Computing method for optimal task issuing and resource association in edge computing scene

Through the combination of genetic algorithm and Levenberg-Marquardt algorithm, the problem of limited resources in edge computing is solved, the optimization and issuance of computing tasks is realized, delay and energy consumption are reduced, and resource utilization efficiency is improved.

CN119065809APending Publication Date: 2024-12-03BEIJING INST OF COMP TECH & APPL
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411094847.0
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-11
Publication Date
2024-12-03

AI Technical Summary

Technical Problem

In edge computing scenarios, when resources are limited, how to reasonably choose a computing task issuance plan to optimize resource utilization, reduce data transmission delay and energy consumption, and enhance information security.

Method used

Using genetic algorithm variable step length gene fragment cross-interchange and Levenberg-Marquardt algorithm, reasonable samples are screened through loss function, optimal solutions are determined, and the selective outage of computing tasks is realized, considering time delay, energy consumption loss and resource limitations. .

Benefits of technology

Under the conditions of limited resources, efficient issuance of computing tasks is achieved, time delay and energy consumption loss are optimized, resource utilization is optimal, and has practical value and research prospects.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119065809A_ABST
    Figure CN119065809A_ABST
Patent Text Reader

Abstract

The invention relates to a computing method for optimal task issuing and resource association in an edge computing scene, and belongs to the technical field of Internet of Things, information security and edge computing. According to the method, an optimal solution is searched by utilizing genetic algorithm variable step size gene fragment cross exchange and a Levenberg-Marquardt algorithm, and the method is a solution for selectively issuing a calculation task to an edge side under the condition of limited resources. According to the method, sensitive factors of intensive tasks to time delay and loss factors such as center calculation time consumption, data transmission time consumption and edge calculation time consumption are fully considered, and the method is relatively close to the actual situation and has certain practical value.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the fields of Internet of Things, information security and edge computing technology, and specifically relates to a method for calculating optimal task delivery and resource association in an edge computing scenario. Background Art

[0002] With the continuous development of IoT technology, more and more terminals are connected to the Internet. According to the forecast of international companies, there will be 41.6 billion IoT devices in 2025, and the data generated will be 79.4ZB, and by 2035 the data volume will reach 2142ZB. Faced with such a huge amount of data, how to ensure the security of information transmission and system stability has become a major problem. Cloud computing is generally considered to be an important technical means to solve this problem. However, such a large amount of data is transmitted to the cloud center, which requires the reliability and security of the transmitted data, and the energy loss caused by data transmission must be balanced. In response to this problem, edge computing came into being. The computing tasks are issued on the end side, which not only reduces the spatial distance of data transmission and reduces the delay; the end-side processing also reduces the amount of data transmitted to the cloud center, which invisibly reduces the system pressure on the cloud, strengthens information security, increases the reliability of data transmission, saves network bandwidth, and reduces the energy consumption caused by transmission; in addition, the end-side deployment is more convenient for obtaining user personal information, and the edge end is usually a distributed architecture, which is difficult to attack and the system is more secure. Although there are many benefits to sending computing tasks to the edge, the edge is often limited by the number of resources. The deployment of all computing tasks on the edge requires huge resources, making it difficult to achieve. Therefore, how to reasonably select task delivery solutions within limited resources has become an increasingly discussed hot topic. Summary of the invention

[0003] 1. Technical issues to be resolved

[0004] The technical problem to be solved by the present invention is: to provide a calculation method for optimal task delivery and resource association in an edge computing scenario, so as to selectively deliver computing tasks to the edge side under the condition of limited resources.

[0005] (II) Technical solution

[0006] In order to solve the above technical problems, the present invention provides a method for calculating optimal task delivery and resource association in an edge computing scenario, comprising the following steps:

[0007] The first step is to collect the number of resources on the edge and cloud computing center, and the number of tasks that need to be managed. The resulting sample array is as shown in formula (1):

[0008] C={C1,C2,C3....C W} (1)

[0009] C i Represents the i-th task, with a value of 0 or 1, representing deployment at the cloud computing center or edge, respectively;

[0010] Step 2: Use random function s k = rand(0,1), assign values ​​to the sample array using formula (2), where 1 represents edge deployment and 0 represents cloud computing center computing. This process stops when the number of generated samples equals the number of tasks, and the resulting sample array is used as the initial condition sample set.

[0011]

[0012] r represents the element C in formula (1) i The value of

[0013] Step 3: Select preliminary reasonable samples from the initial condition sample set: Calculate the loss value through the loss function, select the set of samples whose loss value is less than the preset threshold α from the initial condition sample set, and set the final function threshold β. The relationship between the thresholds α and β satisfies formula (3), where n is a positive integer greater than 1; the number of selected samples Z also needs to satisfy formula (4). If no sample less than the preset threshold α is selected, or the number of samples does not satisfy formula (4), then return to the first step to regenerate a randomly generated sample set;

[0014] α=nβ,n>1 (3)

[0015] Z<ceil(length(C) / 3) (4)

[0016] ceil() is a rounding function, length() is a length function;

[0017] Step 4: Use different step lengths to exchange gene fragments to derive samples: Set different fragment lengths to cross-exchange gene fragments, and then obtain a set of derived samples after the exchange;

[0018] Step 5: Use the Levenberg-Marquardt algorithm to take the loss function as the objective function to determine the optimal solution of the derived samples. If the resources required by the optimal solution at the edge end exceed the expected resource limit or do not reach the set threshold β, the optimal solution is considered to be an invalid solution, and then go back to the first step to recalculate; otherwise, output the optimal solution.

[0019] Preferably, in the fourth step, set sample X = {1, 0, 1, 1, 0, 0, 1, 1, 1}, sample Y = {0, 1, 1, 0, 1, 1, 1, 1, 1}, set the value of d to 1, and perform gene exchange with a step size of 2d to obtain a new derivative sample X = {0, 1, 1, 1, 0, 0, 1, 1, 1}, a new derivative sample Y = {1, 0, 1, 1, 0, 0, 1, 1, 1}, and so on.

[0020] Preferably, the loss function expressions in the third and fifth steps are as shown in formula (5); the coefficient ρ represents the sensitivity of the computing task to delay. For scenarios that focus on delay requirements, the value of the coefficient ρ is increased, and the relationship between ρ and another preset coefficient δ is as shown in formula (6); T w is the total time delay function, as shown in (7), E w is the total energy loss function, as shown in (8);

[0021]

[0022] ρ+δ=1 (6)

[0023]

[0024] Q W Indicates whether the wth task is offloaded.

[0025] Preferably, the energy consumption loss includes both the cloud computing center and the edge. represents the time delay of the cloud computing center, as shown in formula (9), C w represents the number of resources required for the wth task, f l represents the CPU computing power when executing the wth task, and represents the energy loss generated by the cloud computing center, as shown in formula (10), where k is a constant value; represents the delay loss at the edge, as shown in formula (11), which is divided into two parts: one is the delay during data transmission The other part is the delay caused by data calculation; The calculation formula is as follows (12), D w represents the size of the data transferred, and r w represents the uplink rate of the network; the energy consumption generated by the edge computing task is As shown in formula (13), where p t represents the data transmission rate, and the energy loss is the sum of the energy consumption calculated by formula (10) and formula (13);

[0026]

[0027] Preferably, the minimum resource is a combination of the task weight coefficient ρ, the task resource C w , the computing power of the CPU, as shown in formula (14). The sum of the resources at this edge is less than the total resources F. Therefore, the Lagrangian function expression under the constraint of the inequality is as shown in formula (15):

[0028]

[0029] Among them, f w represents the execution capability of the CPU, and θ is the adjustment factor.

[0030] Preferably, each task is a multidimensional nonlinear model. The model formed by taking partial derivatives of the multidimensional nonlinear model is the Jacobian matrix, so as to obtain the Jacobi matrix of the multidimensional nonlinear model. If the expected threshold β is not met after multiple iterations, then the whole process is returned to the first step and re-executed.

[0031] The present invention also provides a method for implementing the selective distribution of computing tasks based on the method.

[0032] The present invention also provides an application of the method in the technical field of Internet of Things.

[0033] The invention also provides an application of the method in the field of information security technology.

[0034] The present invention also provides an application of the method in the field of edge computing technology.

[0035] (III) Beneficial effects

[0036] The present invention provides a method for calculating the optimal task dispatch and resource association in edge computing scenarios. The method uses genetic algorithm variable step length gene fragment crossover and Levenberg-Marquardt algorithm to search for the optimal solution. It is a solution for selectively dispatching computing tasks to the edge side under the condition of limited resources. Among them, the sensitive factors of intensive tasks to time delay, as well as loss factors such as central computing time, data transmission time, and edge computing time are fully considered, which is more in line with the actual situation and has certain practical value. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 The figure is an overall flow chart of the method of the present invention. DETAILED DESCRIPTION

[0038] In order to make the purpose, content and advantages of the present invention more clear, the specific implementation methods of the present invention are further described in detail below in conjunction with the drawings and examples.

[0039] The present invention proposes a solution that uses genetic algorithm variable step size gene crossover and Levenberg-Marquardt algorithm to selectively issue computing tasks under the condition of limited number of edge resources to achieve the best effect (the heat loss caused by multiple tasks processing data and the loss function loss value formed by the required time are minimized). First, the MEC (multi-access edge computing) edge resource deployment conditions, the total number of tasks, the CPU value required for the computing task, the uplink rate of different transmission lines, etc. are collected, and random sequence samples are determined as initial samples. Then, the loss function is calculated, and reasonable samples are screened out through the loss function. Gene fragments are selected through different step sizes for crossover and exchange, and a large number of new samples are derived. Then, the Levenberg-Marquardt algorithm is used for global convergence to obtain the idea of ​​the final solution.

[0040] Combine the following Figure 1 Specific steps to follow:

[0041] The first step is to collect the number of resources on the edge and cloud computing center, the number of tasks that need to be managed, and other conditions. The sample array obtained is as shown in formula (1):

[0042] C={C1,C2,C3....C W} (1)

[0043] C i represents the i-th task, with a value of 0 or 1, indicating deployment at the center or edge, respectively;

[0044] Step 2: Use random function s k = rand(0,1), assign values ​​to the sample array using formula (2), where 1 represents edge deployment and 0 represents cloud computing center computing. This process stops when the number of generated samples equals the number of tasks, and the resulting sample array is used as the initial condition sample set.

[0045]

[0046] r represents the element C in formula (1) i The value of

[0047] Step 3: Select preliminary reasonable samples from the initial condition sample set: Calculate the loss value through the loss function (5), select a set of samples whose loss value is less than the preset threshold α from the initial condition sample set, and set the final function threshold β. The relationship between the thresholds α and β satisfies formula (3), where n is a positive integer greater than 1; the number of selected samples Z also needs to satisfy formula (4). If no sample less than the preset threshold α is selected, or the number of samples does not satisfy formula (4), then return to the first step to regenerate a randomly generated sample set;

[0048] α=nβ,n>1 (3)

[0049] Z<ceil(length(C) / 3) (4)

[0050] ceil() is a rounding function, length() is a length function;

[0051] Step 4: Use different step lengths to exchange gene fragments to derive samples: Set different fragment lengths to cross-exchange gene fragments, and then obtain a set of derived samples after the exchange. For example, sample X = {1, 0, 1, 1, 0, 0, 1, 1, 1}, sample Y = {0, 1, 1, 0, 1, 1, 1, 1, 1}, set the value of d to 1, and perform gene exchange with a step length of 2d, and the new derived sample X = {0, 1, 1, 1, 0, 0, 1, 1, 1}, the new derived sample Y = {1, 0, 1, 1, 0, 0, 1, 1, 1}, and so on.

[0052] The samples initially screened out are positive sample data that tend toward the optimal solution. Some of the gene fragments in them are also playing a positive role, but the length cannot be determined. Therefore, the present invention randomly combines the gene fragments in these samples, and has the opportunity to combine all the positive gene fragments to obtain samples with better effects. The remaining steps are how to screen out these better samples from these sample sets, and use this as the most initial condition to determine the optimal solution.

[0053] Step 5: Use the Levenberg-Marquardt algorithm to take the loss function as the objective function to determine the optimal solution of these derived samples. If the resources required by the optimal solution at the edge end exceed the expected resource limit or do not reach the set threshold β, the optimal solution is considered to be an invalid solution, and then go back to the first step to recalculate; otherwise, output the optimal solution.

[0054] Among them, the loss function expression is as shown in formula (5); the coefficient ρ represents the sensitivity of the computing task to delay. For example, unmanned vehicle scenarios, AR game scenarios, and medical scenarios need to pay attention to delay requirements, so it is necessary to increase the weight coefficient of this value, and the relationship between ρ and the coefficient δ is as shown in formula (6); T w is the total time delay function, as shown in (7), E w is the total energy loss function, as shown in (8).

[0055]

[0056] ρ+δ=1 (6)

[0057]

[0058] QW Indicates whether the wth task is offloaded, because there are only two cases for a task: "offload to the local or transmit to the remote end". If it is offloaded locally, it only needs to calculate the local computing time and transmit the result to the remote user receiving segment, so Q W The value of can only be 0 or 1. The energy loss is divided into two parts: the cloud computing center and the edge. represents the time delay of the cloud computing center, as shown in (9), C w represents the number of resources required for the wth task, f l represents the CPU computing power when executing the wth task, and represents the energy loss generated by the cloud computing center, as shown in formula (10), where k is considered to be a constant value and is not affected by external operating factors; represents the delay loss at the edge, as shown in formula (11), which is divided into two parts: one is the delay during data transmission The other part is the delay caused by data calculation (i.e. the computing time loss of the task); The calculation formula is as follows (12), D w represents the size of the data transferred, and r w represents the uplink rate of the network; the energy consumption generated by the edge computing task is As shown in formula (13), where p t represents the data transmission rate, and the energy loss is the sum of the energy consumption calculated by formula (10) and formula (13);

[0059]

[0060]

[0061] The Levenberg-Marquardt algorithm is used in the process of obtaining the local optimal solution. This algorithm is a kind of Newton optimization algorithm for solving nonlinear problems. It has the characteristics of fast convergence speed, accurate calculation and high efficiency. It is a fast search method between the Newton method and the gradient descent method, but it is more dependent on the given initial value. The present invention uses the cross-exchange of different step-length gene fragments of the genetic algorithm to make the initial value characteristics widely representative, and the positive effect of each segment of the gene is highlighted, so that the shortcomings of the Levenberg-Marquardt algorithm are well filled. Through the decomposition transformation of the loss function, it can be known that the minimum resource is composed of the task weight coefficient ρ, the number of task resources C w , which is determined by the computing power of the CPU, as shown in formula (14). The sum of the resources at this edge is less than the total resources F. Therefore, the Lagrangian function expression under the constraint of the inequality is as shown in formula (15):

[0062]

[0063] Each task can be understood as a parameter, that is, a multidimensional nonlinear model. The model formed by taking partial derivatives of the multidimensional nonlinear model is the Jacobian matrix, which can be used to calculate the Jacobi matrix of the multidimensional nonlinear model, where f w represents the execution capability of the CPU, θ is the adjustment factor, and the Levenberg-Marquardt algorithm introduces the adjustment factor to ensure non-singularity, so that the Levenberg-Marquardt algorithm can be used to search for the optimal solution; if multiple iterations do not always meet the expected threshold β, then go back to the first step and re-execute the entire process.

[0064] Embodiment:

[0065] 1. Collect resource constraints on the MEC side and the number of tasks planned to be deployed.

[0066] 2. According to the number of tasks, calculate the corresponding loss value generated by the samples, intercept the samples that are less than the δ threshold, and count whether the number of samples is sufficient.

[0067] 3. Using a=1, the three gene fragments of a, 2a, and 3a are cross-exchanged to obtain the derived data sample.

[0068] 4. The Levenberg-Marquardt algorithm was used to search for the local optimal solution of the initial value and compare it with the α threshold, but no reasonable optimal solution was obtained.

[0069] 5. Regenerate the sample set according to the number of tasks deployed for calculation, start calculation, and repeat steps 2, 3, and 4.

[0070] 6. Get the optimal solution that meets the threshold.

[0071] The present invention proposes to use a variable step size genetic algorithm to generate different gene fragments for cross-exchange and derive offspring, and then use the Levenberg-Marquardt algorithm to perform optimal unlocking to solve the problems of resource scheduling and task assignment in the field of edge computing.

[0072] It can be seen that in the present invention, only some initial attribute information of the MEC system and computing tasks need to be determined to perform preliminary calculations. There are no other complex objective factors. The resources at the edge are included in the measurement constraints, and the impact of intensive tasks is fully considered, which has certain use value. The present invention uses a genetic algorithm to cross-exchange gene fragments of different step lengths, so that the characteristics of the offspring are obvious without deviating from the original parent. Secondly, the Levenberg-Marquardt algorithm is used for local optimal search. The calculation accuracy, calculation efficiency, and convergence speed of the local search algorithm are relatively high, but the dependence on the initial value is relatively strong, and the iteration of the genetic algorithm with different step lengths can complement it and achieve a better effect. With the development of edge computing, how to obtain an efficient and reasonable task offloading solution under limited resources is an important problem, so the present invention has certain research prospects.

[0073] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the technical principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A method for calculating optimal task delivery and resource association in edge computing scenarios. It is characterized in that The following steps are involved: The first step is to collect the number of resources on the edge and cloud computing center, and the number of tasks that need to be managed. The resulting sample array is as shown in formula (1): C={C 1 ,C 2 ,C 3 ....C W } (1) C i Represents the i-th task, with a value of 0 or 1, representing deployment at the cloud computing center or edge, respectively; Step 2: Use random function s k = rand(0,1), assign values ​​to the sample array using formula (2), where 1 represents edge deployment and 0 represents cloud computing center computing. This process stops when the number of generated samples equals the number of tasks, and the resulting sample array is used as the initial condition sample set. r represents the element C in formula (1) i The value of Step 3: Select preliminary reasonable samples from the initial condition sample set: Calculate the loss value through the loss function, select the set of samples whose loss value is less than the preset threshold α from the initial condition sample set, and set the final function threshold β. The relationship between the thresholds α and β satisfies formula (3), where n is a positive integer greater than 1; the number of selected samples Z also needs to satisfy formula (4). If no sample less than the preset threshold α is selected, or the number of samples does not satisfy formula (4), then return to the first step to regenerate a randomly generated sample set; α=nβ,n>1 (3) Z<ceil(length(C) / 3) (4) ceil() is a rounding function, length() is a length function; Step 4: Use different step lengths to exchange gene fragments to derive samples: Set different fragment lengths to cross-exchange gene fragments, and then obtain a set of derived samples after the exchange; Step 5: Use the Levenberg-Marquardt algorithm to take the loss function as the objective function to determine the optimal solution of the derived samples. If the resources required by the optimal solution at the edge end exceed the expected resource limit or do not reach the set threshold β, the optimal solution is considered to be an invalid solution, and then go back to the first step to recalculate; otherwise, output the optimal solution.

2. The method according to claim 1, It is characterized in that In the fourth step, let sample X = {1, 0, 1, 1, 0, 0, 1, 1, 1}, sample Y = {0, 1, 1, 0, 1, 1, 1, 1, 1}, set the value of d to 1, and perform gene exchange with a step size of 2d. The new derived sample X = {0, 1, 1, 1, 0, 0, 1, 1, 1}, the new derived sample Y = {1, 0, 1, 1, 0, 0, 1, 1, 1}, and so on.

3. The method according to claim 1, It is characterized in that The loss function expressions in the third and fifth steps are as shown in formula (5); the coefficient ρ represents the sensitivity of the computing task to delay. For scenarios that focus on delay requirements, the value of the coefficient ρ is increased, and the relationship between ρ and another preset coefficient δ is as shown in formula (6); T w is the total time delay function, as shown in (7), E w is the total energy loss function, as shown in (8); ρ+δ=1 (6) Q W Indicates whether the wth task is offloaded.

4. The method according to claim 3, It is characterized in that Energy consumption loss includes two parts: the cloud computing center and the edge. represents the time delay of the cloud computing center, as shown in formula (9), C w represents the number of resources required for the wth task, f l represents the CPU computing power when executing the wth task, and represents the energy loss generated by the cloud computing center, as shown in formula (10), where k is a constant value; represents the delay loss at the edge, as shown in formula (11), which is divided into two parts: one is the delay during data transmission The other part is the delay caused by data calculation; The calculation formula is as follows (12), D w represents the size of the data transferred, and r w Represents the network's uplink rate; The energy consumption generated by edge computing tasks is As shown in formula (13), where p t represents the data transmission rate, and the energy loss is the sum of the energy consumption calculated by formula (10) and formula (13); 5. The method according to claim 4, It is characterized in that The minimum resource is composed of the task weight coefficient ρ and the task resource C w , the computing power of the CPU, as shown in formula (14). The sum of the resources at this edge is less than the total resources F. Therefore, the Lagrangian function expression under the constraint of the inequality is as shown in formula (15): Among them, f w represents the execution capability of the CPU, and θ is the adjustment factor.

6. The method according to claim 5, It is characterized in that Each task is a multidimensional nonlinear model. The model formed by taking partial derivatives of the multidimensional nonlinear model is the Jacobian matrix, so as to obtain the Jacobi matrix of the multidimensional nonlinear model. If the expected threshold β is not met after multiple iterations, it will fall back to the first step and re-execute the entire process.

7. A method for selectively distributing computing tasks based on the method as claimed in any one of claims 1 to 6.

8. Application of the method as claimed in any one of claims 1 to 6 in the field of Internet of Things technology.

9. Application of the method according to any one of claims 1 to 6 in the field of information security technology.

10. An application of the method as described in any one of claims 1 to 6 in the field of edge computing technology.