Component deployment method for industrial Internet of Things application

The MADPG algorithm addresses the challenge of deploying IIoT components in heterogeneous cloud-edge environments by combining GA and PSO to minimize costs and meet performance constraints, enhancing the efficiency and privacy of industrial IoT systems.

CN119071162AInactive Publication Date: 2024-12-03FUZHOU UNIV
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202411068179.4
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-06
Publication Date
2024-12-03
Estimated Expiration
Not applicable · inactive patent

AI Technical Summary

Technical Problem

The challenge of efficiently deploying industrial Internet of Things (IIoT) components in a cloud-edge collaborative environment is exacerbated by the heterogeneity of data centers, limited computing resources, and the need for real-time data processing, particularly for privacy-sensitive data that cannot be offloaded to the cloud.

Method used

A component deployment method using a multi-application deployment algorithm (MADPG) that combines Genetic Algorithm (GA) and Particle Swarm Optimization (PSO) to minimize deployment costs while meeting constraints of application response time, privacy protection, and limited computing power in a cloud-edge environment.

Benefits of technology

MADPG effectively reduces deployment costs while ensuring performance constraints, such as response time and data center capacity, by optimizing the deployment of IIoT components in industrial environments.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119071162A_ABST
    Figure CN119071162A_ABST
Patent Text Reader

Abstract

The invention relates to a component deployment method for industrial Internet of Things application, and belongs to the field of industrial Internet of Things application component deployment. According to the method, the characteristics of the industrial Internet of Things application are fully considered, and a deployment model for the industrial Internet of Things application in the cloud environment with the deployment cost minimized under the constraint of the response time of the application, privacy protection and the limited computing power of a data center is established. On the basis, a multi-application deployment algorithm (MADPG) (Multi-Application Deployment On GA and PSO) which combines GA (Generation Algorithm) and PSO (Particle Swarm Optimization) is provided, and on the premise that application constraints of the industrial Internet of Things are met as much as possible, the application deployment cost is minimized.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the field of industrial Internet of Things application component deployment, and relates to a component deployment method for industrial Internet of Things applications. Background Art

[0002] The deployment and use of wireless sensor networks and actor networks in industry make it possible to optimize production lines with better quality management, energy efficiency, fault prediction, product planning, and resource forecasting. When data is collected from different sensors, actuators, and machines in an industrial environment and the devices that control and process the data are accessed through the Internet, this situation is called the Industrial Internet of Things (IIoT). It not only provides more reliable, real-time, and secure communications, but also enables the concept of smart factories in the fourth industrial revolution. As a promising paradigm, IIoT is becoming an emerging technology for realizing Industry 4.0 and smart manufacturing, and is seen as a key technology for the transformation of the manufacturing industry.

[0003] IIoT uses mobile communications, intelligent analysis and other technologies to collect data from industrial production processes through various sensors or controllers with perception and monitoring capabilities for processing and analysis, thereby improving the efficiency and quality of production products while reducing product costs and resource consumption. Rapid technological advances have enabled more and more sensors and smart devices to be connected, generating a large amount of data. However, sensors and machines "continuously" generate different types of data. This data may contain sensitive information. It may also be time-sensitive. Therefore, many applications need to process a large amount of data from these distributed terminal devices. These applications are composed of many components, and these components work together to complete the analysis and processing of data. Some data is privacy-sensitive, and some data needs to be analyzed and processed in real time to ensure the performance of the industrial Internet of Things system. However, due to the limitations of the computing power, storage capacity and other resources of the smart devices themselves, for the large amount of data collected from the terminal devices, they themselves usually cannot provide enough computing resources to ensure the performance of the industrial Internet of Things system.

[0004] Fortunately, the cloud-edge collaborative computing paradigm can serve as an effective solution to the above problems. As a new computing model that combines the advantages of cloud computing and edge computing, the cloud-edge collaborative computing model utilizes the high-performance storage and computing capabilities of cloud data centers and the advantage of edge data centers being closer to users to deploy computing-intensive components in the cloud and data-sensitive components at the edge to meet the performance requirements of industrial IoT systems. However, due to the large heterogeneity of the cloud-edge collaborative environment, different data centers vary in computing power, bandwidth, and storage. At the same time, considering the data dependencies between each application component. Therefore, how to achieve reasonable and efficient deployment of application components in such an environment remains a major challenging problem.

[0005] At present, some work has begun to study the deployment of applications in cloud-edge collaborative environments. However, these works cannot be directly applied to the industrial Internet of Things environment. First, sensitive data may be generated in industrial scenarios, and these data cannot be offloaded to the cloud data center for processing, but can only be offloaded to the edge data center. Therefore, the components responsible for processing these data can only be deployed in the edge data center; secondly, in the industrial Internet of Things environment, data is continuously generated, so the components that process these data are required to be deployed in the data center for a long time in order to process and analyze the corresponding data in real time to ensure the performance of the industrial Internet of Things system. Summary of the invention

[0006] The purpose of the present invention is to provide a component deployment method for industrial Internet of Things applications. The method fully considers the characteristics of industrial Internet of Things applications and establishes a deployment model for industrial Internet of Things applications in a cloud-edge environment that minimizes deployment costs under the constraints of application response time, privacy protection, and limited computing power of data centers. On this basis, a method is proposed to combine GA (Geneti c The multi-application deployment algorithm MADPG (Multi Application Deployment Algorithm based on GA and PSO) combines GA and PSO (Particle Swarm Optimization) to minimize the application deployment cost while satisfying the application constraints of the industrial Internet of Things as much as possible.

[0007] To achieve the above objectives, the technical solution of the present invention is: a component deployment method for industrial Internet of Things applications, which establishes a deployment model for industrial Internet of Things applications in a cloud-edge environment that minimizes deployment costs under the constraints of application response time, privacy protection, and limited computing power of data centers; on this basis, a multi-application deployment algorithm MADPG that combines GA and PSO is proposed to minimize the application deployment cost while meeting the constraints of industrial Internet of Things applications.

[0008] In one embodiment of the present invention, the following steps are included:

[0009] S1. Build system model: A component deployment system for industrial IoT applications in a cloud-edge collaborative environment, which includes three parts: a multi-application model with constraints provided by multiple industrial IoT terminals, a cloud-edge collaborative environment, and a meta-heuristic deployer;

[0010] S2. Industrial Internet of Things Application Deployment: A multi-application deployment algorithm MADPG that combines GA and PSO is proposed to minimize the application deployment cost while satisfying the industrial Internet of Things application constraints.

[0011] In one embodiment of the present invention, in step S1, the multi-application model is specifically as follows:

[0012] A set of applications is represented as a set A = {a1, a2, a3, ..., a n}; Apply a i Represented as a two-tuple (l i , G i ), l i represents the maximum allowed response delay of the i-th application; G i Represents the data dependency structure of the i-th application;

[0013] The complex dependency structure between each application component will be represented as a directed acyclic graph G i =(V i , E i ), where V i is a set of m components {v i1 , v i2 , ..., v im},v ij represents the jth component in the i-th application, component v ij The required computing resources are denoted as p ij , the amount of computation for each component is expressed as τ ij ,s ij ∈{0, 1} indicates whether the component processes sensitive data. If s ij =1 means that components processing sensitive data can only be placed in edge data centers; Represents the data edge set between components. For an edge Represents component v i,p and component v i,q There is an association between each component v i,p To component v i,q The amount of data transmitted is At the same time, component v i,p is the component v i,q The predecessor of component v i,q is the component v i,p A component can only start executing after all its predecessor components have completed execution and received their transmitted data during application execution.

[0014] In one embodiment of the present invention, in step S1, the cloud-edge collaborative environment is specifically as follows:

[0015] Cloud-edge collaborative environment D = {D cloud , D edge}It consists of cloud data center and edge data center; Cloud D cloud ={d1, d2, ..., d u} represents the set of u cloud data centers, edge D edge ={d u+1 , d u+2 , ..., d u+v} indicates that there are v edge data centers near the factory; data center d k It is expressed as:

[0016]

[0017] where r k Indicates data center d k The computing resources owned by k Indicates data center d k The processing capacity of a unit of computing resources; Indicates that in the data center d k The price of renting a computing unit; f k ∈{0, 1} represents the type of data center. When f k =0, it belongs to the cloud data center, which has more computing resources and stronger unit computing power; when f k =1, d k The edge data center has general computing resources and computing power. It is assumed that the number of computing resources and the computing power per unit resource of the data center are known.

[0018] In one embodiment of the present invention, in step S1, the meta-heuristic deployer is specifically as follows:

[0019] The goal of the meta-heuristic deployer is to minimize the deployment cost of multiple applications given a deployment plan that satisfies the maximum allowable response delay, data protection, and does not exceed the computing capacity constraints of the data center for each industrial IoT application; the deployment strategy is defined as follows:

[0020] Γ=(A,S,M,c t , T f )

[0021] Where M is a function: V→S represents the mapping of component V to cloud-edge collaborative environment S, such as M(v ij )=s k Indicates that the component v ij Deploy to data centers k In, c t represents the deployment cost of multiple industrial IoT applications A in the cloud-edge collaborative environment S, represents the response time of each application under the deployment strategy Γ; in view of the characteristics of industrial IoT applications, the application components are deployed in the data center for a long time, and the deployment cost is optimized under the constraints of application response time, component privacy protection and data center capacity limitation; to calculate the response time of the application, the execution time is calculated according to the computing amount of each component and the computing power of the rented data center, and the response time is calculated according to the transmission data size and bandwidth size; the calculation method of the component computing time and data transmission time in one response is as follows:

[0022]

[0023]

[0024] In a request response component v ij In the data center k The computation time of component v is expressed as formula (1), ij On server d k The computing power it has is p ij o k ; Formula (2) represents the component v p With component v q The resulting transmission time; if the two components are deployed in the same data center, their data transmission time is ignored;

[0025] For a deployment strategy Γ, when the mapping function M is determined, the deployment of components of multiple industrial IoT applications A in the cloud edge environment D is also determined; according to the characteristics of industrial IoT applications, it is necessary to calculate the deployment cost c of multiple industrial IoT applications t and response time T f And each data center k Computing resources used

[0026] According to the mapping function M of multiple industrial IoT applications A to the cloud-edge collaborative environment D, the relevant variables are defined as follows:

[0027] (1)t start (v i,j , d k ): Component v in a request response i,j In the data center k The start time of the i,j It is determined by the completion time of all the predecessor components, as shown in formula (3):

[0028]

[0029] (2)t end (v i,j , dk ): Component v in a request response i,j In the data center k The end time is equal to the component v i,j The start time of the data center d k The sum of the above calculation time is shown in formula (4):

[0030] t end (v ij , d k ) = t start (v ij , d k )+t tc (v ij , d k ) (4)

[0031] (3)cost(v ij , d k ): component v ij In the data center k The deployment cost is equal to the required computing resources and data center d k The product of the calculation unit price is shown in formula (5):

[0032]

[0033] (4)cost(a i ): Apply a i The deployment cost is the sum of the deployment costs of each component in the data center, as shown in (6):

[0034]

[0035] (5) Data Center k The computing resources that have been used are all the computing resources deployed on server d k The sum of the computing resources required by the components on is shown in formula (7):

[0036]

[0037] Based on the above definition, we get the execution cost c of multiple industrial IoT applications: t , one response time As shown in formula (8) and formula (9) respectively:

[0038]

[0039]

[0040] The execution cost c tThe deployment cost of all application components and the application response time The latest completion time of the components in the application;

[0041] For the deployment strategy of multiple industrial IoT applications, its goal is to minimize its deployment cost under the premise of satisfying the maximum allowed response delay of the application, the component data processing protection principle and the data center capacity limit; the problem of multiple application deployment in the industrial IoT environment is formalized as formula (10):

[0042]

[0043] Constraint 1 means that the application response time cannot exceed the maximum allowed response delay of the application, and constraint 2 means that if s ij =1, that is, the component processes private data and must be deployed in the edge data center. Constraint 3 indicates that the data center d k The number of occupied resources cannot exceed the computing resources available in its data center.

[0044] In one embodiment of the present invention, step S2 includes:

[0045] S21. Question code:

[0046] A coding method for mapping application components and data centers is proposed; a potential deployment scheme for multiple industrial IoT applications in a cloud-edge collaborative environment corresponds to a particle in the particle swarm; in the Gth iteration, the kth particle in the particle swarm is represented as As shown in formula (13):

[0047]

[0048] Where NP represents the size of the particle population, represents the jth task v of the i-th industrial IoT application ij The location of the deployed data center; for the privacy-preserving component, particles are randomly mapped to edge data centers when they are initialized.

[0049] In one embodiment of the present invention, step S2 further includes:

[0050] S22. Construct fitness function:

[0051] The problem encoding strategy may have infeasible solutions that do not meet the maximum response delay and data center capacity constraints; therefore, assuming that a given particle corresponds to a deployment scheme containing two industrial IoT applications in a cloud-edge collaborative environment, in order to enable the algorithm to have better search performance in the search space, the fitness function is defined as follows:

[0052] (1) If both solutions are feasible, choose the one with the lower deployment cost:

[0053]

[0054] (2) If one of the two solutions is a feasible solution and the other is an infeasible solution, choose the feasible solution:

[0055]

[0056] (3) If both solutions do not exceed the data center capacity limit but one of them violates the maximum response time constraint, choose the solution that meets the maximum response time of more industrial IoT applications:

[0057]

[0058] (4) If one of the two solutions does not meet the data center capacity limit, choose the solution that meets the data center capacity limit more:

[0059]

[0060] in is an event-indicative function, that is, when When the function is 1, otherwise, the function is 0.

[0061] In one embodiment of the present invention, step S2 further includes:

[0062] S23, particle update strategy:

[0063] In the search process, the evolution of particles is affected by their own state, individual optimal particles and the global optimal particles of the population, which correspond to the inertia part, individual cognitive part and group cognitive part of the particle update strategy respectively; therefore, MADPG introduces the mutation operator and crossover operator of the genetic algorithm in the particle update process to enhance the global and local search capabilities of particles. represents the kth particle in the G+1th iteration, and its update method is shown in formula (18):

[0064]

[0065] The current number of iterations G should be less than the maximum number of iterations G max , ⊙ and They represent the adaptive multi-point mutation operator and the two-point crossover operator respectively. and gB G They represent the individual optimal particle and the global optimal particle respectively;

[0066] For the inertia part, the adaptive multi-point mutation strategy in the genetic algorithm is used for updating. The updating method is shown in formula (19):

[0067]

[0068] Where r is a random number in the interval [0, 1]. If and only if r < ω, the particle Execute the adaptive multi-point mutation operator; randomly select k positions in the particle, the positions in bold are the selected mutation positions, and mutate the data center codes of these k positions. Without violating the principle of data protection, the values ​​​​are randomly mutated in [1, |D|];

[0069] For the individual learning and social learning parts, the two-point crossover strategy in the genetic algorithm is used for updating, and the updating methods are shown in formula (20) and formula (21) respectively:

[0070]

[0071]

[0072] Where r1 and r2 are both random numbers in the interval [0, 1]. If and only if r1 < c1 (r2 < c2), the particle Execute the two-point crossover operator; randomly select two positions in the particle, and then exchange the encoding between the two positions of the particle with the global optimal particle or the local optimal particle.

[0073] In one embodiment of the present invention, step S2 further includes:

[0074] S24, parameter adjustment:

[0075] To improve the effect of MADPG, an adjustment mechanism is used to adjust the current particle size according to the The value of the inertia parameter ω is adaptively adjusted according to the actual situation; its update method is formula (22) and formula (23):

[0076]

[0077]

[0078] in Indicates the current individual and the historical global optimal gBest t-1 The number of different codes, |T| represents the sum of the number of components of all industrial IoT applications; as shown in formula (22), when The smaller it is, the smaller the current particle and the historical global optimal particle gBest t-1 The smaller the difference is, the smaller ω is, thus increasing the local search capability; conversely, the larger ω is, the greater the global search capability is, thereby increasing the overall search performance of the algorithm;

[0079] In MADPG, the inertia part uses an adjustable multi-point mutation strategy, and the adjustment method of the mutation bit k is as shown in formula (24):

[0080]

[0081] where k min and k max They represent the maximum and minimum values ​​of the variation digit k respectively; as shown in formula (24), k increases as ω increases and decreases as ω decreases;

[0082] In addition, a linear increase and decrease strategy is adopted for updating the learning factors c1 and c2.

[0083] In one embodiment of the present invention, step S2 further includes:

[0084] S25. Mapping of coded particles to multiple industrial IoT application deployment scenarios

[0085] In the fuzzy cloud-edge environment, the mapping method of coded particles to multiple industrial IoT application deployment schemes is as follows: To simplify the representation, the superscripts and subscripts of the coded particles are omitted, that is, x is used instead. The input includes multiple industrial IoT applications A, cloud-edge collaborative environment D, and coded particles X. The output is the deployment scheme Γ corresponding to the coded particle X = (A, S, M, c t , T f ); Generate |A|+1 threads to simulate the deployment of multiple industrial Internet of Things applications in the cloud-edge collaborative environment; The main thread calls Algorithm 1 to monitor the arrival of industrial Internet of Things applications in real time, and then maps the corresponding components of the application to the corresponding data center according to the particles; Call Algorithm 2 to deploy multiple applications for all applications to be deployed; Calculate the deployment cost and response time according to equations (8) and (9) respectively; Then determine whether the particle is a feasible solution; Finally, return the industrial Internet of Things application deployment solution Γ = (A, S, m, c t , T f );in,

[0086] The input of Algorithm 1 includes industrial IoT application A, cloud-edge collaborative environment D, and particle code X. During the operation of the algorithm, if industrial IoT application a i Arrival, the computation time of each component’s response in each data center and the transmission time t in each data center are calculated according to equations (1) and (2). tc [|V i |×|D|] and t dt [|E i |,|D|×|D|]; Traverse industrial Internet of Things applications a iAll components are mapped to the corresponding data centers; otherwise, they wait for the arrival of an industrial IoT application; the algorithm ends when all industrial IoT applications have arrived;

[0087] The input of Algorithm 2 includes the industrial IoT application a i , the cloud-edge collaborative environment D and the mapping function M, first complete the deployment of components one by one according to the topological sorting and mapping; secondly, calculate the start time of the component in a response according to formula (3); then calculate the end time of the component according to formula (4); finally calculate the deployment cost of the component in the data center; until all components are traversed and the application deployment cost is calculated based on formula (6), the algorithm ends.

[0088] The present invention also provides a component deployment system for industrial Internet of Things applications, including a memory, a processor, and computer program instructions stored in the memory and capable of being executed by the processor. When the processor executes the computer program instructions, the method steps described above can be implemented.

[0089] The present invention also provides a computer-readable storage medium, on which computer program instructions that can be executed by a processor are stored. When the processor executes the computer program instructions, the method steps described above can be implemented.

[0090] Compared with the prior art, the present invention has the following beneficial effects:

[0091] (1) The present invention takes the perspective of industrial IoT application managers into consideration, and minimizes the deployment cost of application components in a cloud-edge environment under the constraints of application response time, privacy protection, and limited computing power of data centers;

[0092] (2) The present invention designs a MADPG algorithm that combines a genetic algorithm and a particle swarm optimization algorithm for this problem. In the particle swarm algorithm, a two-dimensional discrete particle encoding method that maps components and data centers is used to encode the deployment strategy of industrial Internet of Things applications. During the particle evolution process, the relevant learning factors are dynamically changed and genetic and mutation operations are used to improve the local and global search capabilities of the algorithm.

[0093] (3) Through simulation experiments, it is verified that compared with other benchmark strategies, MADPG can effectively reduce the deployment cost while ensuring the performance constraints of industrial Internet of Things applications. BRIEF DESCRIPTION OF THE DRAWINGS

[0094] Figure 1 It is a deployment framework for industrial Internet of Things applications in the cloud-edge collaborative environment of the present invention.

[0095] Figure 2 Encode strategies for particles.

[0096] Figure 3 It is an adaptive multi-point mutation operator.

[0097] Figure 4 is the two-point crossover operator.

[0098] Figure 5 This is the algorithm flow of MADPG. DETAILED DESCRIPTION

[0099] The technical solution of the present invention is described in detail below in conjunction with the accompanying drawings.

[0100] It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present application. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present application belongs.

[0101] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present application. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprise" and / or "include" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.

[0102] The present invention provides a component deployment method for industrial Internet of Things applications, and establishes a deployment model for industrial Internet of Things applications in a cloud-edge environment that minimizes deployment costs under the constraints of application response time, privacy protection, and limited computing power of data centers; on this basis, a multi-application deployment algorithm MADPG that combines GA with PSO is proposed to minimize the application deployment cost while meeting the constraints of industrial Internet of Things applications.

[0103] The following is the specific implementation process of the present invention.

[0104] 1. System model

[0105] Component deployment framework for industrial IoT applications in a cloud-edge collaborative environment Figure 1 As shown, it mainly includes three parts: multiple applications with constraints provided by multiple industrial Internet of Things terminals, a meta-heuristic deployer, and a cloud-edge collaborative environment.

[0106] 1.1 Multi-application model

[0107] In the industrial Internet of Things environment, intelligent terminal devices will submit applications to the cloud-edge collaborative environment to meet the requirements of industrial automation and intelligent manufacturing. The present invention represents a set of applications as a set A = {a1, a2, a3, ..., a n}. The device can turn complex applications intoi The application is split into stateless, interdependent components (such as microservices), so the present invention represents the application as a binary (l i , G i ), l i represents the maximum allowed response delay of the i-th application; G i Represents the data dependency structure of the i-th application.

[0108] The complex dependency structure between each application component will be represented as a directed acyclic graph G i =(V i , E i ). Where V i is a set of m components {v i1 , v i2 , ..., v im},v ij represents the jth component in the i-th application, component v ij The required computing resources are denoted as p ij (such as the number of CPU cores), the amount of computation of each component is expressed as τ ij ,s ij ∈{0, 1} indicates whether the component processes sensitive data. If s ij =1 means that components processing sensitive data can only be placed in edge data centers. Represents the data edge set between components. For an edge Represents component v i,p and component v i,q There is an association between each component v i,p To component v i,q The amount of data transmitted is At the same time, component v i,p is the component v i,q The predecessor of component v i,q is the component v i,p During application execution, a component can only start executing after all its predecessor components have completed execution and received their transmitted data.

[0109] 1.2 Cloud-Edge Collaborative Environment

[0110] In the industrial IoT environment, the cloud-edge collaborative environment mainly provides computing services to industrial IoT devices. cloud , D edge}It consists of cloud data centers and edge data centers (such as factory boxes). Cloud D cloud ={d1, d2, ..., d u} represents the set of u cloud data centers, edge D edge ={du+1 , d u+2 , ..., d u+v} indicates that there are v edge data centers near the factory. Data center d k It can be expressed as:

[0111]

[0112] where r k Indicates data center d k The computing resources owned (such as the number of CPU cores); o k Indicates data center d k The processing capacity of a unit of computing resources; Indicates that in the data center d k The price of renting a computing unit; f k ∈{0, 1} represents the type of data center. When f k =0, it belongs to the cloud data center, which has more computing resources and stronger unit computing power; when f k =1, d k The edge data center has general computing resources and computing power. It is assumed that the number of computing resources and the computing power per unit resource of the data center are known.

[0113] 1.3 Deployment Scheme for Industrial Internet of Things Applications

[0114] The goal of the metaheuristic deployer in the present invention is to give a deployment plan that minimizes the deployment cost of multiple applications under the conditions that each industrial IoT application meets the maximum allowed response delay, data protection, and does not exceed the computing power constraints of the data center. The deployment strategy is defined as follows:

[0115] Γ=(A,S,M,c t , T f )

[0116] Where M is a function: V→S represents the mapping of component V to cloud-edge collaborative environment S, such as M(v ij )=s k Indicates that the component v ij Deploy to data centers k In, c t represents the deployment cost of multiple industrial IoT applications A in the cloud-edge collaborative environment S, Represents the response time of each application under the deployment strategy Γ. In view of the characteristics of industrial Internet of Things applications, this study mainly focuses on the long-term deployment of application components in the data center, and optimizes the deployment cost under the constraints of application response time, component privacy protection and data center capacity limitations. To calculate the response time of the application. The present invention calculates the execution time based on the computing amount of each component and the computing power of the rented data center, and calculates the response time based on the transmission data size and bandwidth size. Therefore, the calculation method of the component computing time and data transmission time in one response is as follows:

[0117]

[0118]

[0119] In a request response component v ij In the data center k The computation time of component v is expressed as formula (1), ij On server d k The computing power it has is p ij o k Formula (2) represents the component v p With component v q If the two components are deployed in the same data center, the data transmission time can be ignored.

[0120] For a deployment scheme Γ, when the mapping function M is determined, the deployment of components of multiple industrial IoT applications A in the cloud edge environment D is also determined. According to the characteristics of industrial IoT applications, it is necessary to calculate the deployment cost c of multiple industrial IoT applications t and response time T f And each data center k Computing resources used

[0121] According to the mapping function M of multiple industrial IoT applications A to the cloud-edge collaborative environment D, the relevant variables are defined as follows:

[0122] (1)t start (v i,j , d k ): Component v in a request response i,j In the data center k The start time of the i,j It is determined by the completion time of all predecessor components, as shown in formula (3).

[0123]

[0124] (2)t end (vi,j , d k ): Component v in a request response i,j In the data center k The end time is equal to the component v i,j The start time of the data center d k The sum of the above calculation times is shown in formula (4).

[0125] t end (v ij , d k ) = t start (v ij , d k )+t tc (v ij , d k ) (4)

[0126] (3)cost(v ij , d k ): component v ij In the data center k The deployment cost is equal to the required computing resources and data center d k The product of the calculation unit prices is shown in formula (5).

[0127]

[0128] (4)cost(a i ): Apply a i The deployment cost is the sum of the deployment costs of each component in the data center, as shown in (6).

[0129]

[0130] (6) Data Center k The computing resources that have been used are all the computing resources deployed on server d k The sum of the computing resources required by the components on , as shown in formula (7).

[0131]

[0132] Based on the above definition, we can get the execution cost c of multiple industrial IoT applications: t , one response time As shown in formula (8) and formula (9) respectively.

[0133]

[0134]

[0135] The execution cost ct The deployment cost of all application components and the application response time The latest completion time of the components in the application.

[0136] For the deployment strategy of multiple industrial IoT applications, the goal is to minimize the deployment cost under the premise of satisfying the maximum response delay allowed by the application, the component data processing protection principle and the data center capacity limit. Therefore, the problem of multiple application deployment in the industrial IoT environment can be formalized as formula (10).

[0137]

[0138] Constraint 1 means that the application response time cannot exceed the maximum allowed response delay of the application, and constraint 2 means that if s ij =1, that is, the component processes private data and must be deployed in the edge data center. Constraint 3 indicates that the data center d k The number of occupied resources cannot exceed the computing resources available in its data center.

[0139] 2. Industrial Internet of Things Application Deployment Strategy

[0140] First, the classic PSO algorithm is explained, and then the relevant content about MADPG is introduced in detail.

[0141] 2.1 PSO algorithm

[0142] Markov decision process is a theoretical framework for optimizing the objective function through interactive learning with the environment. It is an idealized form of reinforcement learning in mathematics. In the process of using graph reinforcement learning to solve the multi-edge load balancing problem, the most appropriate offloading action between edge servers is decided based on the global edge server information processed by the graph convolutional neural network. For edge servers with high load rate and high task response delay, multi-edge load balancing can be achieved by offloading some arriving tasks to adjacent edge servers. MDP can model the task offloading process to maximize the specified long-term reward. Usually, MDP is defined as a four-tuple <,,,>, which represents the state space, action space, state transfer function and reward function respectively. PSO algorithm is an optimization algorithm based on swarm intelligence. The problem to be optimized is regarded as a particle in a multidimensional search space, and each particle is a solution to be optimized. It will learn from its own optimal and global optimal to update its own encoding. In order to comprehensively measure the pros and cons of each particle, the fitness function is introduced for evaluation. In the iterative process, each particle will dynamically update its speed and position based on its own or other particles' experience to obtain better fitness. Update according to equations (11) and (12).

[0143]

[0144]

[0145] Where t represents the current iteration number; the speed of the tth iteration of the i-th particle is The position of the i-th particle at the t-th iteration; and gBest t They represent the optimal particle after t iterations of the i-th particle and the global optimal particle after t iterations respectively; the optimization performance and convergence of the algorithm depend on the inertia factor ω. c1 and c2 reflect the learning trend of particles towards individual optimality and global optimality and are called learning factors. r1 and r2 are used to enhance the randomness of the algorithm and are random numbers between (0, 1) during the iteration process.

[0146] 2.2 MADPG

[0147] The present invention will specifically explain MADPG from the following five aspects.

[0148] 2.2.1 Question Coding

[0149] In order to enable the PSO algorithm to efficiently solve the combinatorial optimization problem of industrial Internet of Things application deployment, the present invention proposes an encoding method for mapping application components and data centers. A potential deployment scheme of multiple industrial Internet of Things applications in a cloud-edge collaborative environment corresponds to a particle in the particle swarm. In the Gth iteration, the kth particle in the particle swarm is represented as As shown in formula (13):

[0150]

[0151] Where NP represents the size of the particle population, represents the jth task v of the i-th industrial IoT application ij Deployed data center locations. It is worth noting that for the privacy-preserving component, particles are randomly mapped to edge data centers when they are initialized. Figure 2 Given a particle corresponding to a deployment scenario containing two industrial IoT applications in a cloud-edge collaborative environment. Each application contains four tasks, v 1,2 and v 2,4 Indicates that the component needs to process private data. The cloud-edge collaborative environment consists of one cloud data center and two edge data centers, numbered 1, 2, and 3 respectively. 11 and component v 12 For example, component v 11 The corresponding particle mapping is 1, indicating that the component is deployed to the data center numbered 1. 12It is a privacy protection component, so it can only be deployed in the edge data center. Therefore, the component is deployed in the edge data center numbered 3.

[0152] 2.2.2 Fitness Function

[0153] This study aims to obtain a better deployment solution that can meet the constraints of maximum response delay for all industrial IoT applications, privacy protection of application components, and data center capacity constraints to minimize deployment costs. If a solution can meet the constraints of maximum response delay for all industrial IoT applications, privacy protection of application components, and data center capacity, the present invention calls it a feasible solution. Therefore, the problem encoding strategy proposed above may have infeasible solutions that do not meet the constraints of maximum response delay and data center capacity. Therefore, in order to make the algorithm have better search performance in the search space, the fitness function is defined as follows:

[0154] (1) If both solutions are feasible, choose the solution with the lower deployment cost.

[0155]

[0156] (2) If there are two solutions, one is a feasible solution and the other is an infeasible solution, choose the feasible solution.

[0157]

[0158] (3) If both solutions do not exceed the data center capacity limit but one of them violates the maximum response time constraint, select the solution that meets the maximum response time of more industrial IoT applications.

[0159]

[0160] (4) If one of the two solutions does not meet the data center capacity constraints, choose the solution that meets more data center capacity constraints.

[0161]

[0162] in is an event-indicative function, that is, when When the function is 1, otherwise, the function is 0.

[0163] 2.2.3 Particle Update Strategy

[0164] In the search process, the evolution of particles is affected by their own state, individual optimal particles and the global optimal particles of the population, which correspond to the inertia part, individual cognitive part and group cognitive part of the particle update strategy respectively. Therefore, MADPG introduces the mutation operator and crossover operator of the genetic algorithm in the particle update process to enhance the global and local search capabilities of particles. represents the kth particle in the G+1th iteration, and its update method is shown in formula (18).

[0165]

[0166] The current number of iterations G should be less than the maximum number of iterations G max , ⊙ and They represent the adaptive multi-point mutation operator and the two-point crossover operator respectively. and gB G They represent the individual optimal particle and the global optimal particle respectively.

[0167] For the inertia part, the adaptive multi-point mutation strategy in the genetic algorithm is used for updating, and the updating method is shown in formula *19).

[0168]

[0169] Where r is a random number in the interval [0, 1]. If and only if r < ω, the particle Execute the adaptive multi-point mutation operator. Figure 3 As shown, k positions are randomly selected in the particle, and the bold font positions are the selected mutation positions. The data center codes of these k positions are mutated. Without violating the principle of data protection, the values ​​are randomly mutated in [1, |D|].

[0170] For the individual learning and social learning parts, the two-point crossover strategy in the genetic algorithm is used for updating, and the updating methods are shown in Equation (20) and Equation (21) respectively.

[0171]

[0172]

[0173] Where r1 and r2 are both random numbers in the interval [0, 1]. If and only if r1 < c1 (r2 < c2), the particle Perform a two-point crossover operator. Figure 4 As shown, it randomly selects two positions in the particle, and then exchanges the encoding between the two positions of the particle with the global optimal particle or the local optimal particle.

[0174] 2.2.4 Parameter Adjustment

[0175] In the PSO algorithm, the value of the inertia parameter ω affects the local search capability and global search capability of the algorithm. In order to improve the effect of the MADPG algorithm, the present invention adopts an adjustment mechanism that can adjust the search speed according to the current particle swarm. The value of the inertia parameter ω is adaptively adjusted according to the actual situation. The update method is formula (22) and formula (23):

[0176]

[0177]

[0178] in Indicates the current individual and the historical global optimal gBest t-1 The number of different codes, |T| represents the sum of the number of components in all industrial IoT applications. As shown in formula (22), when The smaller it is, the smaller the current particle and the historical global optimal particle gBest t-1 The smaller the difference, the smaller ω is, thus increasing the local search capability. Conversely, the larger ω is, the greater the global search capability is, thereby increasing the overall search performance of the algorithm.

[0179] In the MADPG algorithm, the inertia part uses an adjustable multi-point mutation strategy. The adjustment method of the mutation bit k is as shown in formula (24):

[0180]

[0181] where k min and k max They represent the maximum and minimum values ​​of the number of mutation bits k, respectively. As shown in formula (24), k increases as ω increases and decreases as ω decreases. This enhances the global and local search capabilities of the algorithm and further improves the algorithm performance.

[0182] In addition, the update method for the learning factors c1 and c2 adopts a linear increase and decrease strategy, and the specific update strategy will not be repeated here.

[0183] 2.2.5 Mapping of coded particles to multiple IIoT application deployment scenarios

[0184] In the fuzzy cloud-edge environment, the mapping of encoded particles to multiple IIoT application deployment scenarios is shown in Algorithm 1.

[0185] To simplify the representation, the superscripts and subscripts of the coded particles are omitted in this section, that is, x is used instead of

[0186]

[0187]

[0188] The input of the algorithm includes multiple industrial IoT applications A, cloud-edge collaborative environment D and coded particles X, and the output is the deployment scheme Γ corresponding to the coded particle X = (A, S, M, c t , T f ). It generates |A|+1 threads to simulate the deployment of multiple IIoT applications in the cloud-edge collaborative environment (lines 2-14). The main thread IIoT_Applications_Arriving(A, S, X) monitors the arrival of IIoT applications in real time, and then maps the corresponding components of the applications to the corresponding data centers according to the particles (lines 4-6). For all applications to be deployed, call Algorithm 3: IIoT_Applications_Deployment(a i , D, m) to deploy multiple applications (lines 7-11). According to equations (8) and (9), the deployment cost and response time are calculated respectively (line 12); then, it is determined whether the particle is a feasible solution (lines 13-16); finally, the industrial Internet of Things application deployment solution Γ = (A, S, m, c t , T f )(line 17).

[0189]

[0190]

[0191] Algorithm 2 can detect the arrival of industrial IoT application A in real time and map multiple industrial IoT applications to corresponding data centers. The process is shown in Algorithm 2. The input of the algorithm includes industrial IoT application A, cloud-edge collaborative environment D, and particle code X. During the operation of the algorithm, if industrial IoT application a i Arrival, the computation time of each component’s response in each data center and the transmission time t in each data center are calculated according to equations (1) and (2). tc [|V i |×|D|] and t dt [|E i |, |D|×|D|] (row 3). Traversing industrial IoT applications a i All components are mapped to the corresponding data centers (lines 4-6). Otherwise, it waits for the arrival of an IIoT application (lines 7-9). The algorithm ends when all IIoT applications have arrived.

[0192]

[0193] The simulation process of deploying industrial IoT applications to various data centers and the simulation of the response time and deployment cost is shown in Algorithm 3. The input of the algorithm includes industrial IoT applications ai , cloud-edge collaborative environment D and mapping function M. First, according to the topological sorting and mapping, the components are deployed one by one (line 2); secondly, the start time of the component in a response is calculated according to formula (3) (lines 3-9); then, according to formula (4), the end time of the component is calculated (line 10); finally, the deployment cost of the component in the data center is calculated (line 12); until all components are traversed and the application deployment cost is calculated based on formula (6) (line 14), the algorithm ends.

[0194] 2.3 Algorithm Flow

[0195] The algorithm flow of MADPG mainly includes the following six steps: Figure 5 shown.

[0196] 1) Initialize the relevant parameters of MADPG, set the population size, upper limit of iterations, inertia coefficient, learning rate and number of mutations, and then randomly initialize the population.

[0197] 2) According to Algorithm 1, each particle is mapped to the corresponding multi-industrial Internet of Things application deployment plan; through fitness function calculation, the advantages and disadvantages of each particle in the particle swarm are evaluated, and the individual optimal particle and the global optimal particle are selected.

[0198] 3) Verify whether the algorithm should be terminated, specifically, determine whether the number of iterations has reached the upper limit. If the condition is met, the algorithm is terminated and the process directly jumps to step 5); otherwise, the parameters are adjusted adaptively.

[0199] 4) Based on the proposed particle update strategy, the particle encoding is updated, the fuzzy fitness of each particle in the updated particle swarm is evaluated, and the corresponding individual optimal particles and the global optimal particles are updated at the same time.

[0200] 5) Output the global optimal particle in the particle swarm and terminate the algorithm.

[0201] 3. Simulation experiment and analysis

[0202] To verify the effectiveness of MADPG for industrial IoT application deployment in cloud-edge collaborative environment under the objectives of the present invention, this section will discuss the experimental simulation and result analysis.

[0203] 3.1 Experimental Setup

[0204] The experiments in this invention are simulated on a computer with 16GB memory and 2.60GHz 13th i5-13500H processor under Win 10 operating system. The parameters of MADPG are set as population size np = 100, maximum iteration limit G max =1000, inertia factor ω max =0.9,ω min =0.4; inertia factor Social Learning Factor Number of mutations k min =10.

[0205] The application submitted by each IIoT terminal device is randomly generated according to the following rules: the topology of each application is a directed acyclic graph, each application contains about 30 components, the computing resources required for each component are [1, 4], the computing amount of each component is [3000, 10000] MI, the component privacy protection rate is 1%, and if there is a dependency between components, the data transmission volume is [1, 10] MB. In addition, each application has a corresponding maximum allowable response delay. For an IIoT application a i The maximum allowable response delay is defined as shown in formula (25):

[0206] l i =bl*HEFT(a i ), bl={2.0, 2.5, 3.0, 3.5}(25)

[0207] Among them, HEFT(a i ) represents the deployment of industrial IoT applications using HEFT algorithm i response time.

[0208] Assume that there are 3 cloud data centers (s1, s2, s3) and 5 edge data centers (s4, s5, s6, s7, s8) in the cloud-edge environment. Each data center has a specific number of resources, computing power per unit resource, and rental price per unit resource. The resource volume of cloud data centers s1, s2, and s3 is {150, 200, 250}. The computing resources of edge data centers s4, s5, s6, s7, and s8 are {40, 45, 50, 55, 60}, respectively. The computing power per unit resource of cloud data centers s1, s2, and s3 is {1000, 1100, 1200} MIPS. The computing power per unit resource of edge data centers s4, s5, s6, s7, and s8 is {500, 550, 600, 650, 700} MIPS. Currently, the mainstream commercial cloud services, such as Amazon EC2 cloud data centers s1, s2, s3, have a unit resource rental price of {0.575, 0.735, 0.915}$ / day, and the unit resource rental price of edge data centers s4, s5, s6, s7, s8 is {0.271, 0.303, 0.337, 0.386, 0.422}$ / day.

[0209] 3.2 Benchmark Strategy

[0210] MADPG is a meta-heuristic deployment algorithm. In this experiment, the present invention uses the following three meta-heuristic deployment strategies as benchmark strategies to verify the effectiveness of the algorithm in reducing the deployment cost of multiple industrial IoT applications.

[0211] (1) PSO: This strategy adopts a coding scheme similar to MADPG and uses the traditional PSO update strategy to update the particle coding.

[0212] (2) GA: This strategy uses the same coding scheme as MADPG. According to the traditional GA update strategy, the current population is updated using crossover and multi-point mutation operators, and the elites are selected using binary tournaments. In this process, the crossover probability and mutation probability are taken as 0.7 and 0.1 respectively.

[0213] (3) RS: This strategy adopts the same encoding scheme as MADPG and uses a random search strategy to generate new particles. The particle with the maximum fitness in the search process is randomly recorded 20,000 times, which is the optimal scheduling strategy.

[0214] 3.3 Experimental Results and Analysis

[0215] To compare the performance of MADPG and other benchmark strategies in reducing the application deployment cost under the condition of different scales of industrial IoT applications. The present invention uses 4 industrial IoT applications of different scales, and conducts 10 groups of independent repeated experiments for each scale to take the average of their fitness. Industrial IoT applications are randomly generated by the above rules, where the privacy protection rate is 0.1 and the maximum response delay parameter bl is set to 3.5. Table 1 shows the deployment cost of different deployment strategies for industrial IoT applications of different scales.

[0216] Table 1 Deployment cost of applications of different scales under different strategies

[0217]

[0218]

[0219] Table 1 shows that ADPSO-GA can achieve the lowest deployment cost for industrial IoT applications of different scales, while PSO and GA perform second, and RS performs the worst. The deployment cost increases with the number of applications. The above table shows that compared with other benchmark strategies, the algorithm of the present invention can reduce the deployment cost by about 7.2%-33.5% for the number of industrial IoT applications of different scales. This is because the component deployment problem, as an NP-hard problem, has a large number of local extreme points, and MADPG dynamically changes the relevant learning factors during the particle evolution process and uses genetic and mutation operations to improve the local and global search capabilities of the algorithm, thereby obtaining a better deployment strategy.

[0220] In summary, the experimental results show that compared with other benchmark strategies, it can effectively meet the performance of industrial Internet of Things system applications and effectively reduce the cost of application deployment in applications of different scales.

[0221] The present invention also provides a component deployment system for industrial Internet of Things applications, including a memory, a processor, and computer program instructions stored in the memory and capable of being executed by the processor. When the processor executes the computer program instructions, the method steps described above can be implemented.

[0222] The present invention also provides a computer-readable storage medium, on which computer program instructions that can be executed by a processor are stored. When the processor executes the computer program instructions, the method steps described above can be implemented.

[0223] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that include computer-usable program code.

[0224] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0225] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0226] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0227] The above is only a preferred embodiment of the present invention, and does not limit the present invention in other forms. Any technician familiar with the profession may use the above disclosed technical content to change or modify it into an equivalent embodiment with equivalent changes. However, any simple modification, equivalent change and modification made to the above embodiment according to the technical essence of the present invention without departing from the technical solution of the present invention still belongs to the protection scope of the technical solution of the present invention.

Claims

1. A component deployment method for industrial Internet of Things applications, characterized in that: A deployment model for industrial Internet of Things applications in a cloud-edge environment is established to minimize the deployment cost under the constraints of application response time, privacy protection and limited computing power of data centers. On this basis, a multi-application deployment algorithm MADPG that combines GA and PSO is proposed to minimize the application deployment cost while meeting the constraints of industrial Internet of Things applications.

2. A component deployment method for industrial Internet of Things applications according to claim 1, characterized in that: The steps include: S1. Build system model: A component deployment system for industrial IoT applications in a cloud-edge collaborative environment, which includes three parts: a multi-application model with constraints provided by multiple industrial IoT terminals, a cloud-edge collaborative environment, and a meta-heuristic deployer; S2. Industrial Internet of Things Application Deployment: A multi-application deployment algorithm MADPG that combines GA and PSO is proposed to minimize the application deployment cost while satisfying the industrial Internet of Things application constraints.

3. A component deployment method for industrial Internet of Things applications according to claim 2, characterized in that: In step S1, the multi-application model is specifically as follows: A set of applications is represented as a set A = {a1, a2, a3, ..., a n }; Apply a i Represented as a two-tuple (l i , G i ), l i represents the maximum allowed response delay of the i-th application; G i Represents the data dependency structure of the i-th application; The complex dependency structure between each application component will be represented as a directed acyclic graph G i =(V i , E i ), where V i is a set of m components {v i1 , v i2 , …, v im },v ij represents the jth component in the i-th application, component v ij The required computing resources are denoted as p ij , the amount of computation of each component is expressed as τ ij ,s ij ∈{0, 1} indicates whether the component processes sensitive data. If s ij =1 means that components processing sensitive data can only be placed in edge data centers; Represents the data edge set between components. For an edge Represents component v i,p and component v i,q There is an association between each component v i,p To component v i,q The amount of data transmitted is At the same time, component v i,p is the component v i,q The predecessor of component v i,q is the component v i,p A component can only start executing after all its predecessor components have completed execution and received their transmitted data during application execution.

4. A component deployment method for industrial Internet of Things applications according to claim 3, characterized in that: In step S1, the cloud-edge collaborative environment is as follows: Cloud-edge collaborative environment D = {D cloud , D edge }It consists of cloud data centers and edge data centers; Cloud D cloud ={d1, d2, ..., d u } represents the set of u cloud data centers, edge D edge ={d u+1 , d u+2 , ..., d u+v } indicates that there are v edge data centers near the factory; data center d k It is expressed as: where r k Indicates data center d k The computing resources owned by k Indicates data center d k The processing capacity of a unit of computing resources; Indicates that in the data center d k The price of renting a computing unit; f k ∈{0, 1} represents the type of data center. When f k =0, it belongs to the cloud data center, which has more computing resources and stronger unit computing power; when f k =1, d k The edge data center has general computing resources and computing power. It is assumed that the number of computing resources and the computing power per unit resource of the data center are known.

5. A component deployment method for industrial Internet of Things applications according to claim 4, characterized in that: In step S1, the meta-heuristic deployer is as follows: The goal of the meta-heuristic deployer is to minimize the deployment cost of multiple applications given a deployment plan that satisfies the maximum allowable response delay, data protection, and does not exceed the computing capacity constraints of the data center for each industrial IoT application; the deployment strategy is defined as follows: Γ=(A,S,M,c t ,T f ) Where M is a function: V→S represents the mapping of component v to cloud-edge collaborative environment S, such as M(v ij )=s k Indicates that the component v ij Deploy to data centers k In, c t represents the deployment cost of multiple industrial IoT applications A in the cloud-edge collaborative environment S, represents the response time of each application under the deployment strategy Γ; in view of the characteristics of industrial IoT applications, the application components are deployed in the data center for a long time, and the deployment cost is optimized under the constraints of application response time, component privacy protection and data center capacity limitation; to calculate the response time of the application, the execution time is calculated according to the computing amount of each component and the computing power of the rented data center, and the response time is calculated according to the transmission data size and bandwidth size; the calculation method of the component computing time and data transmission time in one response is as follows: In a request response component v ij In the data center k The computation time of component v is expressed as formula (1), ij On server d k The computing power it has is p ij o k ; Formula (2) represents the component v p With component v q The resulting transmission time; if the two components are deployed in the same data center, their data transmission time is ignored; For a deployment strategy Γ, when the mapping function M is determined, the deployment of components of multiple industrial IoT applications A in the cloud edge environment D is also determined; according to the characteristics of industrial IoT applications, it is necessary to calculate the deployment cost c of multiple industrial IoT applications t and response time T f And each data center k Computing resources used According to the mapping function M of multiple industrial IoT applications A to the cloud-edge collaborative environment D, the relevant variables are defined as follows: (1)t start (v i,j , d k ): Component v in a request response i,j In the data center k The start time of the i,j It is determined by the completion time of all the predecessor components, as shown in formula (3): (2)t end (v i,j , d k ): Component v in a request response i,j In the data center k The end time is equal to the component v i,j The start time of the data center d k The sum of the above calculation time is shown in formula (4): t end (v ij ,d k )=t start (v ij ,d k )+t tc (v ij ,d k ) (4) (3)cost(v ij , d k ): component v ij In the data center k The deployment cost is equal to the required computing resources and data center d k The product of the calculation unit price is shown in formula (5): (4)cost(a i ): Apply a i The deployment cost is the sum of the deployment costs of each component in the data center, as shown in (6): (5) Data Center k The computing resources that have been used are all the computing resources deployed on server d k The sum of the computing resources required by the components on is shown in formula (7): Based on the above definition, we get the execution cost c of multiple industrial IoT applications: t , one response time As shown in formula (8) and formula (9) respectively: The execution cost c t The deployment cost of all application components and the application response time The latest completion time of the components in the application; For the deployment strategy of multiple industrial IoT applications, its goal is to minimize its deployment cost under the premise of satisfying the maximum allowed response delay of the application, the component data processing protection principle and the data center capacity limit; the problem of multiple application deployment in the industrial IoT environment is formalized as formula (10): Constraint 1 means that the application response time cannot exceed the maximum allowed response delay of the application, and constraint 2 means that if s ij =1, that is, the component processes private data and must be deployed in the edge data center. Constraint 3 indicates that the data center d k The number of occupied resources cannot exceed the computing resources available in its data center.

6. A component deployment method for industrial Internet of Things applications according to claim 5, characterized in that: Step S2 comprises: S21. Question code: A coding method for mapping application components and data centers is proposed; a potential deployment scheme for multiple industrial IoT applications in a cloud-edge collaborative environment corresponds to a particle in the particle swarm; in the Gth iteration, the kth particle in the particle swarm is represented as As shown in formula (13): Where NP represents the size of the particle population, represents the jth task v of the i-th industrial IoT application ij The location of the deployed data center; for the privacy-preserving component, particles are randomly mapped to edge data centers when they are initialized.

7. A component deployment method for industrial Internet of Things applications according to claim 6, characterized in that: Step S2 further includes: S22. Construct fitness function: The problem encoding strategy may have infeasible solutions that do not meet the maximum response delay and data center capacity constraints; therefore, assuming that a given particle corresponds to a deployment scheme containing two industrial IoT applications in a cloud-edge collaborative environment, in order to enable the algorithm to have better search performance in the search space, the fitness function is defined as follows: (1) If both solutions are feasible, choose the one with the lower deployment cost: (2) If one of the two solutions is a feasible solution and the other is an infeasible solution, choose the feasible solution: (3) If both solutions do not exceed the data center capacity limit but one of them violates the maximum response time constraint, choose the solution that meets the maximum response time of more industrial IoT applications: (4) If one of the two solutions does not meet the data center capacity limit, choose the solution that meets the data center capacity limit more: in is an event-indicative function, that is, when When the function is 1, otherwise, the function is 0.

8. The component deployment method for industrial Internet of Things applications according to claim 7, characterized in that: Step S2 further includes: S23, particle update strategy: In the search process, the evolution of particles is affected by their own state, individual optimal particles and the global optimal particles of the population, which correspond to the inertia part, individual cognitive part and group cognitive part of the particle update strategy respectively; therefore, MADPG introduces the mutation operator and crossover operator of the genetic algorithm in the particle update process to enhance the global and local search capabilities of particles. represents the kth particle in the G+1th iteration, and its update method is shown in formula *18): The current number of iterations G should be less than the maximum number of iterations G max , ⊙ and They represent the adaptive multi-point mutation operator and the two-point crossover operator respectively. and gB G They represent the individual optimal particle and the global optimal particle respectively; For the inertia part, the adaptive multi-point mutation strategy in the genetic algorithm is used for updating. The updating method is shown in formula (19): Where r is a random number in the interval [0, 1]. If and only if r < ω, the particle Execute the adaptive multi-point mutation operator; randomly select k positions in the particle, the positions in bold are the selected mutation positions, and mutate the data center codes of these k positions. Without violating the principle of data protection, the values ​​​​are randomly mutated in [1, |D|]; For the individual learning and social learning parts, the two-point crossover strategy in the genetic algorithm is used for updating, and the updating methods are shown in formula (20) and formula (21) respectively: Where r1 and r2 are both random numbers in the interval [0, 1]. If and only if r1 < c1 (r2 < c2), the particle Execute the two-point crossover operator; randomly select two positions in the particle, and then exchange the encoding between the two positions of the particle with the global optimal particle or the local optimal particle.

9. A component deployment method for industrial Internet of Things applications according to claim 8, characterized in that: Step S2 further includes: S24, parameter adjustment: To improve the effect of MADPG, an adjustment mechanism is used to adjust the current particle size according to the The value of the inertia parameter ω is adaptively adjusted according to the actual situation; its update method is formula (22) and formula (23): in Indicates the current individual and the historical global optimal gBest t-1 The number of different codes, |T| represents the sum of the number of components of all industrial IoT applications; as shown in formula (22), when The smaller it is, the smaller the current particle and the historical global optimal particle gBest t-1 The smaller the difference is, the smaller ω is, thus increasing the local search capability; conversely, the larger ω is, the greater the global search capability is, thereby increasing the overall search performance of the algorithm; In MADPG, the inertia part uses an adjustable multi-point mutation strategy, and the adjustment method of the mutation bit k is as shown in formula (24): where k min and k max They represent the maximum and minimum values ​​of the variation digit k respectively; as shown in formula (24), k increases as ω increases and decreases as ω decreases; In addition, a linear increase and decrease strategy is adopted for updating the learning factors c1 and c2.

10. A component deployment method for industrial Internet of Things applications according to claim 9, characterized in that: Step S2 further includes: S25. Mapping of coded particles to multiple industrial IoT application deployment scenarios In the fuzzy cloud-edge environment, the mapping method of coded particles to multiple industrial IoT application deployment schemes is as follows: To simplify the representation, the superscripts and subscripts of the coded particles are omitted, that is, x is used instead. The input includes multiple industrial IoT applications A, cloud-edge collaborative environment D, and coded particles X. The output is the deployment scheme Γ corresponding to the coded particle X = (A, s, M, c t , T f ); Generate |A|+1 threads to simulate the deployment of multiple industrial Internet of Things applications in the cloud-edge collaborative environment; The main thread calls Algorithm 1 to monitor the arrival of industrial Internet of Things applications in real time, and then maps the corresponding components of the application to the corresponding data center according to the particles; Call Algorithm 2 to deploy multiple applications for all applications to be deployed; Calculate the deployment cost and response time according to equations (8) and (9) respectively; Then determine whether the particle is a feasible solution; Finally, return the industrial Internet of Things application deployment solution Γ = (A, S, m, c t , T f );in, The input of Algorithm 1 includes industrial IoT application A, cloud-edge collaborative environment D, and particle code X. During the operation of the algorithm, if industrial IoT application a i Arrival, the computation time of each component’s response in each data center and the transmission time t in each data center are calculated according to equations (1) and (2). tc [|V i |×|D|] and t dt [|E i |,|D|×|D|]; Traverse industrial Internet of Things applications a i All components are mapped to the corresponding data centers; otherwise, they wait for the arrival of an industrial IoT application; the algorithm ends when all industrial IoT applications have arrived; The input of Algorithm 2 includes the industrial IoT application a i , the cloud-edge collaborative environment D and the mapping function M, first complete the deployment of components one by one according to the topological sorting and mapping; secondly, calculate the start time of the component in a response according to formula (3); then calculate the end time of the component according to formula (4); finally calculate the deployment cost of the component in the data center; until all components are traversed and the application deployment cost is calculated based on formula (6), the algorithm ends.

Citation Information

Patent Citations

  • Multi-workflow application scheduling method in fuzzy edge cloud collaborative environment

    CN115168011A

  • Method for supporting adaptive unloading of multi-internet of things (IOT) applications in edge environment

    US20240192995A1