Software-defined vehicle-mounted task offloading migration planning method, SDN controller and deployment method

Abstract

The present invention discloses a software-defined vehicle-mounted task offloading migration planning method based on distributed robustness, an SDN controller and a deployment method, including the following steps: 1. Obtaining information about vehicle tasks that can be accessed to RSUs, information about vehicle-mounted tasks, etc.; 2. Obtaining the optimal set of k paths between the starting point and the end point of the vehicle-mounted task based on the currently available data between RSUs; 3. Converting the vehicle-mounted task offloading migration planning problem into a mathematical problem, and transforming the original model; 4. Using the branch-and-bound implicit enumeration method to solve the one-stage planning problem, and then using semidefinite programming to solve the two-stage distributed robustness problem; 5. Deploying the algorithm to the controller in the software-defined vehicle-mounted network. The present invention fully considers the task transmission status of each vehicle-mounted task in the edge computing network, and minimizes the time cost and energy consumption cost of the vehicle-mounted task offloading migration planning while ensuring that the vehicle-mounted task is completed within the specified time and the transmission energy consumption is limited.

Description

Technical Field

[0001] The present invention belongs to the field of vehicle-mounted mobile edge computing, and is a vehicle-mounted task offloading migration planning method in a software-defined network environment, and is particularly suitable for executing delay-sensitive tasks in a local area network and reducing the energy consumption of executing tasks in a small base station environment. Background Art

[0002] With the development of modern science and technology, the emergence of the Internet of Things has brought up the concept of Internet of Vehicles (IoV). The emergence of various in-vehicle applications, such as autonomous driving and heterogeneous in-vehicle networks, has facilitated people's lives, but at the same time, these applications have certain requirements for data collection, transmission speed, and server computing power. Based on cloud computing, the concept of mobile edge computing was proposed, which has developed into vehicle edge computing (VEC) in the vehicle environment.

[0003] In the vehicle edge network, vehicle applications effectively complete tasks by seeking computing resources from nearby base stations and edge computing nodes, usually roadside nodes (RSU), which is called task offloading. Figure 1 As shown in the figure. With the continuous increase and development of vehicle applications, people are more sensitive to task delays and hope to save as much time as possible on task transmission. The current mainstream migration selection scheme is to greedily allocate tasks according to the status of the current node or the status of local nodes in the same time period after arriving at a certain node in real time. This scheme can achieve better efficiency when there are fewer tasks, but when traffic congestion occurs, local tasks increase dramatically and the global optimal solution cannot be obtained.

[0004] The present invention mainly studies the acquisition of information such as the current task distribution, data transmission status and channel congestion status of the vehicle edge network through software defined networking (SDN), and on this basis combines the linear programming model to construct an adaptive optimization decision through distributional robust optimization (DRO) to give the optimal global unloading migration planning decision.

[0005] In summary, a DRO-based task offloading migration planning method can be constructed in SDN to cope with the vehicle-mounted task offloading scenario and consider various influencing factors. Summary of the invention

[0006] In view of the above problems, the present invention proposes a distributed robust software-defined vehicle task offloading migration planning method, such as Figure 3As shown, in order to solve the problems of high latency, high energy consumption and low reliability caused by vehicle-mounted task unloading, the method includes the following steps:

[0007] Step 1: Get the set M of RSUs that the vehicle can access, the set N of vehicle task sizes requested to be unloaded in the RSU area, the RSU node distance D, the unloading task starting RSU set S and target set O, the network bandwidth B between RSUs, and the task completion constraint time window L time , task execution energy consumption constraint L energy , limit the number of tasks through the channel L channel ;

[0008] Step 2: Calculate the optimal k paths for the vehicle-borne task to be transmitted in the edge network based on the RSU information in step 1

[0009] Step 3: Convert the onboard task offloading migration decision method into a mathematical problem;

[0010] Step 4. Use planning to solve the math problem in step 3.

[0011] Further, the information in step 1 includes:

[0012] ① The set of RSUs that vehicles can access within the coverage of SDN is M = {m1,…m i ,…m M};

[0013] ② The size of the relevant information uploaded by the vehicle task α, the size of the relevant information downloaded β, and the task workload ω,

[0014] n i =(α,β,ω), that is, the uploaded vehicle task set is N = {n1,…n i ,...n N}

[0015] ③The distance between RSUs is defined as

[0016] ④ On-board task unloading, the starting point RSU node set is denoted as S = {s1,…s i ,…s N}; The target RSU node set is defined as O = {o1, ...o i ,…o N};

[0017] ⑤The bandwidth of RSU is recorded as where b ij represents the network bandwidth between nodes i and j;

[0018] ⑥ The end time constraint of the vehicle task execution is recorded as The energy consumption constraint during vehicle mission execution is denoted as The number of tasks that pass through the channel is limited, denoted by Furthermore, the optimal k paths between every two RSUs obtained in step 2 are:

[0019] Step 2.1: Collect the average signal power of RSU, recorded as P; collect the noise power of RSU, recorded as Np; record the link loss power between RSU and vehicle as L p ;

[0020] Step 2.2, the transmission delay of the onboard task is determined by the bandwidth B of the RSU and the task offloading planning The size of the onboard task is determined by M, and the transmission rate V of the onboard task in the offloading network can be expressed as:

[0021]

[0022] Where, d (km) is the distance between the vehicle and the RSU or the distance between the RSUs, and f (MHz) is the signal frequency of the RSU.

[0023] Step 2.3, here calculate the time required for a unit size task to be transmitted in a certain transmission path, that is:

[0024]

[0025] Step 2.4, use the algorithm to find the optimal k paths between RSU nodes, such as Figure 2 As shown, the algorithm flow is as follows:

[0026] ① Use Dijkstra algorithm to find task i, from source point s i To the end point i The shortest path is denoted as R i1 ,

[0027] Let j=2.

[0028] ② Place the ij-1 Each node on (except the end point d i ) are regarded as deviation points (total |E ij-1 |-1),

[0029] Let each deviation point be v j-1p (p=1,2,…,|E ij-1 |-1).

[0030] ③ Find v j-1p To the end point i The second shortest path.

[0031] ④Splicing R ij-1From the starting point to v j-1p The path and v j-1p The second shortest path to the end point d is obtained, and the length of the path and the energy consumption through the path are calculated as R ij The candidate paths are put into the candidate path set S, and are sorted from small to large by comprehensively considering the time and energy consumption of the unit size task through the path.

[0032] ⑤Loop ②, ③ and ④ to traverse all deviation points.

[0033] ⑥ Determine whether S is empty:

[0034] a) If S is empty, the algorithm ends;

[0035] b) If S is not empty, select the path with the shortest length from S to obtain R ij , remove it from S;

[0036] ⑦ Set i=i+1, and determine whether i≤k:

[0037] a) The inequality holds, return to step ②;

[0038] b) The inequality does not hold and the algorithm ends;

[0039] ⑧ Where R ij represents the jth optional path to complete task i. The set of edges in the path is denoted by E; the time it takes for a unit size task to pass through the path is denoted by Length; and the energy consumed by passing through the path is denoted by Energy. In actual operation, The matrix is ​​split into three matrices. R is a matrix composed of tuples. Each tuple contains three members: E, Length, and Energy. They are abbreviated as: E, Len, and En, that is, R = (E, Length, Energy). The Length matrix can be expressed by the formula Length = dot (C unittime ,E) is obtained, dot is the product of the corresponding positions of the matrix.

[0040] Furthermore, in step 3, the onboard task offloading migration decision method is mathematically modeled:

[0041] Step 3.1, determine the one-stage offloading decision model of the vehicle upload task within the SDN coverage area. Since all parameters are fixed, the route planning choices for the starting point and the end point are the same. Define X as the vehicle task offloading path planning matrix, X = (x1,…x i ,…x |N| ) T , x i represents the task offloading path planning vector selected by task i, x iis a unit vector containing only 0 and 1, and ||x i ||2=1,||X||2=|N|,X∈R k ×R |N| , |N| represents the number of tasks, k is the optimal k paths, and the one-stage offloading decision model is as follows:

[0042]

[0043]

[0044] C time Offloading the path time cost for the vehicle-borne task, C energy Offloading path energy costs for onboard tasks, and Respectively represent the time cost of task uploading and task downloading, is the average waiting time, The time cost of task execution, and They represent the time and energy consumption of task uploading and task downloading, respectively, and C computing Calculate the resource vector for the RSU node, Len forward is the optional path distance when uploading the task, E ij represents the channel occupancy vector of task i selecting the kth path, λ time is the weight of time cost in the objective function, λ energy is the weight of the energy consumption window in the objective function, and λ energy +λ time =1.

[0045] tr() is used to find the trace of the corresponding matrix; diag() is used to find the vector composed of the elements on the main diagonal of the matrix; e and They represent the execution rate and arrival rate of tasks offloaded to the RSU node respectively.

[0046] Step 3.2, the objective function of the second stage means that when the starting point RSU and the target point RSU of the vehicle task are determined during the transmission process, the time cost and energy consumption cost of uploading, calculating and downloading the vehicle task are combined according to the weight λ to minimize the overall cost. In actual situations, the transmission rate of the vehicle task in the channel and the processing rate at the RSU node will be affected by the real-time signal-to-noise ratio and the computing resource status of the RSU node, respectively. At the same time, the path selection that originally meets the constraints may be affected by the change of the remaining time. Therefore, it is necessary to re-plan a reasonable return route.

[0047] X is the path selected for task upload obtained in the first stage planning, and Y represents the path selected for task return in the second stage. X,Y∈Rk ×R N ,||X||=||Y||=|N|.

[0048] First, the cost of transmitting the vehicle-mounted task in SDN is fully represented by the model, that is, the objective function is:

[0049]

[0050] Then, the two-stage decision model is designed as a distributionally robust optimization model under moment uncertainty:

[0051]

[0052] with h(x)=λ time ×α×tr(Len forward ×X)+λ energy ×tr(En forward ×X),

[0053]

[0054]

[0055] Among them Len forward is the time matrix of the unit size task through the optional upload path, En forward is the energy consumption matrix of unit size tasks through the optional upload path, En back is the energy consumption matrix of unit size tasks through the optional download path, To obtain the expectation of the parameters, is the probability distribution of the random parameter x.

[0056] Step 3.3, due to the existence of uncertain parameters, That is, the time required for the task to return and the time spent on task calculation, so fuzzy sets are needed to represent them.

[0057] To facilitate the representation of variables, the uncertainty of the time spent on the vehicle task due to interference caused by the signal-to-noise ratio or other factors when selecting the path is recorded as The same RSU may also have certain uncertainties due to system occupancy and equipment problems, so it is recorded as A ij is a matrix with all zeros except the i-th row and j-th column which are 1; ζ i is a unit vector whose i-th bit is 1.

[0058] x=(x1,…x k ,…x i×k+1 ,...xn×k ,x n×k+1 ,...x (n+1)×k )

[0059]

[0060] The first constraint states that the sum of the probability densities of the desired parameter within the range of possible values ​​is 1; the second constraint assumes that the mean of x lies on an ellipsoid of size γ1, centered at the moment estimate μ0; the third constraint requires that the covariance matrix lies in a semi-positive definite cone defined by a matrix inequality. In other words, it describes how likely x is to be close to μ0 under the condition of ∑ as the correlation.

[0061] Step 3.4, organize the model content of the second-stage planning optimization and express it in a simple form to facilitate subsequent operations:

[0062]

[0063] with h(x)=λ time ×(α×tr(Len forward ×X)+λ energy ×(β×tr(En forward ×X),

[0064]

[0065]

[0066]

[0067]

[0068] The parameters are described as follows:

[0069] ①x is a random parameter vector; G is any closed convex set known to contain the random vector x; P is the possible probability distribution of the unknown parameter x; and ∑ represent the expectation and variance of the obtained parameters respectively; γ1≥0 and γ2≥0 are two control parameters;

[0070] ②0≤λ is a trade-off coefficient, representing the risk aversion rate of the decision maker. The larger the coefficient λ, the more risk-averse the decision maker is;

[0071] ③CVaR (Conditional Value at Risk) is a risk measurement technique that is better than VaR. It means that the consumption of the selected unloading countermeasure combination exceeds the average consumption value under a given VaR value.

[0072] Right now:

[0073]

[0074] Where φ is the confidence interval.

[0075] Step 3.5, use Lagrangian duality to process the second stage model in the original problem:

[0076] ① Define the Lagrangian function. The original problem can be transformed into the following semi-infinite conic linear problem:

[0077]

[0078]

[0079] in

[0080] ②Define Lagrangian Duality:

[0081]

[0082]

[0083] Among them, r is the dual variable of the equality constraint, H is the dual variable of the inequality constraint, is the dual variable of the second-order moment inequality constraint.

[0084] ③ Function simplification:

[0085]

[0086]

[0087] Where s0 indicates that the matrix is ​​semidefinite.

[0088] Further, the specific steps of step 4 are:

[0089] Step 4.1, solve the one-stage decision optimization problem. The one-stage problem is a relatively simple 01 integer programming problem, so it can be solved by implicit enumeration, which is a special branch and bound method.

[0090] 1) Convert the one-stage unloading decision model into the standard form of linear programming;

[0091] 2) Branch and bound: Planning problems in standard form prioritize x ij = 0, randomly select a variable value that may be better and meets the constraints, find the value of the objective function, and determine whether to perform a branch operation based on the following principles:

[0092] ① When the sub-problem of the current branch is a feasible solution, stop branching, keep the branch with the smallest objective function value obtained by the current feasible solution, and delete the branch with a larger boundary value;

[0093] ② Regardless of whether it is a feasible solution or not, as long as the boundary value of the current branch is greater than the objective function value of the feasible solution obtained, the branch is stopped;

[0094] ③ When some variables are fixed, the branch will be stopped directly if one constraint condition is not met;

[0095] 3) Branch operation: Determine one of the free variables as a fixed variable. The branch has two options: set the fixed variable to 1 or 0;

[0096] 4) When all other branches are deleted, the feasible solution that remains is the optimal solution of this model;

[0097] Step 4.2, according to the obtained one-stage decision model, substitute it into the two-stage decision model to solve the two-stage planning problem. The two-stage problem is a distributed robust vehicle task offloading migration planning model based on CVaR constraints.

[0098] In step 3, the original complex and difficult-to-solve DRO problem has actually been converted into a semi-definite programming (SDP) problem, which can be solved.

[0099] Based on the above method, the present invention further proposes an SDN controller, in which the above method is deployed. The specific operation of deploying the algorithm to the SDN controller is:

[0100] The SDN data layer uploads the current data to the SDN control layer. After the offloading balancing decision is determined, the optimal solution is obtained through the above-established model according to the current status of the RSU and the channel status between the nodes. The allocation information is transmitted to the application layer, and the vehicle-mounted task is offloaded and the result is returned according to the given route planning.

[0101] The beneficial effects of the present invention are:

[0102] The present invention aims to solve the defects of the prior art by planning the unloading route of vehicle-mounted tasks in the edge computing network and completing the omissions of the edge computing network in task transmission. At the same time, distributed robust optimization is used to estimate the changes in the edge computing environment during task transmission and execution, so as to make the best decision. BRIEF DESCRIPTION OF THE DRAWINGS

[0103] Figure 1 Onboard task offloading planning environment diagram

[0104] Figure 2 Flowchart for solving the optimal K paths

[0105] Figure 3 On-board task migration planning flow chart DETAILED DESCRIPTION

[0106] The present invention is further described below in conjunction with the accompanying drawings. It should be noted that the specific implementation of this example is based on the present technology, and a detailed implementation process and implementation steps are given, but the protection scope of the present invention is not limited to this implementation example.

[0107] Assuming that the vehicle is ready to unload the onboard task set N at this time, the specific implementation method according to the present invention is as follows:

[0108] (1) Use the SDN controller in step 1. The set M of RSUs in each LAN, the vehicle task N that is requested to be unloaded in the RSU area, the adjacent RSU and target RSU nodes for vehicle task unloading, the network bandwidth B of the RSU, and the instant computing capacity C of the RSU computing When information changes, it can be summarized in the SDN controller;

[0109] (2) Based on the onboard task offloading information obtained in step 1 and the channel information between RSUs, the optimal K paths from the starting RSU to the target RSU for each task are obtained according to the optimal K paths algorithm in step 2;

[0110] (3) The SDN controller aggregates the vehicle unloading task set N, according to the formula in step 3

[0111]

[0112] Transform the onboard task offloading migration planning problem into a mathematical problem to be solved;

[0113] According to step 4, the implicit enumeration method with branch and bound property is used to solve the one-stage 0-1 integer programming problem of path planning selection and the two-stage CVaR-constrained distributed robust vehicle task offloading migration planning problem using semidefinite programming.

[0114] According to step 5, the algorithm is deployed to the SDN controller.

[0115] The series of detailed descriptions listed above are only specific descriptions of feasible implementation methods of the present invention. They are not intended to limit the scope of protection of the present invention. All equivalent methods or changes that do not deviate from the technical creation of the present invention should be included in the scope of protection of the present invention.

Claims

1. A distributed robust software-defined vehicle task offloading migration planning method, characterized in that: These include: S1. Obtain relevant information: the set of RSUs that the vehicle can access M, the set of vehicle task sizes N requested to be unloaded in the RSU area, the RSU node distance D, the unloading task starting RSU set S and target set O, the network bandwidth B between RSUs, and the task completion constraint time window L time , task execution energy consumption constraint L energy , limit the number of tasks through the channel L channel ; S2, based on the information in S1, calculate the optimal set of k paths for the vehicle-borne task to be transmitted in the edge network S3, mathematically modeling the onboard task offloading migration decision method; S4, use the model in S3; in, The optimal k paths in S2 The calculation methods include: S2.1, collect the average signal power of RSU, recorded as P; collect the noise power of RSU, recorded as Np; record the link loss power between RSU and vehicle as L p ; S2.2, the transmission delay of the onboard task is determined by the bandwidth B of the RSU and the task offloading planning The size of the on-board task M determines the transmission rate V of the on-board task in the offloading network, which can be expressed as: Where d (km) is the distance between the vehicle and the RSU or the distance between the RSUs, and f (MHz) is the signal frequency of the RSU; S2.3, calculate the time required for a unit size task to be transmitted in a certain transmission path, that is: S2.4, find the optimal k paths between RSU nodes, the calculation method is as follows: ① Use Dijkstra algorithm to find task i, from source point s i To the end point i The shortest path is denoted by R i1 , let j = 2; ② Place the ij-1 Each node on (except the end point d i ) are regarded as deviation points (total |E ij-1 |-1), and each deviation point is recorded as v j-1p (p=1,2,…,|E ij-1 |-1); ③ Find v j-1p To the end point i The second shortest path; ④Splicing R ij-1 From the starting point to v j-1p The path and v j-1p The second shortest path to the end point d is obtained, and the length of the path and the energy consumption through the path are calculated as R ij The candidate paths are put into the candidate path set S, and are sorted by comprehensively considering the time and energy consumption of the unit size task through the path; ⑤Loop ②, ③ and ④ to traverse all deviation points; ⑥ Determine whether S is empty: a) If S is empty, the algorithm ends; b) If S is not empty, select the path with the shortest length from S to obtain R ij , remove it from S; ⑦ Set i=i+1, and determine whether i≤k: a) The inequality holds, return to step ②; b) The inequality does not hold and the algorithm ends; Where R ij It represents the jth optional path to complete task i. The set of edges in the path is recorded as E. The time it takes for a unit size task to pass through the path is recorded as Length. The energy consumption through the path is recorded as Energy. In actual operation, The matrix is ​​split into three matrices and abbreviated as: E, Len, En, that is, R = (E, Length, Energy), where the Length matrix can be obtained by the formula Length = dot (C unittime ,E) get, dot is the product of the corresponding position of the matrix; In S3, a vehicle-mounted task offloading migration decision method is mathematically modeled, including: S3.1, determine the one-stage offloading decision model of the vehicle upload task within the SDN coverage area; since all parameters are fixed, the route planning choices for the starting point and the end point are the same, and define X as the vehicle task offloading path planning matrix, X = (x1,…x i ,…x |N| ), x i represents the task offloading path planning vector selected by task i, x i is a unit vector containing only 0 and 1, and ||x i ||2=1,||X||2=|N|,X∈R |N| ×R k , |N| represents the number of tasks, k is the optimal k paths, and the one-stage offloading decision model is as follows: C time Offloading the path time cost for the vehicle-borne task, C energy Offloading path energy costs for onboard tasks, and Respectively represent the time cost of task uploading and task downloading, is the average waiting time, The time cost of task execution, and They represent the time and energy consumption of task uploading and task downloading, respectively, and C computing Calculate the resource vector for the RSU node, Len forward is the optional path distance when uploading the task, E ij represents the channel occupancy vector of task i selecting the kth path, λ time is the weight of time cost in the objective function, λ energy is the weight of the energy consumption window in the objective function, and λ energy +λ time =1; tr() is used to find the trace of the corresponding matrix; diag() is used to find the vector composed of the elements on the main diagonal of the matrix; e and They represent the execution rate and arrival rate of tasks offloaded to the RSU node respectively; The S3 mathematically models the vehicle-mounted task offloading migration decision method, and also includes: Step 3.2, the objective function of the second stage means that when the starting point RSU and the target point RSU of the vehicle task are determined during the transmission process, the time cost and energy consumption cost of uploading, calculating and downloading the vehicle task are combined according to the weight λ to minimize the overall cost. In actual situations, the transmission rate of the vehicle task in the channel and the processing rate at the RSU node are affected by the real-time signal-to-noise ratio and the computing resource status of the RSU node respectively. At the same time, the path selection that originally meets the constraints may be affected by the change of the remaining time. Therefore, it is necessary to re-plan a reasonable return route: Let X be the path selected by the task upload obtained in the first stage planning, and Y be the path selected when the task returns in the second stage, X,Y∈R |N| ×R k ,||X||=||Y||=|N|, First, the cost of transmitting vehicle-mounted tasks in SDN is fully represented by a model: Then, the two-stage decision model is designed as a distributed robust optimization model under moment uncertainty: with h(x)=λ time ×α×tr(Len forward ×X)+λ energy ×tr(En forward ×X), Among them Len forward is the time matrix of the unit size task through the optional upload path, En forward is the energy consumption matrix of unit size tasks through the optional upload path, En back is the energy consumption matrix of the unit size task through the optional download path, To obtain the expectation of the parameters, is the probability distribution of the random parameter ξ; The S3 mathematically models the vehicle-mounted task offloading migration decision method, and further includes: Step 3.3, for uncertain parameters That is, the time required for the task to return and the time spent on task calculation, using fuzzy sets to represent them; To facilitate the representation of variables, the uncertainty of the time spent on the vehicle task due to interference caused by the signal-to-noise ratio or other factors when selecting the path is recorded as The same RSU may also have certain uncertainties due to system occupancy and equipment problems, so it is recorded as A ij is a matrix with all zeros except the i-th row and j-th column which are 1; ζ i is a unit vector with the i-th bit being 1, ξ=(ξ1,…ξ k ,…x i×k+1 ,…x n×k ,x n×k+1 ,…x (n+1)×k ) The first constraint states that the sum of the probability densities of the desired parameter within the range of possible values ​​is 1; the second constraint assumes that the mean of ξ lies on an ellipsoid of size γ1, centered at the moment estimate μ0; the third constraint requires that the covariance matrix lies in a semi-positive definite cone defined by a matrix inequality, which describes how likely ξ is to be close to μ0 under the condition of ∑ as the correlation; The S3 mathematically models the vehicle-mounted task offloading migration decision method, and also includes: Step 3.4, organize the model content of the second-stage planning optimization and express it in a simple form to facilitate subsequent operations: with h(x)=λ time ×(α×tr(Len forward ×X)+λ energy ×(β×tr(En forward ×X), The parameters are described as follows: ①ξ is a random parameter vector; G is any closed convex set known to contain the random vector ξ; P is the possible probability distribution of the unknown parameter ξ; and ∑ represent the expectation and variance of the obtained parameters respectively; γ1≥0 and γ2≥0 are two control parameters; ②0≤λ is a trade-off coefficient, which represents the risk aversion rate of the decision maker. The larger the coefficient λ, the more risk-averse the decision maker is. ③CVaR (Conditional Value at Risk) is a risk measurement technique that is better than VaR. It means that the consumption of the selected unloading countermeasure combination exceeds the average consumption value under a given VaR value. Right now: Where φ is the confidence interval; Step 3.5, use Lagrangian duality to process the second stage model in the original problem: ① Define the Lagrangian function. The original problem can be transformed into the following semi-infinite conic linear problem: in ②Define Lagrangian Duality: Among them, r is the dual variable of the equality constraint, H is the dual variable of the inequality constraint, is the dual variable of the second-order moment inequality constraint; ③ Function simplification: in Indicates that the matrix is ​​semidefinite; The specific steps of S4 are: S4.1, solve the one-stage decision optimization problem. The one-stage problem is a relatively simple 01 integer programming problem, which is solved by implicit enumeration method, which is a special branch and bound method; 1) Convert the one-stage unloading decision model into the standard form of linear programming; 2) Branch and bound: Planning problems in standard form prioritize x ij = 0, randomly select a variable value that may be better and meets the constraints, find the value of the objective function, and determine whether to perform a branch operation based on the following principles: ① When the sub-problem of the current branch is a feasible solution, stop branching, keep the branch with the smallest objective function value obtained by the current feasible solution, and delete the branch with a larger boundary value; ② Regardless of whether it is a feasible solution or not, as long as the boundary value of the current branch is greater than the objective function value of the feasible solution obtained, the branch is stopped; ③ When some variables are fixed, the branch will be stopped directly if one constraint condition is not met; 3) Branch operation: Determine one of the free variables as a fixed variable. The branch has two options: set the fixed variable to 1 or 0; 4) When all other branches are deleted, the feasible solution that remains is the optimal solution of this model; In step 4.2, the first-stage decision model is substituted into the second-stage decision model to solve the second-stage planning problem. The second-stage problem is a distributed robust vehicle task offloading migration planning model based on CVaR constraints. The original complex and difficult-to-solve DRO problem is converted into a semi-definite programming (SDP) problem to achieve the solution.

2. According to claim 1, a distributed robust software-defined vehicle task offloading migration planning method is characterized in that: The specific information in S1 is as follows: ① The set of RSUs that vehicles can access within the coverage of SDN is M = {m1,…m i ,...m |M| }; ② The size of the relevant information uploaded by the vehicle task α, the size of the relevant information downloaded β, and the task workload ω, n i =(α,β,ω), that is, the uploaded vehicle task set is N = {n1,...n i ,…n |N| } ③The distance between RSUs is defined as ④ On-board task unloading, the starting point RSU node set is denoted as S = {s1,…s i ,…s |N| }; The target RSU node set is defined as O = {o1, ...o i ,…o |N| }; ⑤The bandwidth of RSU is recorded as where b ij represents the network bandwidth between nodes i and j; The end time constraint of the vehicle task execution is denoted as The energy consumption constraint during vehicle mission execution is denoted as The number of tasks that pass through the channel is limited, denoted by 3. A deployment method of a vehicle-mounted task offloading migration planning method of an SDN controller, characterized in that: The vehicle-mounted task offloading migration planning method is a distributed robust software-defined vehicle-mounted task offloading migration planning method according to any one of claims 1-2, and the deployment method is as follows: The SDN data layer uploads the current data to the SDN control layer. After determining the offloading balancing decision, the optimal solution is obtained through the established mathematical model based on the current status of the RSU and the channel status between the nodes. The allocation information is then transmitted to the application layer. The on-board tasks are offloaded and the results are returned according to the given route planning.

Images