An Edge Computing Task Offloading Method and Computer Device Considering Conditional Value at Risk in Industrial Internet

Abstract

The present invention discloses an edge computing task offloading method and a computer device considering conditional value at risk in industrial Internet. Considering the high latency risk that may occur in tasks due to channel uncertainty and the uncertainty of the MEC set available for task offloading in edge computing, a local computing model, an edge computing model, and a latency risk value model of computing tasks are established. With the minimization of the sum of latency and latency risk value as the optimization goal, considering the task offloading constraint and the latency constraint, through the first-order moment and second-order moment distributionally robust fuzzy set method, the latency problem is approximated as a semidefinite programming problem (SDP). An improved latency optimization offloading algorithm based on the branch and bound method is proposed to rationally allocate the task offloading problem, and to provide better services for users by achieving lower offloading latency in edge computing. Finally, by specifying the value of the confidence level β and changing the weight of the conditional value at risk (CVaR), the high latency risk is avoided.

Description

Technical Field

[0001] The present invention belongs to the field of industrial Internet mobile edge computing, and specifically relates to a two-stage distributionally robust optimization method for task offloading considering delay risk values. Background Art

[0002] With the rapid development of Internet technology, Mobile Edge Computing (MEC) has become an important part of Internet technology. As people's demands become higher and higher, the number of industrial devices accessing the Internet is increasing. Relying solely on the traditional cloud computing model, it is difficult to meet the requirements of industrial applications in terms of delay and economy. As an emerging computing model, edge computing processes tasks by transmitting them to edge devices. Compared with cloud computing, it can provide a faster service response time and reduce network overhead.

[0003] An important indicator for evaluating the quality of an edge computing system is service delay. There are already a large number of solutions for reducing the service delay of the system. However, most of the strategies are designed based on the average delay. In a complex industrial Internet environment, some tasks have extremely strict requirements for delay. Delay jitter exceeding the threshold can, in the lightest case, lead to device deadlocks and damage, and in the worst case, even cause significant economic losses. Therefore, we not only need to consider the average performance of delay, but also pay attention to the characterization of transmission performance, that is, risk. Initially, people applied risk control theory in the financial field. After decades of research, Conditional Value at Risk (CVaR) has become a mature tool for characterizing risk, and the risk in an edge computing system can be characterized by means of CVaR.

[0004] One of the main challenges faced by the edge computing problem considering delay risk values is how to handle the uncertainties of the wireless channel and the set of available MECs for transmission. As we all know, the wireless channel is dynamically changing, and due to the mobility of devices and MECs, the set of MECs to which tasks can be transmitted is uncertain. How to handle these problems is a difficult point. Currently, the commonly used methods are stochastic programming and robust optimization. Compared with stochastic programming methods, robust optimization does not require obtaining all the information of the data, so it has better application value in the industrial Internet. Summary of the Invention

[0005] In view of the above problems, the present invention proposes a task offloading method and corresponding computer device in an industrial Internet that takes into account conditional risk value, which can handle industrial Internet offloading scenarios and consider uncertain factors to solve the problem of balancing delay and high delay risk during the task offloading process in the industrial Internet. The solution of the present invention can not only ensure that the task offloading in the industrial Internet remains feasible in any scenario, but also consider how to avoid risks when risks are added. The specific technical solutions are as follows:

[0006] Step 1: Establish a local computing model, an edge computing model, and a delay risk value model for the computing task;

[0007] Step 2: According to the models established in Step 1, considering both the average delay performance and risk, define the objective function as the sum of the average delay and its delay risk value, which is called the delay risk sum. Take minimizing the delay risk sum as the optimization objective to solve the uncertainty of the wireless channel state and the offloading MEC set, which makes it difficult to determine the computing time and high delay risk on the MEC;

[0008] Step 3: Through the distributionally robust method, establish a fuzzy set of the first moment and the second moment, and construct a two-stage distributionally robust model according to the model optimization objective in Step 2. The first stage is the decision-making stage, which decides whether the task is computed locally or offloaded to the MEC for computing. The second stage is the MEC computing stage, assuming that the decision in the first stage is known, and taking minimizing the delay risk sum as the optimization objective.

[0009] Step 4: Transform the optimization problem of minimizing the delay risk sum in Step 3 into a semidefinite programming problem;

[0010] Step 5: Use the branch and bound method to find the optimal offloading scheme for the task, so as to realize the offloading decision.

[0011] Furthermore, the specific methods for establishing the relevant computing models in Step 1 and Step 2 are as follows:

[0012] 1) Local computing model:

[0013]

[0014] Where represents the local computing time of task Q i f i l represents the CPU computing power of the local device, and w i represents the computing amount of task Q i ;

[0015] 2) Edge computing model:

[0016]

[0017] Where represents the edge computing time of task Q i , including the edge execution time and the transmission delay, represents the computing resources allocated by the MEC, r i,k represents the user uplink data rate, s i represents the computing task Q iThe size.

[0018] 3) Latency risk value model: When performing some critical tasks with high latency requirements in the industrial Internet, extremely high reliability needs to be provided to ensure the safe and efficient execution of tasks. Therefore, not only the average latency of computing tasks needs to be concerned, but also the risk of high latency occurring with small probability. The present invention introduces CVaR as a risk metric and establishes the risk model of latency as:

[0019]

[0020] where α is the quantity to be optimized. In an ideal situation, when the optimization is completed, α is the conditional risk value corresponding to the confidence level β. is the expectation with respect to the probability distribution P. 4) Objective function: Minimize the computing time of tasks assigned to local devices and the time later assigned to MEC tasks through the objective function, and add the latency risk value to this objective function. Specifically as follows:

[0021] where x is the offloading decision combination vector, that is, x = (1 - x

[0022]

[0023] ,..., 1 - x 1,k ,..., 1 - x |I|,k ), T If x i,k = 1, it means that task i is offloaded to the MEC for processing. If x i,k = 0, it means that task i is processed locally. E P [Q(x, ε)] represents the expected time for tasks to be processed on the MEC. CVaR β [Q(x, ε)] represents the latency risk. I represents the set of tasks. Q(x, ε) represents the task latency under the offloading decision combination vector x in the industrial Internet scenario ε. λ ∈ [0, 1] is the latency risk weight, and the value of λ is adjusted to balance the average latency and the risk.

[0024] Furthermore, the specific process of step 3 for transforming the original problem into a two-stage distributionally robust model considering the latency risk value is as follows: 1) Through the definition of CVaR, replace Therefore, the latency risk value is denoted as

[0025]

[0026] 2) The objective function can be transformed into

[0027]

[0028] where tT Represents the set of task local computing times, \(t\) T \(x\) is the local computing time for the first-stage task to make an offloading decision

[0029] 3) For the convenience of calculation, let \(\varepsilon\) be equivalent to \(t\) E , where represents the computing time vector of the task on the MEC. Due to the uncertainty of the channel and the set of available MECs, it cannot be determined Use some historical data to establish a fuzzy set based on the first moment and second moment

[0030]

[0031] where is the measure space The set of all probability measures, \(B\) is the Borel \(\sigma -\)algebra of \(\Omega\), \(M\) is any closed convex set containing the support of the probability distribution \(P\) Denote by the standard deviation of \(\xi\), and \(\Lambda\) are the parameter vectors controlling the size of the fuzzy set Denote the semidefinite constraint (\( then represents that \(B - A\) is a positive definite matrix), and \(\sum\) t are the estimates of the mean and covariance matrix of the random vector \(t\) E respectively.

[0032] 4) Based on the fuzzy set of the first moment and second moment, construct a two-stage distributionally robust model for task offloading.

[0033]

[0034] where represents using the worst-case scenario to calculate the average delay and delay risk in the objective function to ensure the stable feasibility of the model.

[0035] 5) According to the strong duality theory, transform the distributionally robust model into

[0036]

[0037] Furthermore, the specific process of step 4 transforming the two-stage distributionally robust problem into a semidefinite programming problem is as follows:[[]]

[0038] 1) Define the intermediate auxiliary variable

[0039] Since in (8) It is an infinite-dimensional optimization problem with a probability distribution P. Therefore, the objective function and the first and second moment constraints in the second stage are as follows:

[0040]

[0041] s,t

[0042]

[0043] 2) Introduce the dual variable z of constraint (9), z ,Z, and consider the dual problem of the problem

[0044]

[0045] where indicates that z is a real number, indicates and z are m-dimensional real vectors,

[0046] indicates that Z is an m-dimensional symmetric positive semi-definite matrix.

[0047] 3) Transform the dual problem into a tractable semidefinite programming problem.

[0048]

[0049] where the defined intermediate variable is mainly defined for computational convenience. y = (x 1 ,k,…,x |I| ,k) T is the offloading decision vector in the second stage, representing which MEC the task i is offloaded to for processing.

[0050] Furthermore, the specific process of step 5 to find the offloading decision variable x in the first stage and update the offloading decision y in the second stage is as follows: 1) First, without considering the offloading decision constraints of the original two-stage latency problem, solve the corresponding relaxation problem, and denote the result as T min .

[0051] 2) If the obtained T min just satisfies the binary constraint condition, then this solution is the optimal solution of the original problem.

[0052] 3) Branch. According to the priority P of the tasks described in step 1 i select the task Q i , and construct two constraint conditions x i,k = 1 and x i,k= 0 represents whether the task is offloaded to the edge computing or computed on the local device. Adding these two constraints forms two sub-problems of the original problem.

[0053] 4) Bounding. First, solve whether there is a feasible solution for the two sub-problems. If there is, denote it as T 1 and T 2 . Judge the sizes of T 1 and T 2 . If T 1 < T 2 , then let T 1 = T min , otherwise, let T 2 = T min .

[0054] 5) If the solution of the sub-problem does not reach the optimal solution, or the offloading decision x i,k does not satisfy all being 0-1 variables, then it is necessary to continue with branch and bound (repeat steps 3) and 4)) until the offloading decision is all 0-1 variables. Finally, determine the first-stage decision variable x according to the offloading decision x i,k and update the second-stage decision variable y.

[0055] A computer device of the present invention has the execution code or storage code of the above-mentioned method for offloading edge computing tasks considering conditional value at risk in an industrial Internet built therein.

[0056] Advantages of the present invention:

[0057] (1) The present invention provides a two-stage task offloading problem considering conditional risk value. Aiming at the latency problem in the industrial Internet, a two-stage distributionally robust model based on the first-order moment and second-order moment fuzzy set is constructed, and the problem is transformed into a semidefinite programming problem for solution. Compared with the existing edge computing task offloading schemes, it ensures that it can still operate safely under the worst application scenarios.

[0058] (2) For some critical tasks with high latency requirements in the industrial Internet, extremely high reliability is required to ensure the safe and efficient operation of the tasks. The present invention not only pays attention to the average latency of the computing tasks, but also considers the risk of high latency occurring with small probability. By specifically designing the conditional risk value, it can well balance the average latency and the risk. Description of the Drawings

[0059] Figure 1 is the flowchart of the task offloading method in the industrial Internet of the present invention;

[0060] Figure 2 is the flowchart of the branch and bound method for solving the first-stage offloading decision x;

[0061] Figure 3 The influence of λ and β on the expected offloading time;

[0062] Figure 4 The influence of the distributionally robust uncertainty set on the expected offloading time. Specific implementation manners

[0063] The present invention proposes an edge computing task offloading method and a computer device considering conditional value at risk in an industrial Internet, and the overall technical idea is as follows:

[0064] Considering the high delay risk that may occur in tasks due to channel uncertainty and the uncertainty of the MEC set available for task offloading in edge computing, a local computing model, an edge computing model, and a delay risk value model of the computing task are established. Taking the minimization of the sum of the delay and the delay risk value as the optimization objective, considering the task offloading constraint and the delay constraint, through the first-order moment and second-order moment distributionally robust fuzzy set method, the delay problem is approximated into a semidefinite programming problem (SDP). A delay optimization offloading algorithm improved based on the branch and bound method is proposed to rationally allocate the task offloading problem, and a better service is provided for users by achieving a lower offloading delay in edge computing. Finally, the value of the confidence level β is specified, and the high delay risk is avoided by changing the weight of the conditional value at risk (CVaR).

[0065] The present invention will be further described below with reference to the accompanying drawings.

[0066] As Figure 1 shown, the present invention provides a two-stage distributionally robust optimization task offloading method for an industrial Internet considering conditional risk value, including the following steps:

[0067] Step S1: Taking the tasks of the industrial Internet and the operating conditions of the edge server as constraints. A scheduling model for task offloading is constructed with the goal of minimizing the delay risk of the industrial Internet.

[0068]

[0069] Where x is the offloading decision combination vector, that is, x = (1 - x 1,k ,..., 1 - x |I|,k )T. If x i,k = 1, it means that task i is offloaded to the MEC for processing. If x i,k = 0, it means that task i is processed locally. E P [Q(x, ε)] represents the expected time for the task to be processed on the MEC, and CVaR β[Q(x, ε)] is used to characterize the latency risk. I represents the set of tasks, Q(x, ε) represents the task latency under the offloading decision combination vector x in the industrial Internet scenario ε, and λ ∈ [0, 1] is the latency risk weight, and the value of λ is adjusted to balance the average latency and risk.

[0070] Step S2: Construct an uncertainty set based on the first moment and second moment for the uncertain quantities in the model.

[0071] Due to the uncertainty of the wireless channel and the set of MECs that can be offloaded during the offloading process, it is difficult to determine the computing time on the edge device. Some historical data is used to predict the mean and variance of the uncertain variables, and then an uncertainty set is constructed using the first moment and second moment.

[0072]

[0073] where is a measure space the set of all probability measures, B is the Borel σ - algebra (Borel algebra) of is denoted by and represents the standard deviation of and are parameters that control the size of the fuzzy set. represents a semidefinite constraint ( then B - A being a positive definite matrix), and ∑ t are certain estimates of the mean and covariance matrix of the random vector t E respectively.

[0074] Step S3: Take the offloading decision in the industrial Internet as the decision variable in the first stage, and take the optimization variables of the latency risk sum other than the offloading decision as the decision variables in the second stage. With the minimum latency risk sum as the optimization objective and the relevant operating conditions and uncertainty set as the constraints, construct a two - stage distributionally robust model considering conditional value at risk.

[0075]

[0076] s, t

[0077]

[0078] where the objective function is the sum of the latency of local computing of the first - stage tasks, the running time of offloading to the edge device in the second stage, and a conditional value at risk, with the relevant operating conditions and uncertainty set as the constraints. μt It is an auxiliary variable introduced to transform the problem form. The operating conditions here include the 0-1 variable for offloading. A task can only be offloaded to one edge device, and each edge device has sufficient resources to handle the assigned tasks, etc., which will not be described here. The main discussion is on the objective function and the uncertainty set constraints.

[0079] Step S4: Use the branch and bound method to find the offloading decision variables in the first stage. Then transform the original problem into an SDP problem.

[0080] Since the objective value in the second stage is affected by the uncertainty set constraints, the problem in the second stage is mainly considered. Because both the objective function and the constraint conditions contain uncertain variables, and it is unknown whether the function is a convex function, it brings great difficulties to the solution. By introducing the dual variable z, z, Z, consider the dual problem of the second stage problem:

[0081]

[0082] Obviously, the inequality constraints in the dual problem are not easy to handle, and it is further transformed into a semidefinite programming problem:

[0083]

[0084] where is mainly defined for calculation convenience, y = (x 1,k , …, x |I|,k ) T is the offloading decision vector in the second stage, representing which MEC the task i is offloaded to for processing.

[0085] Determine the decision variable x in the first stage through the branch and bound method, and update the decision variable y in the second stage. The process is as Figure 2 shown, and the specific steps are as follows:

[0086] 6) First, without considering the offloading decision constraints of the original two-stage time delay problem, solve the corresponding relaxation problem, and record the result as T min .

[0087] 7) If the obtained T min just satisfies the binary constraint condition, then this solution is the optimal solution of the original problem.

[0088] 8) Branch. According to the priority P of the tasks described in step 1 i select the task Q i , construct two constraint conditions x i,k = 1 and x i,k = 0 representing whether the task is offloaded to the edge computing or calculated on the local device, and add these two constraint conditions to form two sub-problems of the original problem.

[0089] 9) Delimitation. First, solve whether there are feasible solutions for the two sub-problems. If there are, denote them as T 1 and T 2 . Judge the sizes of T 1 and T 2 . If T 1 <T 2 , then let T 1 = T min , otherwise, let T 2 = T min .

[0090] 10) If the solution of the sub-problem does not reach the optimal solution, or the offloading decision x i,k does not satisfy all being 0-1 variables, then it is necessary to continue with branch and bound (repeat steps 3) and 4)) until the offloading decision is all 0-1 variables. Finally, determine the first-stage decision variable x according to the offloading decision x i,k and update the second-stage decision variable y.

[0091] An embodiment of the present invention further includes a computer device, and the computer device has the execution program code or stored program code of the above-mentioned edge computing task offloading method considering conditional value at risk in an industrial Internet built therein.

[0092] The present invention combines Figures 3 to 4 to analyze and introduce the proposed two-stage distributionally robust optimization task offloading method for an industrial Internet considering conditional risk value.

[0093] During the verification process, the present invention sets the scale of the problem to 5*5, that is, 5 tasks need to be processed, and there are 5 edge devices at the same time. Randomly generate the positions where these 5 tasks are generated and the positions of the 5 devices within the unit square range, and calculate the time t_mec spent on offloading the random tasks to the edge devices, which is mainly used to estimate the mean and variance of t E . The data used are 10,000 independent samples randomly generated, and these samples come from the interval [0.5*t_mec, 1.5*t_mec]. Set the latest completion time for each task to 1 s. We assume that the computational amount w i and s i data sizes are generated by a probability distribution, and calculate the local computing time through these data.

[0094] Verify the influence on the expected offloading time by setting different λ and β, as Figure 3 shown. Through Figure 3It can be seen that by setting different risk parameters λ and confidence levels β, it can be observed that the expected time for a higher β value is always higher than that for a lower β value. Note that at a larger β value, the impact of λ on the expected cost is more significant. Since λ is a risk parameter, an increase in the parameter λ will lead to a higher level of risk aversion. To avoid a higher level of risk, it is often necessary to sacrifice some offloading time. Obviously, there is a trade-off between the mean and risk. Therefore, in the actual application process, the delay risk-sensitive parameters λ and confidence levels β can be carefully selected according to different delay requirements. The model proposed in the present invention not only considers the delay risk but can also be flexibly used according to different application scenarios. If the task does not need to consider a large risk, appropriate delay risk-sensitive parameters λ and confidence levels β can be selected to obtain a smaller average delay. If the task has a higher requirement for delay risk, the risk can be avoided by selecting λ and β.

[0095] Verify the impact on the expected offloading time by setting different uncertainty sets, as Figure 4 shown, by Figure 4 It can be seen that by setting different uncertainty sets, it is concluded that for a given λ and β, a larger family of distributions will provide a higher offloading time. This is because the worst-case scenario of a larger family of distributions will result in a worse situation than that of a relatively smaller family of distributions. If there is enough historical data, a more accurate uncertainty set can be characterized, and in this case, a better result (smaller delay risk and) can be obtained. Even if there is not enough historical data, under a series of constraints that satisfy the scenario, a relatively conservative result can be obtained by characterizing a larger uncertainty set.

[0096] The series of detailed descriptions listed above are only specific descriptions of the feasible implementation manners of the present invention, and they are not intended to limit the protection scope of the present invention. Any equivalent manners or changes that do not depart from the technology created by the present invention should be included in the protection scope of the present invention.

Claims

1. An edge computing task offloading method considering conditional value at risk in industrial Internet, characterized in that, it includes the following: S1. Establish a local computing model, an edge computing model and a delay risk value model of the computing task; S2. According to the models in S1, considering both the average performance and risk of delay, define the objective function as the sum of the average delay and its delay risk value, which is called the delay risk sum, and establish an objective function with minimizing the delay risk sum as the optimization goal; S3. Establish a fuzzy set of the first moment and the second moment, and construct a two-stage distributionally robust model. The first stage is the decision-making stage, which decides whether the task is computed locally or offloaded to the MEC for computing. The second stage is the MEC computing stage, with minimizing the delay risk sum as the optimization goal; S4. Transform the two-stage distributionally robust model in step S3 into a semidefinite programming problem; S5. Use the branch and bound method to find the optimal offloading scheme of the task and achieve the offloading decision; The local computing model in S1 is: Among them represents the local computing time of task Q i , f i l represents the CPU computing power of the local device, w i represents the amount of computation of task Q i ; The edge computing model in S1 is: Among them represents the edge computing time of task Q i including edge execution time and transmission delay represents the computing resources allocated by MEC, r i,k represents the user uplink data rate, s i represents the computing task Q i size; The establishment of the delay risk value model in S1: Introduce CVaR as the risk measure, and establish the delay risk value model as: Among them, α is the quantity to be optimized. In an ideal situation, when the optimization is completed, α is the conditional risk value corresponding to the confidence level β. is the expectation with respect to the probability distribution P; the objective function in S2 is: where \(x\) is the offloading decision combination vector, i.e., \(x=(1 - x 1,k ,\cdots,1 - x |I|,k ) T . If \(x i,k = 1\), it means that task \(i\) is offloaded to the MEC for processing. If \(x i,k = 0\), it means that task \(i\) is processed locally. \(E P [Q(x,\varepsilon)]\) represents the expected time for the task to be processed on the MEC. CVaR β [Q(x,\varepsilon)]\) is used to represent the delay risk. \(I\) represents the set of tasks, \(Q(x,\varepsilon)\) represents the task delay under the offloading decision combination vector \(x\) in the industrial Internet scenario \(\varepsilon\). \(\lambda\in[0,1]\) is the delay risk weight, and the value of \(\lambda\) is adjusted to balance the average delay and risk.

2. The edge computing task offloading method considering conditional value at risk in industrial Internet according to claim 1, characterized in that, the specific implementation of S3 includes: S3.1 Replace with Q(x, ε) according to the CVaR definition and denote the delay risk value as Denote the delay risk value as S3.2 The objective function is where t T represents the set of task local computing times, and t T x is the local computing time for the first-stage task to make an offloading decision; For the convenience of calculation, ε is equivalent to t E , both representing the computing time of the task on the MEC. Due to the uncertainty of the channel and the set of available MECs, t cannot be determined E , and a fuzzy set based on the first and second moments is established using historical data Among them is a measure space is the set of all probability measures, and B is the Borel σ - algebra of, and M is any closed convex set containing the support of the probability distribution P Denoted by the standard deviation of is and are parameters that control the size of the fuzzy set represents the semi - definite constraint and ∑ t are certain estimates of the mean and covariance matrix of the random vector t E respectively S3.4 Based on the fuzzy set of the first moment and the second moment, construct a two-stage distributionally robust model for task offloading:

3. The edge computing task offloading method considering conditional value at risk in industrial Internet according to claim 2, characterized in that, it also includes changing the distributionally robust model into 4. The edge computing task offloading method considering conditional value at risk in industrial Internet according to claim 3, characterized in that, the specific implementation of S4 is as follows: S4.1 Define intermediate variables Since in (8) is an infinite-dimensional optimization problem with probability distribution P, the objective function and the first and second moment constraints of the second stage are established as follows: S4.2 Introduce the dual variable z of the constraint (9), z , Z, where indicates that z is a real number, indicates and z is an m-dimensional real vector, indicates that Z is an m-dimensional symmetric positive semi-definite matrix, and consider the dual problem of the problem S4.3 Transform the dual problem into a tractable semidefinite programming problem: The intermediate variables defined therein y = (x 1,k , …, x |I|,k ) T is the offloading decision vector in the second stage, representing which MEC the task i is offloaded to for processing.

5. The edge computing task offloading method considering conditional value at risk in industrial Internet according to claim 1, characterized in that, the implementation of S5 is to find the first-stage offloading decision variable x and update the second-stage offloading decision y. The specific process is: S5.1 First, without considering the offloading decision constraints of the original two-stage time delay problem, solve the corresponding relaxed problem, and denote the result as T min ; S5.2 If the obtained T min exactly satisfies the binary constraint conditions, then this solution T min is the optimal solution to the original problem; S5.3 Branching; according to the priority P of the task i Select task Q i , construct two constraint conditions x i,k = 1 and x i,k = 0 represent that the task is offloaded to the edge computing and computed on the local device respectively. Add these two constraint conditions to form two sub-problems of the original problem; S5.4 Delimitation; First, solve whether there are feasible solutions for the two sub-problems. If there are, denote them as T 1 and T 2 , determine T 1 and T 2 's magnitudes. If T 1 < T 2 , then let T 1 = T min , otherwise, let T 2 = T min ; S5.5 If the solution of the sub-problem does not reach the optimal solution, or the offloading decision x i,k does not satisfy all being 0-1 variables, then it is necessary to continue with branch and bound until the offloading decision all satisfies 0-1 variables. Finally, according to the offloading decision x i,k determine the decision variables x in the first stage and update the decision variables y in the second stage.

6. A computer device, characterized in that, the computer device stores an execution code, and the execution code can be executed by a processor to implement the edge computing task offloading method considering conditional value at risk in industrial Internet according to any one of claims 1-5.

Images