A cloud-edge-end system resource collaboration method for intelligent production line management

Abstract

The present invention discloses a resource collaboration method for a cloud-edge-end system of intelligent production line control, which belongs to the field of cloud-edge-end technology. By combining economic models and optimization methods, the edge server and terminal devices can cooperate well. The specific contents are as follows: (1) The resource pricing and transaction model of the cloud-edge-end system of intelligent production line control uses economic principles and capital pricing to dynamically quantify the resources of the edge server; (2) A terminal periodic task unloading method based on a multi-armed bandit reinforcement learning method uses reinforcement learning to optimize the resource configuration required for periodic task unloading; (3) A resource collaboration method based on Stackelberg game takes the maximization of the overall benefits of the system as the goal to achieve the purpose of optimizing the overall resource utilization of the system. The present invention adopts the above method, targeting the cloud-edge-end system of intelligent production line control, so that the tasks unloaded by the terminal device can minimize the operating costs within the specified operating time.

Description

Technical Field

[0001] The present invention relates to the field of cloud-edge-end technology, and in particular to a method for coordinating cloud-edge-end system resources for intelligent production line management and control. Background Art

[0002] With the maturity of key technologies such as 5G networks and the Internet of Things (IoT), Industrial IoT cloud computing has shifted from a traditional cloud platform computing model to a "cloud + edge" edge computing model. Production line management and control systems, as a specialized scenario within the Industrial IoT, are characterized by cross-regional operations, long process flows, heterogeneous facilities, and diverse resource resources. These characteristics complicate resource scheduling for edge facilities on the production line, leading to uneven resource allocation and low resource utilization. Furthermore, edge devices often have limited resources, and inappropriate resource scheduling strategies can prevent production line terminal devices from accessing sufficient resources in a timely manner, impacting the real-time performance and efficiency of production line tasks. Therefore, for production line management and control systems, a rational cloud-edge-end resource coordination approach is needed to optimize resource allocation and task offloading across the cloud, edge, and end of the production line management IoT. This approach can effectively improve resource utilization of production line equipment and reduce task latency within the production line management and control system. Summary of the Invention

[0003] The purpose of the present invention is to provide a resource collaboration method for the cloud-edge-end system of intelligent production line management and control. For the cloud-edge-end system of intelligent production line management and control, a resource collaboration model based on resource pricing and transaction is first established between the edge server and the terminal device to drive and stimulate the resource scheduling of the edge server and the terminal device; for the specific periodic tasks to be unloaded by the terminal device on the production line, the reinforcement learning method based on the multi-armed bandit is used to select the best resource scheduling configuration scheme, so that the task can minimize the operating cost within the specified running time.

[0004] To achieve the above objectives, the present invention provides a method for intelligent production line management and control of cloud-edge-end system resource collaboration, comprising the following steps:

[0005] S1: Consider the edge server in the production line environment as a service provider, denoted as E = {E1, E2, ..., E n}, the terminal devices on the production line are regarded as users, denoted as U = {U1, U2, ..., U m}, the terminal server has sufficient funds; each time the terminal device U i On the edge server E j A task offloading process is regarded as a resource consumption process. Each edge server holds a limited number of k resources R = {R1, R2, ..., R k}, each resource R i The remaining available quantity is recorded as A i, define r=(r1,r2,…,r k ) represents a resource allocation plan, where r i represents the amount of resource i used, r i The value range is 0≤r i ≤A i ;

[0006] S2: Edge server sets initial resource price plan

[0007] S3: Initial resource configuration for terminal device to set task offloading (1) ;

[0008] S4: Before the tth round of task unloading, the terminal device obtains the resource price plan of all current edge servers from the edge server The terminal device uses a multi-armed bandit reinforcement learning method for each edge server to obtain the optimal resource configuration for task offloading, and selects the global optimal resource configuration from all the optimal resource configurations;

[0009] S5: The terminal performs task offloading according to the global optimal resource configuration, and the edge server executes the terminal task;

[0010] S6: Edge Server E i Record all resource allocation plans for terminal devices that are offloaded to themselves, and generate a new round of resource pricing plans based on terminal resource usage

[0011] S7: Implement the resource coordination process between edge and terminal devices based on the Stackelberg game, repeating steps S4 to S6, with the edge server as the leader and the terminal device as the follower.

[0012] Preferably, in step S4, the specific process of the terminal device selecting the global optimal resource is as follows:

[0013] S41: The formula for defining the optimal resource allocation is as follows:

[0014]

[0015] Set the target optimization function to Ω(r) = p(r) t(r), and the constraint formula is Where p(r) is the unit time price of resource configuration r provided by the edge server, t(r) is the running time of the task under r, is the set of all optional resource configurations, T is the maximum tolerable runtime limit of the task;

[0016] S42: Adjust the format of the target optimization function:

[0017] Let f(r) = 1 / t(r) represent the running rate of the task, then the formula of the objective optimization function of the multi-armed bandit machine is as follows:

[0018]

[0019] The formula for the optimal resource allocation after optimization is as follows:

[0020]

[0021] S43: The total resource configuration set on the edge server is is a finite set, where Indicates a value range of 0 to R i An integer, each resource configuration are all multidimensional integer vectors. The terminal device initially selects the initial resource configuration r in step S3. (1) ;

[0022] S44: Each resource configuration Each is defined as an "arm" of the slot machine, and the optimal "arm" is selected using multiple rounds of iterations;

[0023] S45: In the i-th iteration (i=1, 2, 3, ...), the terminal device uses resource configuration r (i) The task is offloaded to the edge server, and the task running rate formula is as follows:

[0024] y (i) =f(r (i) )+∈ (i)

[0025] In the above formula, y (i) is the actual observed task execution time, ∈ (i) is the observation interference term, ∈ (i) ~N(0,σ 2 ) is a disturbance term that obeys the normal distribution;

[0026] S46: Calculate the next round of iterative resource allocation based on the Gaussian process-confidence upper bound interval principle. The process is as follows:

[0027] The running rate function f(r) of the task is defined as a sampling of the Gaussian process GP(m(r),k(r,r')), where m(r) is the mean function of f(r), and k(r,r') is the covariance of f(r) and f(r'), that is, any The finite subsets of all obey the joint normal distribution N(m(r),k(r,r'));

[0028] After the tth iteration, the resource configuration {r (1) ,r(2) ,…,r (t)} and the noisy samples {y (1) ,y (2) ,…,y (t)}, we can get the posterior distribution of f to be normal distribution, for any r, The formulas for the expected function, covariance function, and variance function of f(r) are as follows:

[0029] m (t) (r)=(k (t) (r)) T (K (t) +σ 2 I) -1 y (t)

[0030] k (t) (r,r')=k(r,r')-(k (t) (r)) T (K (t) +σ 2 I) -1 k (t) (r')

[0031] (σ (t) ) 2 (r) = k t (r,r)

[0032] In the above formula, k (t) (r) and K (t) The formula is as follows:

[0033] k (t) (r)=[k(r (1) ,r),k(r (2) ,r),…,k(r (t) ,r)] T

[0034]

[0035] The square exponential kernel function is used as the prior covariance function of f, and the formula is as follows:

[0036]

[0037] In the above formula, σ 2 is the observed interference term ∈ (i) The variance of l 2 is the length scale parameter of the kernel function;

[0038] The terminal device uses the posterior distribution to predict the expected function and variance function of f after round t, and selects the resource configuration used for the t+1th round of iteration. The formula is as follows:

[0039]

[0040]

[0041] In the above formula, the parameters δ is a constant term;

[0042] S47: Repeat steps S45 and S46 until the function y is observed (t) Convergence, the convergence condition is σ (t) <θ·σ 2 , where θ is the convergence threshold weight, and the optimal resource configuration r of the edge server is obtained (t) ;

[0043] S48: Execute steps S41 to S47 for each edge server to obtain the optimal resource configuration for all edge servers calculate The maximum value of , the formula for the global optimal resource allocation is as follows:

[0044]

[0045] In the above formula, is the global optimal resource configuration, and the edge server selected for task offloading is E j .

[0046] Preferably, in step S41, there are two cases in which the edge server prices the resource p(r) of the terminal device:

[0047] In the fixed price case, when a terminal device requests a resource configuration r from an edge server, the resource pricing p(r) provided by the edge server to the terminal device is always fixed;

[0048] In the case of dynamic price, the terminal device requests resource allocation from the edge server in each iteration. (t) When the edge server provides different resource prices p (t) (r), before selecting the resource configuration before the next iteration, the terminal device uses the previous price to pre-calculate the resource pricing of the next round of resource configuration r:

[0049]

[0050] In the above formula, p (t-1) (r) is the price of the previous round, p (t-2) (r) is the pricing of the last two rounds.

[0051] Preferably, in step S6, the process of the edge server generating a new round of resource price scheme is as follows:

[0052] S61: According to the initial resource price plan of step S2 Setting up the resource R i The initial unit price p i The formula for ' is as follows:

[0053]

[0054] S62: Set resource supply and demand factor α for all terminal devices s , the formula is as follows:

[0055]

[0056] In the above formula, Request resource R from the edge server for all terminal devices in this round of task offloading i The total amount, S i For edge server resources R i The residual amount, γ is a constant;

[0057] S63: For all terminal devices, set the environmental competition factor α c , the formula is as follows:

[0058]

[0059] In the above formula, is the number of edge servers in the previous round for resource R i Average pricing;

[0060] S64: For specific terminal equipment U j , set the discrimination factor α d , the formula is as follows:

[0061]

[0062] In the above formula, It was the last round of U j Application Resource R i the number of This round of U j Apply for R i the number of The last round was U j Provided R i resource pricing;

[0063] S65: Summarize steps S62 to S64, and set the resource R for all terminal devices. iThe formula for the resource price scheme is as follows:

[0064]

[0065] In the above formula, is the resource price plan of the edge server for all terminal devices in this round;

[0066] For specific terminal equipment U j , the edge server sets the resource R i U j The resource price scheme formula is as follows:

[0067]

[0068] In the above formula, It is the resource price plan of the edge server for specific terminal devices in this round.

[0069] Therefore, the present invention adopts the above-mentioned method to provide a cloud-edge-device system resource collaboration method for intelligent production line management, which has the following advantages:

[0070] (1) In the present invention, the resource pricing and transaction model is used to effectively measure the weight of each resource by setting a price for the resources of the edge server.

[0071] (2) In the present invention, the goal of the edge server is to maximize its own benefits through resource trading, which can effectively motivate the edge server to share its own resources, thereby improving its own resource utilization.

[0072] (3) In the present invention, the terminal device uses a task offloading method based on multi-armed bandit reinforcement learning to obtain the lowest-cost resource configuration through multiple rounds of iteration, thereby ensuring the optimality of its own task offloading; in addition, the task running rate function f is modeled using the Gaussian process-confidence interval upper bound model, which improves the dynamic adjustment ability of the prediction result of the function f, so that the reinforcement learning process can converge to the optimal value quickly, effectively reducing the time required for resource scheduling while ensuring resource utilization.

[0073] (4) In the present invention, the Stackelberg game model is used to coordinate the resource supply and demand conflicts between the edge server and the terminal device. Both parties continuously adjust their own parameters to maximize their own interests, thereby indirectly achieving cooperation to maximize the overall interests of the system and ensure that the overall resource utilization of the system reaches the optimal state.

[0074] (5) In the present invention, the resource utilization of the edge server is quantified using resource pricing: for the resource set R of the edge server = {R1, R2, ..., R k}, if resource R iThe price per unit time is p i , then for a resource allocation scheme r=(r1,r2,…,r k ), the resource utilization of the edge server is This facilitates subsequent statistical calculations.

[0075] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0076] Figure 1 A schematic diagram of the cloud-edge-end in a resource collaboration method for a cloud-edge-end system for intelligent production line management and control according to the present invention;

[0077] Figure 2 This is a flow chart of the optimal resource configuration in a resource collaboration method for a cloud-edge-end system for intelligent production line management and control according to the present invention;

[0078] Figure 3 This is a flowchart based on Stackelberg game in the resource collaboration method of cloud-edge-end system for intelligent production line management and control of the present invention. DETAILED DESCRIPTION

[0079] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the invention claimed for protection, but merely represents selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0080] Example

[0081] The present invention provides a method for intelligent production line management and control of cloud-edge system resources collaboration, such as Figure 1 For the existing cloud-edge-end structure, the unified cloud center is responsible for the centralized control of the entire production line environment and has sufficient software and hardware resources; the edge server is far away from the cloud center but close to the production line terminal equipment, with limited resources and certain local computing and communication capabilities; the terminal equipment is mainly used for monitoring and scheduling the production process of the production line. Its geographical location is relatively dispersed but relatively fixed, with limited hardware resources and software operating environment. Many of its computing, communication and other types of tasks need to be completed with the help of edge servers. The tasks that the terminal device needs to offload are periodic tasks, that is, tasks that can terminate by themselves and need to be repeatedly executed.

[0082] Set n edge servers E={E1,E2,…,En}As the leader, each of them first sets resource pricing; m terminal devices U={U1,U2,…,U m As a follower, after obtaining the resource pricing of the edge server, the resource allocation plan is decided, which includes the following steps:

[0083] S1: Consider the edge server in the production line environment as a service provider, denoted as E = {E1, E2, ..., E n}, the terminal devices on the production line are regarded as users, denoted as U = {U1, U2, ..., U m}, the terminal server has sufficient funds; each time the terminal device U i On the edge server E j A task offloading process is regarded as a resource consumption process. Each edge server holds a limited number of k resources R = {R1, R2, ..., R k}, each resource R i The remaining available quantity is recorded as A i In each round of task offloading, each edge server needs to i Make reasonable pricing;

[0084] Each terminal device uses a multi-armed bandit-based reinforcement learning method to obtain the optimal offloading plan for periodic tasks under the current pricing scheme, defining r = (r1, r2, ..., r k ) represents a resource allocation plan, r i represents the amount of resource i used, r i The value range is 0≤r i ≤A i ;

[0085] S2: The edge server sets the initial resource price plan based on factors such as resource costs

[0086] S3: Initial resource configuration for terminal device to set task offloading (1) ;

[0087] S4: Before the tth round of task unloading, the terminal device obtains the resource price plan of all current edge servers from the edge server like Figure 2 ,The terminal device uses the multi-armed bandit reinforcement learning method for each edge server to obtain the optimal resource configuration for task offloading, and selects the global optimal resource configuration from all the optimal resource configurations;

[0088] S41: The formula for defining the optimal resource allocation is as follows:

[0089]

[0090] Set the target optimization function to Ω(r) = p(r) t(r), and the constraint formula is Where p(r) is the unit time price of resource configuration r provided by the edge server, t(r) is the running time of the task under r, is the set of all optional resource configurations, T is the maximum tolerable runtime limit of the task;

[0091] S42: Adjust the format of the target optimization function:

[0092] Let f(r) = 1 / t(r) represent the running rate of the task, then the formula of the objective optimization function of the multi-armed bandit machine is as follows:

[0093]

[0094] The formula for the optimal resource allocation after optimization is as follows:

[0095]

[0096] S43: The total resource configuration set on the edge server is is a finite set, where Indicates a value range of 0 to R i An integer, each resource configuration are all multidimensional integer vectors. The terminal device initially selects the initial resource configuration r in step S3. (1) ;

[0097] S44: Each resource configuration Each is defined as an "arm" of the slot machine, and the optimal "arm" is selected using multiple rounds of iterations;

[0098] S45: In the i-th iteration (i=1, 2, 3, ...), the terminal device uses resource configuration r (i) The task is offloaded to the edge server, and the task running rate formula is as follows:

[0099] y (i) =f(r (i) )+∈ (i)

[0100] In the above formula, y (i) is the actual observed task execution time, ∈ (i) is the observation interference term, ∈ (i) ~N(0,σ 2 ) is a disturbance term that obeys the normal distribution;

[0101] S46: Calculate the next round of iterative resource allocation based on the Gaussian process-confidence upper bound interval principle. The process is as follows:

[0102] The running rate function f(r) of the task is defined as a sampling of the Gaussian process GP(m(r),k(r,r')), where m(r) is the mean function of f(r), and k(r,r') is the covariance of f(r) and f(r'), that is, any The finite subsets of all obey the joint normal distribution N(m(r),k(r,r'));

[0103] After the tth iteration, the resource configuration {r (1) ,r (2) ,…,r (t)} and the noisy samples {y (1) ,y (2) ,…,y (t)}, we can get the posterior distribution of f to be normal distribution, for any r, The formulas for the expected function, covariance function, and variance function of f(r) are as follows:

[0104] m (t) (r)=(k (t) (r)) T (K (t) +σ 2 I) -1 y (t)

[0105] k (t) (r,r')=k(r,r')-(k (t) (r)) T (K (t) +σ 2 I) -1 k (t) (r')

[0106] (σ (t) ) 2 (r) = k t (r,r)

[0107] In the above formula, K (t) With k (t) The formula for (r) is as follows:

[0108] k (t) (r)=[k(r (1) ,r),k(r (2) ,r),…,k(r (t) ,r)] T

[0109]

[0110] The square exponential kernel function is used as the prior covariance function of f, and the formula is as follows:

[0111]

[0112] In the above formula, σ 2 is the observed interference term ∈ (i) The variance of l 2 is the length scale parameter of the kernel function;

[0113] The terminal device uses the posterior distribution to predict the expected function and variance function of f after round t, and selects the resource configuration used for the t+1th round of iteration. The formula is as follows:

[0114]

[0115]

[0116] Among them, the parameters It is used to balance the expectation term and the variance term in the confidence upper bound interval, where δ is a constant term;

[0117] S47: Repeat steps S45 and S46 until the function y is observed (t) Convergence, the convergence condition is σ (t) <θ·σ 2 , where θ is the convergence threshold weight, and the optimal resource configuration r of the edge server is obtained (t) ;

[0118] S48: Execute steps S41 to S47 for each edge server to obtain the optimal resource configuration for all edge servers calculate The maximum value of , the formula for the global optimal resource allocation is as follows:

[0119]

[0120] In the above formula, is the global optimal resource configuration, and the edge server selected for task offloading is E j .

[0121] In step S41, the edge server sets resource pricing p(r) for the terminal device in two cases:

[0122] In the case of fixed prices, when a terminal device requests a resource configuration r from an edge server, the resource pricing p(r) provided by the edge server to the terminal device is always fixed;

[0123] In the case of dynamic pricing, the terminal device requests resource allocation from the edge server in each iteration. (t)When the edge server provides different resource prices p (t) Before selecting the resource configuration for the next iteration, the terminal device pre-calculates the resource pricing for the next round of resource configuration r using the previous price:

[0124]

[0125] In the above formula, p (t-1) (r) is the price of the previous round, p (t-2) (r) is the pricing of the last two rounds.

[0126] S5: The terminal performs task offloading according to the global optimal resource configuration, and the edge server executes the terminal task;

[0127] S6: Edge Server E i Record all resource allocation plans for terminal devices that are offloaded to themselves, and generate a new round of resource pricing plans based on terminal resource usage

[0128] S61: According to the initial resource price plan of step S2 Setting up the resource R i The initial unit price p i The formula for ' is as follows:

[0129]

[0130] S62: Set resource supply and demand factor α for all terminal devices s , the formula is as follows:

[0131]

[0132] In the above formula, Request resource R from the edge server for all terminal devices in this round of task offloading i The total amount, S i For edge server resources R i The residual amount, γ is a constant;

[0133] S63: For all terminal devices, set the environmental competition factor α c , the formula is as follows:

[0134]

[0135] In the above formula, is the number of edge servers in the previous round for resource R i Average pricing;

[0136] S64: For specific terminal equipment U j , set the discrimination factor α d, the formula is as follows:

[0137]

[0138] In the above formula, It was the last round of U j Application Resource R i the number of This round of U j Apply for R i the number of The last round was U j Provided R i resource pricing;

[0139] S65: Summarize steps S62 to S64, and set the resource R for all terminal devices. i The formula for the resource price scheme is as follows:

[0140]

[0141] In the above formula, is the resource price plan of the edge server for all terminal devices in this round;

[0142] For specific terminal equipment U j , the edge server sets the resource R i U j The resource price scheme formula is as follows:

[0143]

[0144] In the above formula, This is the resource price plan of the edge server for a specific terminal device in this round

[0145] S7: As Figure 3 ,The resource coordination process of edge and terminal devices is realized based on Stackelberg game, and steps S4 to S6 are repeated, with the edge server as the leader and the terminal device as the follower.

[0146] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit the same. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that they can still modify or replace the technical solutions of the present invention with equivalents, and these modifications or equivalent replacements cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for coordinating cloud-edge-device system resources for intelligent production line management and control, characterized in that: The following steps are involved: S1: Consider the edge server in the production line environment as a service provider, denoted as E = {E1, E2, ..., E n }, the terminal devices on the production line are regarded as users, denoted as U = {U1, U2, ..., U m }, the terminal server has sufficient funds; each time the terminal device U i On the edge server E j A task offloading process is regarded as a resource consumption process. Each edge server holds a limited number of k resources R = {R1, R2, ..., R k }, each resource R i The remaining available quantity is recorded as A i , define r=(r1,r2,…,r k ) represents a resource allocation plan, where r i represents the amount of resource i used, r i The value range is 0≤r i ≤A i ; S2: Edge server sets initial resource price plan S3: Initial resource configuration for terminal device to set task offloading (1) ; S4: Before the tth round of task unloading, the terminal device obtains the resource price plan of all current edge servers from the edge server The terminal device uses a multi-armed bandit reinforcement learning method for each edge server to obtain the optimal resource configuration for task offloading, and selects the global optimal resource configuration from all the optimal resource configurations; S5: The terminal performs task offloading according to the global optimal resource configuration, and the edge server executes the terminal task; S6: Edge Server E i Record all resource allocation plans for terminal devices that are offloaded to themselves, and generate a new round of resource pricing plans based on terminal resource usage S7: Implement the resource coordination process between edge and terminal devices based on the Stackelberg game, repeating steps S4 to S6, with the edge server as the leader and the terminal device as the follower.

2. The method for intelligent production line management and control of cloud-edge-end system resources collaboration according to claim 1 is characterized by: In step S4, the terminal device selects the global optimal resource in the following specific process: S41: The formula for defining the optimal resource allocation is as follows: Set the target optimization function to Ω(r) = p(r) t(r), and the constraint formula is Where p(r) is the unit time price of resource configuration r provided by the edge server, t(r) is the running time of the task under r, is the set of all optional resource configurations, T is the maximum tolerable runtime limit of the task; S42: Adjust the format of the target optimization function: Let f(r) = 1 / t(r) represent the running rate of the task, then the formula of the objective optimization function of the multi-armed bandit machine is as follows: The formula for the optimal resource allocation after optimization is as follows: S43: All resource configuration sets on edge servers is a finite set, where Indicates a value range of 0 to R i An integer, each resource configuration are all multidimensional integer vectors. The terminal device initially selects the initial resource configuration r in step S3. (1) ; S44: Each resource configuration Each is defined as an "arm" of the slot machine, and the optimal "arm" is selected using multiple rounds of iterations; S45: In the i-th iteration (i=1, 2, 3, ...), the terminal device uses resource configuration r (i) The task is offloaded to the edge server, and the task running rate formula is as follows: y (i) =f(r (i) )+∈ (i) In the above formula, y (i) is the actual observed task execution time, ∈ (i) is the observation interference term, ∈ (i) ~N(0,σ 2 ) is a disturbance term that obeys the normal distribution; S46: Calculate the next round of iterative resource allocation based on the Gaussian process-confidence upper bound interval principle. The process is as follows: The running rate function f(r) of the task is defined as a sampling of the Gaussian process GP(m(r),k(r,r')), where m(r) is the mean function of f(r), and k(r,r') is the covariance of f(r) and f(r'), that is, any The finite subsets of all obey the joint normal distribution N(m(r),k(r,r')); After the tth iteration, the resource configuration {r (1) ,r (2) ,…,r (t) } and the noisy samples {y (1) ,y (2) ,…,y (t) }, we can get the posterior distribution of f to obey the normal distribution, for any The formulas for the expected function, covariance function, and variance function of f(r) are as follows: m (t) (r)=(k (t) (r)) T (K (t) +σ 2 I) -1 y (t) k (t) (r,r')=k(r,r')-(k (t) (r)) T (K (t) +σ 2 I) -1 k (t) (r') (s (t) ) 2 (r)=k t (r,r) In the above formula, k (t) (r) and K (t) The formula is as follows: k (t) (r)=[k(r (1) ,r),k(r (2) ,r),…,k(r (t) ,r)] T The square exponential kernel function is used as the prior covariance function of f, and the formula is as follows: In the above formula, σ 2 is the observed interference term ∈ (i) The variance of l 2 is the length scale parameter of the kernel function; The terminal device uses the posterior distribution to predict the expected function and variance function of f after round t, and selects the resource configuration used for the t+1th round of iteration. The formula is as follows: In the above formula, the parameters δ is a constant term; S47: Repeat steps S45 and S46 until the function y is observed (t) Convergence, the convergence condition is σ (t) <θ·σ 2 , where θ is the convergence threshold weight, and the optimal resource configuration r of the edge server is obtained (t) ; S48: Execute steps S41 to S47 for each edge server to obtain the optimal resource configuration for all edge servers calculate The maximum value of , the formula for the global optimal resource allocation is as follows: In the above formula, is the global optimal resource configuration, and the edge server selected for task offloading is E j .

3. The method for intelligent production line management and control of cloud-edge-end system resources collaboration according to claim 2 is characterized by: In step S41, the edge server sets resource pricing p(r) for the terminal device in two cases: In the fixed price case, when a terminal device requests a resource configuration r from an edge server, the resource pricing p(r) provided by the edge server to the terminal device is always fixed; In the case of dynamic price, the terminal device requests resource allocation from the edge server in each iteration. (t) When the edge server provides different resource prices p (t) (r), before selecting the resource configuration before the next iteration, the terminal device uses the previous price to pre-calculate the resource pricing of the next round of resource configuration r: In the above formula, p (t-1) (r) is the price of the previous round, p (t-2) (r) is the pricing of the last two rounds.

4. The method for intelligent production line management and control of cloud-edge-end system resources collaboration according to claim 1 is characterized by: In step S6, the process of the edge server generating a new round of resource price plan is as follows: S61: According to the initial resource price plan of step S2 Setting up the resource R i The initial unit price p i The formula for ' is as follows: S62: Set resource supply and demand factor α for all terminal devices s , the formula is as follows: In the above formula, Request resource R from the edge server for all terminal devices in this round of task offloading i The total amount, S i For edge server resources R i The residual amount, γ is a constant; S63: For all terminal devices, set the environmental competition factor α c , the formula is as follows: In the above formula, is the number of edge servers in the previous round for resource R i Average pricing; S64: For specific terminal equipment U j , set the discrimination factor α d , the formula is as follows: In the above formula, It was the last round of U j Application Resource R i the number of This round of U j Apply for R i the number of The last round was U j Provided R i resource pricing; S65: Summarize steps S62 to S64, and set the resource R for all terminal devices. i The formula for the resource price scheme is as follows: In the above formula, is the resource price plan of the edge server for all terminal devices in this round; For specific terminal equipment U j , the edge server sets the resource R i U j The resource price scheme formula is as follows: In the above formula, It is the resource price plan of the edge server for specific terminal devices in this round.

Images