the number of digital twin servers, deployment location, and association relationship between each digital twin and the digital twin server according to the target deployment scheme, and complete the deployment of the digital twin server.

[0015] In this embodiment, a single-interaction latency model is constructed by distinguishing between cases within and outside the coverage area. This allows the optimization process to accurately characterize the latency differences between the two differentiated transmission paths: direct upload of state data and gateway relay forwarding, thus avoiding a disconnect between the deployment scheme and actual network conditions. By jointly encoding deployment decisions and related decisions into a two-segment chromosome and combining it with four types of constraint repair operations, the number and location of digital twin servers can adaptively increase or decrease according to the constraint satisfaction situation, without the need for pre-determining the deployment quantity. Through multi-objective iterative optimization with average interaction latency, load balancing, and total energy consumption as objectives, the real-time performance, stability, and economy of digital twin services are taken into account, providing a digital twin server deployment scheme with superior overall performance for industrial IoT scenarios. Attached Figure Description

[0016] Figure 1 A schematic diagram illustrating the principle of the digital twin server deployment architecture for the Industrial Internet of Things provided in this application embodiment; Figure 2 This application provides a schematic diagram illustrating the principle of a two-stage interactive link and its delay within the coverage area, as shown in the embodiments of this application. Figure 2 Image (a) is a schematic diagram of the principle of the two-stage interactive link structure within and outside the coverage area. Figure 2 (b) is a schematic diagram illustrating the principle of single-interaction delay in two scenarios: within and outside the coverage area. Figure 3 A schematic diagram illustrating the principle of deployment-association two-segment chromosome encoding and decoding provided in an embodiment of this application; Figure 4 This is a schematic diagram comparing the initialization of the optimal point set and the population distribution of random initialization provided in an embodiment of this application. Figure 4 Image (a) is a schematic diagram of initializing the population distribution using the optimal point set. Figure 4 (b) is a schematic diagram of the random initialization of the population distribution; Figure 5 The schematic diagram illustrates the principle of four types of constraint repair operations provided in the embodiments of this application, wherein, Figure 5 (a) is a schematic diagram illustrating the principle of failure-related repair. Figure 5 (b) is a schematic diagram illustrating the principle of resource over-limit migration and repair. Figure 5 Image (c) is a schematic diagram illustrating the principle of adaptive delay over-limit repair. Figure 5(d) is a schematic diagram illustrating the principle of unloaded removal and repair; Figure 6 A schematic diagram comparing the Pareto fronts obtained by different methods provided in the embodiments of this application; Figure 7 A schematic diagram illustrating how the average interaction delay varies with the number of iterations for different methods provided in the embodiments of this application; Figure 8 A schematic diagram illustrating how the load balancing index changes with the number of iterations for different methods provided in the embodiments of this application; Figure 9 A schematic diagram illustrating the change in total energy consumption of the digital twin server as a function of the number of iterations for different methods provided in the embodiments of this application; Figure 10 This is a schematic diagram illustrating the ablation comparison of constraint repair mechanisms provided in the embodiments of this application, wherein... Figure 10 (a) is a schematic diagram showing the change of the normalized hypervolume index with the number of iterations. Figure 10 (b) is a schematic diagram showing how the proportion of feasible solutions in the population changes with the number of iterations. Detailed Implementation

[0017] The present solution will now be described in conjunction with the accompanying drawings and specific embodiments.

[0018] See Figure 1 The adaptive deployment method for digital twin servers for the Industrial Internet of Things provided in this embodiment includes: S101, construct a digital twin deployment architecture that includes a physical entity layer and a digital twin layer, and obtain the location information, status information and resource information of each intelligent manufacturing device and each intelligent gateway.

[0019] This embodiment uses a smart manufacturing plant as an example, with a floor area of ​​500 m × 500 m. The physical entity layer includes... =300 intelligent manufacturing equipment and =50 intelligent gateways. The intelligent manufacturing equipment includes CNC machine tools, industrial robots, automated guided vehicles, etc. The intelligent gateways are deployed in a grid pattern along the production line and storage area. Each intelligent gateway is connected to the others via a wired backhaul link through an industrial Ethernet network. The intelligent manufacturing equipment connects to the nearest intelligent gateway through an industrial wireless network. The digital twin server is deployed on the intelligent gateway as a candidate deployment location, that is, the digital twin server is deployed on the intelligent gateway, and each intelligent gateway can deploy at most one digital twin server. The digital twin layer includes the digital twin maintained by the digital twin server. Each intelligent manufacturing equipment corresponds to one digital twin. The digital twin continuously receives status data uploaded by the corresponding equipment to complete synchronization and sends control feedback to the equipment.

[0020] Let the set of intelligent manufacturing equipment in the physical entity layer be denoted as... Let the set of smart gateways be denoted as It will be deployed on the smart gateway. The digital twin server on the server is denoted as , represented as a quadruple ,in, This indicates the location information of the digital twin server. Indicates that the digital twin server is in Status information at any given moment Indicates that the digital twin server is in Real-time resource information, Indicates that the digital twin server is in A collection of digital twins maintained at all times. This includes intelligent manufacturing equipment. The corresponding digital twin is denoted as , represented as a quadruple ,in, This indicates the location information of the corresponding intelligent manufacturing equipment. Indicating digital twins in Real-time synchronized device status information, This indicates the resource information required for the operation of the digital twin. This represents the maximum tolerable interaction latency of the digital twin. Define the associated variable. , Indicates device The digital twin is a smart gateway Maintenance of the digital twin server.

[0021] In this embodiment, the planar coordinates of each intelligent manufacturing device and each intelligent gateway are obtained as location information through the factory positioning system; the device operating status is periodically collected as status information through the device access protocol, and the amount of status data uploaded in a single instance is recorded as follows. Obtain the computing resource capacity of each smart gateway through the resource management interface. With memory resource capacity And the computing resources required for each digital twin to run. With memory resources , as resource information.

[0022] S102, based on location information to determine coverage relationships, construct single-interaction latency model, load balancing model and energy consumption model, and establish a multi-objective constrained optimization problem for the deployment of digital twin servers.

[0023] See Figure 2 (a) Digital twin server With communication coverage radius In this embodiment Take 60m. When intelligent manufacturing equipment... With digital twin server Euclidean distance between Not greater than When the device is within coverage area, status data is directly uploaded to the digital twin server via a wireless link; when Greater than When the device is outside the coverage area, the status data is first uploaded to the smart gateway to which the device is connected via a wireless link. Then by the smart gateway Relayed to the digital twin server via wired backhaul link .

[0024] Computational intelligent manufacturing equipment With digital twin server Wireless transmission rate between: ; in: Indicates the wireless channel bandwidth. Indicating intelligent manufacturing equipment The transmission power, Indicating intelligent manufacturing equipment Channel gain between the smart gateway and the access gateway Indicates noise power.

[0025] Distinguishing between cases within and outside coverage, the uplink transmission delay is obtained: ;

[0026] in: Indicating intelligent manufacturing equipment The amount of status data uploaded in a single instance. Indicating intelligent manufacturing equipment Its connected smart gateway Wireless transmission rate between Indicates smart gateway With digital twin server The distance between the wired links, This represents the relay delay coefficient per unit distance for a unit amount of data on a wired backhaul link.

[0027] Calculate the synchronization computation latency for the digital twin server to complete one digital twin state synchronization: ;

[0028] in: This indicates the number of processing cycles required to synchronize a unit of data. Digital twin server Assigned to digital twin Computing resources.

[0029] The single-interaction delay is expressed as the sum of the uplink transmission delay and the synchronization calculation delay, and the observation duration is... The average latency of a single interaction within the time frame is calculated as follows: ; ;

[0030] The calculation process of single-interaction latency is illustrated using a specific device as an example: In this embodiment, the wireless channel bandwidth... Take 1 MHz as the device's transmit power. Take 0.5 W, channel gain Pick noise power Pick Substituting W into the formula for calculating the wireless transmission rate, we get the signal-to-noise ratio as follows: wireless transmission rate Approximately 14.61 Mbit / s. This represents the amount of status data uploaded by the device in a single upload. Taking 2.0 kbit as an example, the number of processing cycles required to synchronize a unit of data volume. Allocate computing resources at 20 cycles / bit. Using 0.6 GHz as an example, for devices within coverage area, the wireless transmission latency is approximately 0.137 ms (2.0 kbit / s divided by 14.61 Mbit / s), the synchronization calculation latency is approximately 0.067 ms (20 × 2000 cycles divided by 0.6 GHz), and the single interaction latency is approximately 0.204 ms. For devices outside coverage area, the relay latency coefficient... Pick s / (bit·m), wired link distance Taking 120 m as an example, the wired repeater delay is... ×2000×120, approximately 0.240ms, with a single interaction latency of approximately 0.444ms, still less than the device's maximum tolerable interaction latency of 0.60ms. Figure 2 Figure (b) provides a comparison of the time delays in the two scenarios described above.

[0031] Computational digital twin server exist Load and load balancing at any given time: ; ; in: , These represent digital twin servers. The computing resource capacity and memory resource capacity, Indicates allocation to digital twin memory resources, , These represent the weighting coefficients for computational resource utilization and memory resource utilization, respectively, in this embodiment. Take 0.6, Take 0.4; Indicates smart gateway Deployment variables for whether to deploy a digital twin server. Indicates the number of deployments. This represents the average load on deployed digital twin servers. Load balancing. This represents the standard deviation of the load on the deployed digital twin servers. The smaller the value, the more balanced the load. The optimization objective is to minimize the load balancing.

[0032] The operating power of the digital twin server is calculated using a power model that is linearly related to processor utilization: ; in: This represents the idle power; in this embodiment, it is taken as 8 W. This represents the full-load power; in this embodiment, it is taken as 25 W. Digital twin server exist Processor utilization at any given moment.

[0033] With the optimization objectives of minimizing average interaction latency, load balancing, and total energy consumption, a multi-objective constrained optimization problem is established: ;

[0034] in: This represents the deployment decision vector composed of deployment variables. This represents the association decision matrix composed of associated variables. Constraints Ensure that each digital twin is maintained by one and only one digital twin server; constraints Ensure that digital twins can only be associated with smart gateways that have deployed digital twin servers, and that each smart gateway can deploy at most one digital twin server; constraints Ensure that the computing and memory resource requirements of each digital twin server do not exceed its resource capacity; constrain... Ensure that the latency of a single interaction of each digital twin does not exceed its maximum tolerable interaction latency; constraints The range of values ​​for decision variables is limited. Average interaction latency characterizes the real-time performance of the digital twin service, load balancing characterizes the stability of the server cluster operation, and total energy consumption characterizes the economics of the deployment scheme. These three objectives are interdependent: the more deployments, the lower the average interaction latency, but the accumulation of idle power increases the total energy consumption; centralizing the digital twins with at least a few servers can reduce energy consumption, but it will exacerbate load imbalance and increase synchronization computation latency. Therefore, a balance needs to be achieved among the three through multi-objective optimization.

[0035] S103, jointly encodes the deployment decision of the digital twin server and the association decision of the digital twin into a two-segment chromosome, and performs initialization on the population.

[0036] See Figure 3 Chromosomes are constructed by sequentially assembling deployment gene segments and associated gene segments. Deployment gene segments contain... The first deployment gene using 0-1 encoding, the... Deployment genes Setting it to 1 indicates that in the smart gateway Deploy a digital twin server, with 0 indicating no deployment; associated gene segments include The associated genes encoded with integers, the first Related genes , Indicates device The digital twin is a smart gateway The digital twin server is maintained. Deployment variables and associated variables are configured according to... Decoding, in which This indicates the indicator function. This two-stage encoding incorporates deployment quantity, deployment location, and association relationships into the same chromosome joint optimization, including the deployment quantity. The number of genes that take the value 1 in the deployed gene segment is given directly, and it adapts to the evolutionary process without needing to be given in advance.

[0037] Figure 3 by =8、 A small example with =12 demonstrates the encoding and decoding process: the deployed gene segment is (1,0,1,0,0,1,1,0), representing the smart gateway. , , , A digital twin server is deployed, with a total of 4 servers; the associated gene segment is (1,1,3,3,1,6,6,1,3,6,3,7), representing the device. , , , Digital twins are made by Server maintenance, equipment , , , Digital twins are made by Server maintenance, equipment , , Digital twins are made by Server maintenance, equipment Digital twins are made by Server maintenance.

[0038] The initial population is generated using the optimal point set method. Constructing optimal points: ; in: This indicates taking the decimal part. The genetic dimension of a chromosome. Indicates satisfaction The smallest prime number.

[0039] The first The individual Dimensional gene mapping to value range : ; in: , They represent the first The lower and upper bounds of the value range of the dimensional gene are defined. The deployed gene segment is discretized according to a threshold; the deployed gene is set to 1 if the mapping value is not less than 0.5, otherwise it is set to 0. For associated gene segments, the following is defined: Round up to obtain the associated gene value.

[0040] Taking the two-dimensional case as an example, the effect of initializing the optimal point set is explained: when When =2, it satisfies The smallest prime number =7, the optimal point for construction is ≈0.2470 ≈0.5550. See also Figure 4 Generate 30 individuals on a unit plane. Figure 4 The point sequence obtained by initializing the optimal point set shown in (a) exhibits a uniform and non-clustered distribution in both dimensions, while Figure 4 As shown in (b), the random initialization point sequence of the same size exhibits obvious local clusters and blank areas. A uniform initial distribution allows the population to cover a larger solution space in the early stages of iteration, which helps reduce the risk of getting trapped in local optima and accelerates convergence.

[0041] S104 employs an improved non-dominated sorting genetic algorithm to iteratively solve a multi-objective constrained optimization problem, performing constraint repair operations on chromosomes that do not meet the constraints during the iteration process.

[0042] Because genetic operations such as crossover and mutation can disrupt the constraint satisfaction of chromosomes—for example, associating genes with undeployed smart gateways, some servers exceeding their resource capacity, or some digital twins experiencing interaction latency exceeding the tolerance limit—this embodiment performs a constraint repair operation immediately after each generation of offspring chromosomes to restore infeasible solutions to feasible ones. See also... Figure 5 Constraint repair operations include the following four categories: Failure Association Repair: When an associated gene points to a smart gateway that has not deployed a digital twin server, iterate through the deployed digital twin servers and redirect the associated gene to the digital twin server that meets resource capacity constraints and has the smallest incremental interaction latency. For example... Figure 5 As shown in (a), the equipment The associated genes are taken as 2 and the deployment genes are taken as 2. =0, association invalid; among candidate servers, redirect to The latency increment for interactions with the server is 0.42 ms, redirected to... The latency increment for interaction with the server is 0.31 ms and With sufficient remaining resources, the associated gene was repaired from 2 to 3.

[0043] Resource Exceedance Migration and Repair: When the computing or memory resource requirements of a digital twin server exceed its resource capacity, the digital twins maintained by that server are migrated one by one to other deployed digital twin servers with sufficient remaining resources and the smallest increase in interaction latency, in ascending order of interaction latency increment, until the resource capacity constraint is met. For example... Figure 5 As shown in (b), after repairing the failed association The computing resource requirements of the server reached 28.0 GHz, exceeding its capacity of 25.0 GHz; the candidate equipment was moved out. Migrate to The incremental latency of interaction with the server is minimal after migration. The computing resource requirement for the server has been reduced to 22.0 GHz. The server's computing resource requirement is 21.0GHz, which meets the capacity constraint; therefore, the equipment... The associated gene was repaired from 3 to 1.

[0044] Latency Exceedance Adaptive Addition and Repair: When the maximum tolerable interaction latency cannot be met by any deployed digital twin server associated with a digital twin, a new digital twin server is added on the smart gateway closest to the corresponding smart manufacturing equipment that has not yet deployed a digital twin server. The corresponding deployment gene is set to 1, and the associated gene of the digital twin is redirected to the newly added digital twin server. For example... Figure 5 As shown in (c), the device Maximum tolerable interaction latency It is 0.60 ms, and it is associated with , The single-interaction latency to the server was 0.86 ms and 0.92 ms respectively, both exceeding the tolerance limit; at the distance from the device Recent undeployed smart gateway After adding a digital twin server, the latency for a single interaction was reduced to 0.29 ms, therefore, gene sequencing will be deployed. The number of deployments is adaptively increased by changing the value from 0 to 1 and the associated gene from 6 to 7.

[0045] Unloaded Deployment and Repair: When the number of digital twins maintained by a deployed digital twin server is 0, the corresponding deployment gene will be set to 0 to revoke the deployment. For example... Figure 5 As shown in (d), After the aforementioned repairs, the server no longer maintains any digital twins and only generates idle power; therefore, the deployed genes will be deployed. The system is changed from 1 to 0, eliminating idle energy consumption and adaptively reducing the number of deployments.

[0046] After the above four types of repair operations are performed sequentially, the chromosomes simultaneously satisfy the constraints. To constraint Furthermore, the number of deployments adaptively increases or decreases depending on the constraint satisfaction. Compared to handling constraints using penalty functions, constraint repair operations ensure that the population participating in evolution always consists of feasible solutions, avoiding the waste of a large amount of search resources in infeasible regions.

[0047] The specific process of iterative solution is as follows: First, an adaptive linear ranking selection strategy is used to calculate the selection probability of individuals. After sorting individuals within the population according to their non-dominant level and crowding degree, the ranking is determined... The probability that an individual in position 1 will be selected is: ; in: Indicates population size, , Let represent the expected selection coefficients of the best and worst individuals, respectively, and satisfy . , In this embodiment Take 1.8 Take 0.2; Indicates the current iteration number. Indicates the maximum number of iterations. Adaptive exponent. As the number of iterations increases, The probability is increased to 1 to keep the selection pressure low in the early stages of iteration to maintain population diversity, and high in the later stages of iteration to accelerate convergence. The calculated selection probability is normalized and then used for roulette wheel selection.

[0048] Secondly, a multi-point crossover strategy is adopted, in which the number of crossover points adaptively increases with the number of iterations. The number of intersections in the next iteration is: ; in: This indicates chromosome length. Fewer crossover points occur in the early stages of iteration, which helps preserve superior gene combinations; more crossover points occur in the later stages of iteration, enhancing local search capabilities. Crossover operations are performed separately within the deployed gene segment and the associated gene segment to avoid cross-location between the two gene segments.

[0049] Secondly, a polynomial mutation strategy is used to perturb the gene: ; in: , These represent the gene values ​​before and after the mutation, respectively. This represents a random number that follows a uniform distribution. This represents the variation distribution index, which is set to 20 in this embodiment. , The mutation results of the deployed gene segments are discretized to 0 or 1 according to a threshold, and the mutation results of the associated gene segments are rounded up and truncated to the nearest integer. This allows the mutation operator to be adapted to two-part coding.

[0050] Then, the aforementioned four types of constraint repair operations are performed on the offspring chromosomes generated through selection, crossover, and mutation. The parent population is merged with the repaired offspring population, and a fast non-dominated sort is performed on the merged population. Individuals are selected to enter the next generation population in order of increasing non-dominated hierarchy. When not all individuals at a certain non-dominated hierarchy can enter the next generation population, the crowding degree of each individual at that hierarchy is calculated. ; in: , They represent according to the number After ranking the optimization objectives, and comparing them with the individual The objective function values ​​of two adjacent individuals , These represent the first in this level. The maximum and minimum values ​​of the optimization objective are determined. Individuals are selected to fill the next generation of the population in descending order of crowding to maintain the diversity of Pareto front solutions.

[0051] Finally, in each iteration, the overall aggregated evaluation value of each individual is calculated and the elite archive set is maintained: ; in: Indicates the first The weight coefficients for each optimization objective are set; in this embodiment, the weight of each objective is 1 / 3. A higher overall aggregate evaluation value indicates better overall performance. The overall aggregate evaluation value should not be less than a threshold. Individuals are stored in the elite archive, threshold Take the first The overall aggregated evaluation value of the worst individual in the next iteration of the population. The elite archive avoids the loss of high-quality compromise solutions due to random operations during evolution. Repeat the above process until the maximum number of iterations is reached.

[0052] S105: Determine the target deployment scheme from the Pareto optimal solution set obtained by iterative solution, and complete the deployment of the digital twin server.

[0053] After iteration, the elite archive is merged with the final population, and the Pareto optimal solution set is extracted. For each solution in the Pareto optimal solution set, the weighted Euclidean distance between its normalized objective vector and the ideal point is calculated: ; The solution with the minimum weighted Euclidean distance is selected as the target deployment scheme, which achieves the optimal trade-off among the three objectives of average interaction latency, load balancing, and total energy consumption. According to the deployment gene segments of the target deployment scheme, digital twin servers are instantiated on the corresponding smart gateways. The association relationships between each digital twin and the digital twin server are established according to the associated gene segments, and the twin's operational data is synchronized, thus completing the deployment.

[0054] During the operation phase, when the number of intelligent manufacturing equipment increases or decreases, the spatial distribution of equipment is adjusted, or the proportion of digital twins whose interaction latency exceeds the tolerance limit is detected to exceed the preset threshold, the current deployment scheme is used as the elite individuals in the initial population to re-execute the iterative solution, thereby realizing the periodic adaptive update of the deployment scheme.

[0055] To further illustrate the feasibility of this solution, the key parameter settings and experimental verification results of this embodiment are given below.

[0056] In this embodiment, the key parameters of the simulation scenario and algorithm are set as follows: the factory area is 500 m × 500 m; the number of smart gateways... Take 50, the number of intelligent manufacturing equipment Take 300; Computing resource capacity of each smart gateway. Values ​​are taken within the range of 20GHz to 30GHz, representing memory resource capacity. Values ​​range from 16 GB to 32 GB; the communication coverage radius of the digital twin server. Take 60 m; the maximum tolerable interaction delay of the digital twin. Values ​​are taken within 0.4 ms to 0.8 ms; the amount of status data uploaded by the device in a single upload. Values ​​range from 1.0 kbit to 3.0 kbit; computational resources allocated to a single digital twin. Values ​​are taken within the range of 0.4 GHz to 0.8 GHz, memory resources Values ​​range from 0.2 GB to 0.5 GB; wireless channel bandwidth. Take 1 MHz, noise power Pick W, relay delay coefficient Pick s / (bit·m); Idle power Take 8 W, full load power Take 25 W, energy consumption observation time Take 1 second; population size Set the number to 30, which represents the maximum number of iterations. The crossover probability is set to 0.9, and the mutation probability is set to 0.1. The comparison method selects three classic multi-objective evolutionary algorithms: NSGA-II, NSGA-III, and MOEA / D. The comparison method uses the same two-stage encoding as in this embodiment, and constraints are handled using penalty functions.

[0057] See Figure 6 Under the same number of iterations, the Pareto front obtained by the method in this embodiment is located to the lower left of the front of the three comparative methods. That is, it is superior in both average interaction delay and total energy consumption. Moreover, the front solution is evenly distributed and has a complete coverage, indicating that the initialization of the optimal point set and the adaptive genetic operator effectively improve the convergence and diversity of the solution set.

[0058] See Figure 7The average interaction latency of the method in this embodiment converges to around 0.25 ms after approximately 200 iterations, with a final convergence value of approximately 0.248 ms. The final convergence values ​​of NSGA-II, NSGA-III, and MOEA / D are approximately 0.462 ms, 0.401 ms, and 0.352 ms, respectively. Compared to NSGA-II, the average interaction latency of the method in this embodiment is reduced by approximately 46.3%; compared to NSGA-III and MOEA / D, it is reduced by approximately 38.2% and 29.5%, respectively. The latency advantage mainly comes from two aspects: First, the differentiated latency modeling inside and outside the coverage area allows the optimization process to accurately perceive the additional cost of relay forwarding, tending to associate twins with directly connectable nearby servers; second, the adaptive addition and repair of latency exceeding limits ensures that the interaction latency of each twin is within the tolerable upper limit.

[0059] See Figure 8 In this embodiment, the load balancing index of the method continuously decreases with iteration and converges to about 0.052. The final values ​​of NSGA-II, NSGA-III and MOEA / D are about 0.118, 0.131 and 0.169, respectively. The method of this embodiment reduces the load by about 55.9% compared with NSGA-II, indicating that the resource overload migration repair and multi-objective collaborative optimization make the load distribution of each digital twin server more balanced, avoiding the overload operation of some servers.

[0060] See Figure 9 The total energy consumption of the method in this embodiment fluctuated significantly in the early stages of iteration, then rapidly decreased and converged to approximately 167.0 J. This corresponds to the final deployment scheme, where 9 digital twin servers were adaptively selected from 50 candidate smart gateways, with an average processor utilization of approximately 0.62. NSGA-II ultimately converged to approximately 173.4 J, corresponding to the deployment of 10 servers with an average utilization of approximately 0.55. NSGA-III and MOEA / D converged to approximately 178.2 J and 170.5 J, respectively. The method in this embodiment reduces total energy consumption by approximately 3.7% compared to NSGA-II. Its energy efficiency advantage stems from the elimination of ineffective power from idle servers through idle removal and repair, and a better trade-off between the number of deployments and utilization.

[0061] See Figure 10 To verify the effectiveness of the constraint repair operation, the method in this embodiment is compared with the ablation version that removes constraint repair and uses penalty functions to handle constraints. Figure 10 Figure (a) shows that with constraint repair, the normalized hypervolume index reaches 0.85 in about 260 iterations and eventually converges to about 0.923; without constraint repair, it takes about 935 iterations to reach 0.85 and eventually converges to about 0.851. That is, constraint repair reduces the number of iterations required to achieve the same solution set quality by about 72% and ultimately improves the hypervolume by about 8.5%. Figure 10 Figure (b) shows that with constraint repair, the proportion of feasible solutions in the population remains constant at 100%, while the proportion of feasible solutions in the penalty function version slowly increases from the initial 31% to 86%, indicating that a large amount of search resources are consumed in the infeasible region. The above ablation results verify the contribution of the four types of constraint repair operations to the convergence speed and solution set quality.

[0062] Based on the above experimental results, the method in this embodiment outperforms the comparative methods in terms of average interaction latency, load balancing, and total energy consumption. Furthermore, the deployment quantity does not require manual pre-setting and can be adaptively determined according to the constraint satisfaction situation, verifying the effectiveness and feasibility of this solution in the deployment scenario of digital twin servers for industrial IoT.

[0063] Corresponding to the adaptive deployment method of digital twin servers for industrial IoT provided in the above embodiments, this application also provides an embodiment of an adaptive deployment system of digital twin servers for industrial IoT.

[0064] The adaptive deployment system for digital twin servers for the Industrial Internet of Things in this embodiment includes: An architecture building module is used to build a digital twin deployment architecture that includes a physical entity layer and a digital twin layer. The physical entity layer includes multiple smart manufacturing devices and multiple smart gateways. The digital twin server uses the smart gateways as candidate deployment locations. The digital twin layer includes digital twins maintained by the digital twin server that correspond one-to-one with each smart manufacturing device, and obtains the location information, status information and resource information of each smart manufacturing device and each smart gateway. The optimization modeling module is used to determine the coverage relationship between the intelligent manufacturing equipment and the digital twin server based on the location information. It constructs a single interaction latency model for the digital twin for both cases where the equipment is within the coverage area and outside the coverage area. It also constructs a load balancing model and an energy consumption model for the digital twin server. With average interaction latency, load balancing degree and total energy consumption as optimization objectives, and with association uniqueness, deployment mutual exclusion, resource capacity and interaction latency tolerance as constraints, a multi-objective constraint optimization problem for the deployment of the digital twin server is established. The encoding module is used to jointly encode the deployment decision of the digital twin server and the association decision of the digital twin into a two-segment chromosome. The two-segment chromosome includes a deployment gene segment that characterizes whether each smart gateway deploys a digital twin server, and an association gene segment that characterizes the digital twin server associated with each digital twin, and performs initialization on the population. The iterative solution module is used to iteratively solve the multi-objective constrained optimization problem using an improved non-dominated sorting genetic algorithm. During the iteration process, constraint repair operations are performed on chromosomes that do not meet the constraints. The constraint repair operations are carried out by redirecting failure associations, migrating overloaded loads, adding digital twin servers, and removing idle digital twin servers, so that the number and location of digital twin servers are adaptively adjusted according to the constraint satisfaction status. The deployment decision module is used to determine the target deployment scheme from the Pareto optimal solution set obtained by iterative solution, determine the number of digital twin servers, deployment location, and association relationship between each digital twin and the digital twin server according to the target deployment scheme, and complete the deployment of the digital twin server.

[0065] In this application embodiment, "at least one" refers to one or more, and "more than one" refers to two or more. "And / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent three cases: A alone, A and B simultaneously, and B alone, where A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. "At least one of the following" or similar expressions refer to any combination of these items, including any combination of single or plural items.

[0066] The above description is merely a specific embodiment of this application. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the protection scope of this application. The protection scope of this application shall be determined by the scope of the claims.

Claims

1. An adaptive deployment method for digital twin servers for the Industrial Internet of Things, characterized in that, include: A digital twin deployment architecture is constructed, comprising a physical entity layer and a digital twin layer. The physical entity layer includes multiple smart manufacturing devices and multiple smart gateways. The digital twin server uses the smart gateways as candidate deployment locations. The digital twin layer includes digital twins maintained by the digital twin server that correspond one-to-one with each smart manufacturing device, and obtains the location information, status information, and resource information of each smart manufacturing device and each smart gateway. Based on the location information, the coverage relationship between the intelligent manufacturing equipment and the digital twin server is determined. For the two cases of being within the coverage area and being outside the coverage area, a single interaction latency model of the digital twin is constructed. A load balancing model and an energy consumption model of the digital twin server are also constructed. With average interaction latency, load balancing degree and total energy consumption as optimization objectives, and with association uniqueness, deployment mutual exclusion, resource capacity and interaction latency tolerance as constraints, a multi-objective constraint optimization problem for the deployment of the digital twin server is established. The deployment decision of the digital twin server and the association decision of the digital twin are jointly encoded into a two-segment chromosome. The two-segment chromosome includes a deployment gene segment that characterizes whether each smart gateway deploys a digital twin server, and an association gene segment that characterizes the digital twin server associated with each digital twin. The population is then initialized. An improved non-dominated sorting genetic algorithm is used to iteratively solve the multi-objective constrained optimization problem. During the iteration process, constraint repair operations are performed on chromosomes that do not meet the constraints. The constraint repair operations are carried out by redirecting failure associations, migrating overloaded loads, adding digital twin servers, and removing idle digital twin servers, so that the number and location of digital twin servers are adaptively adjusted according to the constraint satisfaction status. The target deployment scheme is determined from the Pareto optimal solution set obtained by iterative solution. The number of digital twin servers, their deployment locations, and the association between each digital twin and the digital twin server are determined according to the target deployment scheme. The deployment of the digital twin servers is then completed.

2. The adaptive deployment method for digital twin servers for the Industrial Internet of Things according to claim 1, characterized in that, The constructed digital twin deployment architecture, comprising a physical entity layer and a digital twin layer, includes: Let the set of intelligent manufacturing equipment in the physical entity layer be denoted as... Let the set of smart gateways be denoted as ,in, This indicates the number of intelligent manufacturing equipment. Indicates the number of smart gateways; Will be deployed on smart gateways The digital twin server on the server is denoted as Digital twin server Represented as a quadruple ,in, This indicates the location information of the digital twin server. Indicates that the digital twin server is in Status information at any given moment Indicates that the digital twin server is in Real-time resource information, Indicates that the digital twin server is in The set of digital twins maintained at any given moment; Intelligent manufacturing equipment The corresponding digital twin is denoted as Digital twin Represented as a quadruple ,in, This indicates the location information of the corresponding intelligent manufacturing equipment. Indicating digital twins in Real-time synchronized device status information, This indicates the resource information required for the operation of the digital twin. This represents the maximum tolerable interaction latency of the digital twin; Define related variables , Indicating intelligent manufacturing equipment Digital twins By smart gateway Maintenance of the digital twin server on the platform. This indicates that there is no maintenance relationship.

3. The adaptive deployment method for digital twin servers for the Industrial Internet of Things according to claim 2, characterized in that, The method for constructing single-interaction latency models for digital twins in two scenarios—one within the coverage area and the other outside the coverage area—includes: Computational intelligent manufacturing equipment With digital twin server Wireless transmission rate between: ； in: Indicates wireless channel bandwidth. Indicating intelligent manufacturing equipment The transmission power, Indicating intelligent manufacturing equipment Channel gain between the smart gateway and the access gateway Indicates noise power; When intelligent manufacturing equipment With digital twin server Distance between No larger than the communication coverage radius of the digital twin server When the intelligent manufacturing equipment is within coverage area, its status data is directly uploaded to the digital twin server via a wireless link; when the distance... Greater than the communication coverage radius When it is determined that the intelligent manufacturing equipment is outside the coverage area, the status data is first uploaded to the intelligent gateway to which the equipment is connected via a wireless link. Then by the smart gateway Relayed to the digital twin server via wired backhaul link The uplink transmission delay is obtained as follows: ； in: Indicating intelligent manufacturing equipment The amount of status data uploaded in a single session. Indicating intelligent manufacturing equipment Its connected smart gateway Wireless transmission rate between Indicates smart gateway With digital twin server The distance between the wired links, Represents the relay delay coefficient per unit distance of data volume on a wired backhaul link; Calculate the synchronization computation latency for the digital twin server to complete one digital twin state synchronization: ； in: This indicates the number of processing cycles required to synchronize a unit of data. Digital twin server Assigned to digital twin Computing resources; The latency of a single interaction is expressed as the sum of the uplink transmission latency and the synchronization calculation latency: ； For observation duration The time-averaged interaction latency is obtained by taking the average latency of each single interaction within the time frame. 。 4. The adaptive deployment method for digital twin servers for the Industrial Internet of Things according to claim 3, characterized in that, The load balancing model and energy consumption model for constructing the digital twin server include: Computational digital twin server exist Load at any moment: ； in: Digital twin server The computing resource capacity, Digital twin server Memory resource capacity, Digital twin server Assigned to digital twin memory resources, , These represent the weighting coefficients for computing resource utilization and memory resource utilization, respectively. Load balancing is measured by the standard deviation of the load on deployed digital twin servers: ； in: Indicates smart gateway Deployment variables for whether to deploy a digital twin server. This indicates the number of digital twin servers deployed. This indicates the average load and load balancing of the deployed digital twin servers. The smaller the value, the more balanced the load is among the digital twin servers; The power model is used to calculate the digital twin server. exist Operating power at any given time: ； in: This indicates the idle power of the digital twin server. This indicates the full-load power of the digital twin server. Digital twin server exist Processor utilization at any given moment.

5. The adaptive deployment method for digital twin servers for the Industrial Internet of Things according to claim 4, characterized in that, The multi-objective constrained optimization problem for establishing a digital twin server deployment includes: With the optimization objectives of minimizing average interaction latency, load balancing, and total energy consumption, the following multi-objective constrained optimization problem is established: ； in: Indicates by deployment variables The deployment decision vector constituted Indicates the association variables The constructed association decision matrix; constraints This means that each digital twin is maintained by one and only one digital twin server, which is a constraint. This means that a digital twin can only be associated with a smart gateway that has a deployed digital twin server, and each smart gateway can deploy at most one digital twin server, constraining... This indicates that the computing and memory resource requirements of each digital twin server do not exceed its resource capacity, constraining... This indicates that the latency of a single interaction of each digital twin does not exceed its maximum tolerable interaction latency, constraining... This indicates the range of values ​​for deployment variables and associated variables.

6. The adaptive deployment method for digital twin servers for the Industrial Internet of Things according to claim 5, characterized in that, The step of jointly encoding the deployment decision of the digital twin server and the association decision of the digital twin into a two-segment chromosome and performing population initialization includes: The chromosome is divided into deployment gene segments and associated gene segments, wherein the deployment gene segments include The first deployment gene using 0-1 encoding, the... A deployment gene of 1 indicates that in the smart gateway A digital twin server is deployed on top of the system, with 0 indicating no deployment; the associated gene segment contains The associated genes encoded with integers, the first Related genes , Indicating intelligent manufacturing equipment The digital twin is a smart gateway The digital twin server is maintained, and the deployment variables and associated variables are arranged according to... Decoding, in which Indicates an indicator function; The optimal point set method is used to generate an initial population within the gene value space, constructing optimal points: ； in: This indicates taking the decimal part. The genetic dimension of a chromosome. Indicates satisfaction The smallest prime number; The first The individual Dimensional gene mapping to value range : ； in: , They represent the first The lower and upper bounds of the value range of the dimensional gene; Deployed gene segments are discretized according to a threshold; if the mapping value is not less than 0.5, the deployed gene is set to 1; otherwise, it is set to 0. Associated gene segments are then discretized according to... Round up to obtain the associated gene value.

7. The adaptive deployment method for digital twin servers for the Industrial Internet of Things according to claim 6, characterized in that, The constraint repair operation performed on chromosomes that do not meet the constraints during the iteration process includes: Failure association repair: When the associated gene points to a smart gateway that has not deployed a digital twin server, traverse the deployed digital twin servers and redirect the associated gene to the digital twin server that meets the resource capacity constraints and has the smallest interaction latency increment. Resource Exceedance Migration and Repair: When the computing or memory resource requirements of a digital twin server exceed its resource capacity, the digital twins maintained by the digital twin server are migrated one by one to other deployed digital twin servers with sufficient remaining resources and the smallest interaction latency increment, in order of increasing interaction latency increment, until the resource capacity constraint is met. Latency Exceedance Adaptive Addition and Repair: When the maximum tolerable interaction latency cannot be met by any deployed digital twin server associated with a digital twin, a digital twin server is added on the smart gateway that is closest to the smart manufacturing equipment corresponding to the digital twin and has not yet deployed a digital twin server. The corresponding deployment gene is set to 1, and the associated gene of the digital twin is redirected to the added digital twin server, so that the number of deployments increases adaptively. Idle Deployment Repair: When the number of digital twins maintained by a certain deployed digital twin server is 0, the corresponding deployment gene is set to 0 to cancel the deployment, eliminate idle energy consumption, and allow the number of deployments to decrease adaptively.

8. The adaptive deployment method for digital twin servers for the Industrial Internet of Things according to claim 7, characterized in that, The iterative solution of the multi-objective constrained optimization problem using the improved non-dominated sorting genetic algorithm includes: The parent population is merged with the offspring population after selection, crossover, mutation, and constraint repair. A fast non-dominated sort is performed on the merged population, selecting individuals to enter the next generation population in ascending order of non-dominated hierarchy. When not all individuals at a certain non-dominated hierarchy can enter the next generation population, the crowding degree of each individual at that hierarchy is calculated. ； in: , They represent according to the number After ranking the optimization objectives, and comparing them with the individual The objective function values ​​of two adjacent individuals , These represent the first in this level. The maximum and minimum values ​​of each optimization objective; Individuals are selected to fill the next generation of the population in descending order of crowding density in order to maintain the diversity of Pareto front solutions. An adaptive linear ranking selection strategy is used to calculate the selection probability of individuals. After sorting the individuals in the population according to their non-dominance level and crowding degree, the ranking is determined. The probability that an individual in position 1 will be selected is: ； in: Indicates population size, , Let represent the expected selection coefficients of the best and worst individuals, respectively, and satisfy ... , , Indicates the current iteration number. Indicates the maximum number of iterations. The adaptive exponent increases with the number of iterations, so that the selection pressure is small in the early stage of the iteration to maintain population diversity and increases in the later stage of the iteration to accelerate convergence. The calculated selection probability is normalized and then used for roulette wheel selection. A multi-point crossover strategy is adopted, in which the number of crossover points adaptively increases with the number of iterations. The number of intersections in the next iteration is: ； in: Indicates chromosome length; The gene is perturbed using a polynomial mutation strategy: ； ； in: , These represent the gene values ​​before and after the mutation, respectively. This represents a random number that follows a uniform distribution. Indicates the distribution index of variation. , The mutation results of the deployed gene segments are discretized into 0 or 1 according to a threshold, and the mutation results of the associated gene segments are rounded up and truncated to the nearest integer. .

9. The adaptive deployment method for digital twin servers for the Industrial Internet of Things according to claim 8, characterized in that, Determining the target deployment scheme from the Pareto optimal solution set obtained through iterative solving includes: Calculate the overall aggregate evaluation value of the individual in each iteration: ； in: Indicates the first The weighting coefficients of each optimization objective. Represents an individual The The objective function value is considered, and a higher comprehensive aggregate evaluation value indicates better overall performance of the individual; the comprehensive aggregate evaluation value should not be less than a threshold. Individuals are stored in an elite archive, the threshold. Take the first The comprehensive aggregated evaluation value of the worst individual in the next iteration population is used. After the iteration is completed, the elite archive set is merged with the final population and the Pareto optimal solution set is extracted. For each solution in the Pareto optimal solution set, calculate the weighted Euclidean distance between its normalized objective vector and the ideal point: ； The solution with the minimum weighted Euclidean distance is selected as the target deployment scheme.

10. An adaptive deployment system for digital twin servers for the Industrial Internet of Things, characterized in that, include: An architecture building module is used to build a digital twin deployment architecture that includes a physical entity layer and a digital twin layer. The physical entity layer includes multiple smart manufacturing devices and multiple smart gateways. The digital twin server uses the smart gateways as candidate deployment locations. The digital twin layer includes digital twins maintained by the digital twin server that correspond one-to-one with each smart manufacturing device, and obtains the location information, status information and resource information of each smart manufacturing device and each smart gateway. The optimization modeling module is used to determine the coverage relationship between the intelligent manufacturing equipment and the digital twin server based on the location information. It constructs a single interaction latency model for the digital twin for both cases where the equipment is within the coverage area and outside the coverage area. It also constructs a load balancing model and an energy consumption model for the digital twin server. With average interaction latency, load balancing degree and total energy consumption as optimization objectives, and with association uniqueness, deployment mutual exclusion, resource capacity and interaction latency tolerance as constraints, a multi-objective constraint optimization problem for the deployment of the digital twin server is established. The encoding module is used to jointly encode the deployment decision of the digital twin server and the association decision of the digital twin into a two-segment chromosome. The two-segment chromosome includes a deployment gene segment that characterizes whether each smart gateway deploys a digital twin server, and an association gene segment that characterizes the digital twin server associated with each digital twin, and performs initialization on the population. The iterative solution module is used to iteratively solve the multi-objective constrained optimization problem using an improved non-dominated sorting genetic algorithm. During the iteration process, constraint repair operations are performed on chromosomes that do not meet the constraints. The constraint repair operations are carried out by redirecting failure associations, migrating overloaded loads, adding digital twin servers, and removing idle digital twin servers, so that the number and location of digital twin servers are adaptively adjusted according to the constraint satisfaction status. The deployment decision module is used to determine the target deployment scheme from the Pareto optimal solution set obtained by iterative solution, determine the number of digital twin servers, deployment location, and association relationship between each digital twin and the digital twin server according to the target deployment scheme, and complete the deployment of the digital twin server.

Images