CN117156493A - QoE-based vehicle networking task unloading and resource allocation method - Google Patents

Abstract

The invention discloses a QoE-based vehicle networking task unloading and resource allocation method, which comprises the following steps: firstly, constructing a vehicle networking edge calculation VEC system model according to the calculation and communication capacity of a vehicle end; based on the VEC system, the vehicle may communicate with the road side unit RSU through V2I technology, such that the computing tasks may be offloaded to the RSU for execution; by introducing a task unloading decision, vehicle transmission power, vehicle communication bandwidth and RSU computing resource constraint, the QoE problem of the maximized system is constructed; according to the problem structure, the problem is decomposed into a resource allocation problem for fixed task unloading decision and a task unloading problem for optimizing the task corresponding to the resource allocation problem, the resource allocation problem is solved by using a gradient iteration method, and a simulated annealing algorithm is selected to achieve the maximization of the system QoE, so that the problem solving time is remarkably reduced.

Description

QoE-based vehicle networking task unloading and resource allocation method

Technical Field

The invention belongs to the field of internet of vehicles edge computing, and particularly relates to an internet of vehicles task unloading and resource allocation method based on QoE.

Background

With the rapid development of intelligent traffic systems, the rapid development of internet of vehicles applications such as automatic driving, high-precision map navigation and the like is greatly facilitated for our travel life. However, the rapid development of applications is accompanied by a heavy computational task. The vehicle side has limited computing and communication resources, so it is necessary to study efficient real-time task offloading and resource allocation in the vehicle network. In recent years, cloud computing has rapidly developed, and due to huge computing resources in the cloud, some researches propose to move a vehicle task to the cloud for execution. However, the cloud server is usually far away from the vehicle end, and a long transmission distance can bring a great task time delay, so that the requirement of real-time task processing cannot be met. To address these challenges, some research has considered moving edge calculations into vehicle networking systems, with internet of vehicles edge calculations occurring. A large number of servers are deployed at the RSU end, and the vehicle can transmit the task to the RSU end for processing through V2I communication, so that the problem of weak calculation capability of the vehicle section is solved, and the real-time performance of the task can be met. However, due to the limited RSU resources, the limited communication distance between the vehicles and the RSU, and the time delay and energy consumption generated by task unloading, when more vehicles are unloaded at the same time, the execution efficiency of the task may be significantly affected, and even the task execution may fail. Therefore, under the condition of meeting the real-time performance, the research on the problem of vehicle network task unloading and resource allocation has important practical significance.

For the internet of vehicles edge computing system, there has been a great deal of research considering internet of vehicles computing offloading. At present, in the research of internet of vehicles edge computing task offloading and resource allocation, many research results have been achieved, but the following problems still exist:

1) In the problems of task unloading and resource allocation, most researches are about considering the problems of task unloading or resource allocation, and the researches on collaborative optimization are less;

2) In the QoE-based task offloading method, most researches only consider computation delay and do not consider computation and communication energy consumption, but the energy consumption is not negligible in task offloading and resource allocation;

3) Aiming at the vehicle networking edge computing platform, the task offloading and resource allocation problems based on the QoE of the vehicle user have higher computational complexity, and the problems need to be decomposed and the system utility is maximized by using a low-computational complexity algorithm.

Disclosure of Invention

In order to solve the problems, the invention discloses a quality of experience (QoE) -based vehicle networking task unloading and resource allocation method, which comprises the steps of firstly constructing a vehicle networking edge computing (VEC) system model according to computing and communication capabilities of a vehicle end; based on the VEC system, the vehicle may communicate with a Road Side Unit (RSU) through V2I technology so that computing tasks may be offloaded to the RSU for execution; by introducing a task unloading decision, vehicle transmission power, vehicle communication bandwidth and RSU computing resource constraint, the QoE problem of the maximized system is constructed; according to the problem structure, the problem is decomposed into a resource allocation problem for fixed task unloading decision and a task unloading problem for optimizing the task corresponding to the resource allocation problem, the resource allocation problem is solved by using a gradient iteration method, and a simulated annealing algorithm is selected to achieve the maximization of the system QoE, so that the problem solving time is remarkably reduced.

In order to achieve the above purpose, the technical scheme of the invention is as follows:

a QoE-based vehicle networking task unloading and resource allocation method mainly comprises the following steps:

(1) Building a vehicle networking edge calculation VEC system model according to the vehicle end calculation and communication capability;

(2) Based on the VEC system, the vehicle can communicate with the road side unit RSU through the wireless communication technology V2I technology, so that the calculation task can be unloaded to the RSU for execution;

(3) By introducing a task unloading decision, vehicle transmission power, vehicle communication bandwidth and RSU computing resource constraint, the QoE problem of the maximized system is constructed;

(4) According to the problem structure, constraint conditions are mutually independent, so that the problem is resolved, the problem of resource allocation is solved by using a gradient iteration method, and the maximum system QoE is realized by using a simulated annealing algorithm, so that the problem solving time is remarkably reduced.

Further, in the step (1), the vehicle end calculates and communicates the ability to choose to offload the task to RSU to carry out or carry out locally; each vehicle user selects whether to offload tasks to the RSU; describing relevant characteristics of the RSU through a tuple, wherein the relevant characteristics comprise communication range, RSU communication bandwidth and computing capacity; the tuples may be represented by quaternions: { W, F ^max R, D }, where W represents the total bandwidth available to the RSU; f (F) ^max Representing the maximum calculation frequency of the RSU; r represents the communication radius of the RSU; d represents the vertical distance between the RSU and the roadThe method comprises the steps of carrying out a first treatment on the surface of the Defining a set of vehicles generating a task asWherein N is the total number of vehicles, i is the number of each vehicle; wherein the vehicle i has a speed v _i Assuming that the vehicle is kept traveling at a constant speed within the RSU range; finally, describing the characteristics of each vehicle computing task through a tuple, wherein the characteristics comprise a task execution period and task input data; the tuple can be a two-tuple { C _i ,L _i Described by }, wherein C _i A task execution cycle representing the vehicle i; l (L) _i The task input data size of the vehicle i is indicated.

Further, in step (2), d is defined in consideration of mobility between vehicles _i The distance between the vehicle i and the RSU is set; the vehicle starts to follow a constant speed at the intersection with the RSU coverage edge, d _i =r; at this time, the communication time of the vehicle in the range of the RSU is defined asThe vehicle i and RSU transmission rates are expressed by shannon's theorem as:

S _i ＝w _i Wlog ₂ (1+SNR _i ) #(1)

wherein w is _i Denoted as the portion of bandwidth allocated to vehicle i; w is the total bandwidth available. SNR between vehicle i and RSU _i The formula is:

wherein p is _i Is the signal transmission power of vehicle i;a distance path loss based on delta coefficient, denoted as vehicle i; sigma (sigma) ² Is the noise variance;

further, in step (3), the decision variable alpha is introduced _i To indicate at which end the task is performed: when the vehicle i is to perform a task locally,α _i =0; if unloading to RSU execution, alpha _i =1. Thus, the total execution time and total energy consumption of the task of the vehicle i are expressed as:

wherein the method comprises the steps ofTime delay and energy consumption generated by unloading execution of the vehicle i are represented; />Represented by the time delay and energy consumption that the vehicle i generates when executing locally.

Specifically, the vehicle-carried task is performed in the local section:

the local CPU computing power of the vehicle is expressed as:the task carried by the vehicle i is executed in the local section for the following time:

processor energy consumption is modeled as:wherein->Representing the energy consumption coefficient of the vehicle processor. By multiplying the above equation with equation (5), the energy consumed to process the ith task is obtained:

when the RSU end uninstalls and executes:

the time required for task offloading typically includes three parts: and uploading the data carried by the task to the RSU at the roadside by the vehicle, wherein the RSU receives the data and then processes the data to obtain the execution time and the output time required by the RSU for returning the data to the vehicle after the RSU processes the data. The transmission delay required by the vehicle to transmit a task to the RSU is expressed as:

upon receiving the offloaded task from the vehicle, the RSU will perform the task on behalf of the vehicle and return the output to the user after completion. The task execution time delay is expressed as:

in summary, the total time delay of task execution at the RSU is:

the vehicle generates energy consumption due to task transmission to the RSUI.e. the product of the transmission power of the vehicle i and the time required for the task transmission, expressed by the formula:

defining the system utility as the experience quality of a vehicle end user, namely QoE; whereas in VEC systems QoE appears to be improved in terms of time and energy consumption compared to vehicle on-site calculationsHow much, specifically described as:andthe formula is:

when the vehicle i performs a task locally, namely:the vehicle user QoE is 0. Meanwhile, if the task is unloaded to the RSU for execution, because the RSU resources are abundant, when the task execution time delay and the energy consumption are lower: />At this timePositive, vehicle user QoE positive. However, with excessive offloading tasks, since the RSU computing communication resources are limited, offloading to the RSU computing may result in longer latency and energy consumption compared to the local computing when the latency and energy consumption of the remote computing are more, i.e.: />The vehicle user QoE is negative. In addition, a->And is also provided withThe concrete explanation is as follows: vehicle end users prefer time delay and energy consumption. For example: the user with shorter battery life at the vehicle end can add +.>While reducing->Thereby saving more energy at the cost of longer task completion times. In practical use, the vehicle side is set by different power saving modes>Is a value of (2). Such as: in extreme energy saving mode: />Or under maximum performance conditions: />

As can be seen from equation (11), the task offload decision variable a, i.e., a= { i|α _i Vehicle-side task transmission power =1 }Bandwidth allocation +.>And frequency scaling can affect the magnitude of the vehicle QoE. In order to illustrate the QoE-based internet of vehicles task offloading and resource allocation method, the following constraints need to be introduced:

(1) Task offloading constraints: regardless of the task migration technique, the tasks of vehicle i are completely offloaded to the RSU or are performed at the vehicle's local end. There is therefore a need to introduce the following constraints in terms of task offloading:

(2) Bandwidth allocation constraints: due to the limited bandwidth resources of the vehicle to RSU communication, it must be ensured that the bandwidth of the total offloaded vehicle does not exceed the total available bandwidth. The following constraints are thus introduced:

(3) Transmission power constraint: when the vehicle transmits tasks to the RSU, the transmission power needs to be ensured not to exceed the maximum transmission power of the vehicleThe following constraints are thus introduced:

(4) Calculating a frequency constraint: the RSU side has limited computational resources, and it must be ensured that the total computational resources required for the task transmitted to the RSU should not exceed the available computational resources of the RSU. The following constraints are thus introduced:

(5) Task real-time constraints: the vehicles can communicate only within the range of the RSU, and when the vehicles leave the range, the vehicles are disconnected with the RSU, so that task unloading failure is caused, and therefore, the task unloading calculation time must be ensuredIn vehicle communication time->Within a range of (2). The following constraints are thus introduced:

to improve system QoE, the task offloading problem can be expressed as the following model:

s.t.(12)-(18)。

further, in the step (4), the problem model established in the step (3) is decomposed. The method comprises the steps of decomposing a resource allocation problem into a fixed task unloading decision and optimizing a task unloading problem corresponding to the resource allocation problem.

Resource allocation problem for fixed task offloading:

the objective function model is to solve the most available resource allocation strategy to maximize the system QoE, if the task offloading decision is known, namely:

s.t.(13)-(18)

because for a given offloading decision-making,to be fixed, question->The adaptation is to minimize the overhead of unloading the vehicle user, namely: />According to the above formula, question->The equivalent is:

s.t.(13)-(18)

wherein the method comprises the steps ofIn order to find the most power and bandwidth allocation, the problem +.>The conversion is to solve the Lagrange dual function, which is specifically expressed as:

s.t.χ _p,w,f ≥0

wherein the Lagrangian function is expressed as:

wherein,represented as non-negative lagrangian multipliers; f (F) ^max Representing the maximum computing power of the RSU, f _i Representing a frequency scaling factor; p is p _i Is the signal transmission power of vehicle i; />Is the maximum signal transmission power of vehicle i; x-shaped articles _p,w,f And the equal to or greater than 0 is Lagrange multiplier constraint. The specific solving steps are as follows:

first, according to the Karush-Kuhn-Tucker (KKT) condition, the Lagrangian function is relative to p _i ,w _i ,f _i The derivatives of (2) are as follows:

because of the difficulty in finding the closed form expression from (24), the dichotomy is used to find p _i Is a root of (2). The method comprises the following specific steps:

1) Initializing an interval: setting p _i Is a range of values of (a)Wherein->So that

2) The mid-point of the interval is calculated at each step:

3) Checking convergence: if it isWhere ε is a given small positive number, then the iteration is stopped and p _i As p _i Is a root of (2);

4) Determining the next subinterval: if it isThen say root is located in interval->In, thus let->If->Then say root is located in interval->In, thus let->

5) Repeating the steps until the convergence condition is satisfied.

Second, to obtain the best bandwidth allocation, the Lagrangian function is calculated relative to w _i The result is shown in equations (27) (28).

After solving the problem by dichotomy and equations (27), (28)After the solution of (2), the lagrangian multiplier is updated as:

wherein s is ₁ ，s ₂ ，s ₃ ，s ₄ ，s ₅ Sum s ₆ Is a positive step size. The optimal power and bandwidth allocation is sought through a gradient iteration method, and the process is as follows:

1) Initializing: setting the transmission power p of each vehicle _i Bandwidth and computing resource allocation omega _i ,f _i The method comprises the steps of carrying out a first treatment on the surface of the Initializing Lagrangian multiplier χ _p,w,f The method comprises the steps of carrying out a first treatment on the surface of the Setting the iteration times t=0; setting an initial objective function value:

2) Iterative optimization: in each iteration, the resource allocation is first performed to determine the transmission power according to the dichotomy, and the bandwidth and the calculation resource allocation ω are calculated by the formulas (27), (28) _i ,f _i . The lagrangian multiplier χ is then updated by equations (29) - (34) _p,w,f 3) Convergence check: after each iteration is completed, the objective function value Θt is recalculated]. If Θ [ t ]]And Θ [ t-1]]If the difference is smaller than the preset threshold, the algorithm is considered to be converged, and the iterative process is ended.

For a given offloading decision, the allocation of computational and communication resources has been solved in the above steps, in particular: according to the formula and the algorithm, the system utility is obtained as follows:

solving the current unloading decision A according to a gradient iteration methodOptimal resource and communication allocation strategyAt this time, problem->Can be reduced to the question->Namely:

s.t.|A|≤N

the problem is solved by selecting a simulated annealing algorithm, and the specific steps are as follows:

the first step: let t=t ₀ Indicated as the temperature at which annealing was started. An initial solution is randomly generated, namely: unloading decision set A ₀ 。

And a second step of: obtaining an optimal resource allocation strategy under the condition of the current solution according to a gradient iteration method, and combining (35) to calculate a corresponding QoE value J (A ₀ )。

And a third step of: randomly perturbing the current solution, wherein the perturbation rule is as follows: randomly modifying the unloading decision of the vehicle to generate a new solution and calculating the QoE value of the current solution according to step 2).

Fourth step: assume that the current solution is J (A _i ) The newly solved QoE value is J (a _i+1 ). By comparing the sizes of the current solution and the new solution, if the current solution is larger than the current solution, the unloading strategy is accepted, the current solution of the function is updated, otherwise, the current solution is according to the formulaAnd obtaining the cooling probability p, generating a random number which is uniformly distributed, and accepting the unloading strategy if the cooling probability p is larger than the random number, otherwise, giving up the unloading strategy.

Fifth step: according to the cooling coefficient and the temperature updating rule: t (T) _k ＝T ₀ γ ^k ，γ<1, wherein γ is the annealing rate. And (3) adjusting the current temperature, and repeatedly executing the steps 3) -4).

Sixth step: and when the termination temperature or the specified iteration times are reached, terminating annealing, and solving the maximum unloading effect of the vehicle user.

The invention has the beneficial effects that:

1. according to the method, the relation between the calculation time delay and the energy consumption is comprehensively considered, and QoE of a vehicle user is modeled as improvement of the local calculation of comprehensively considered task execution time delay and energy consumption;

2. the invention provides a method for reasonably unloading tasks under the condition of considering the limited RSU computing resources so as to maximize the utility of the system;

3. aiming at the MINLP problem with high complexity, the method decomposes the problem into two sub-problems of task unloading and resource allocation, reduces the complexity of the problem and greatly improves the utility of the system.

Drawings

FIG. 1 is a flow chart of the steps of the inventive concept and solution;

FIG. 2 is a schematic diagram of a vehicle network task offload framework in accordance with the present invention;

FIG. 3 is a schematic diagram of the present invention for solving the vehicle transmission power using dichotomy;

FIG. 4 is a schematic diagram of the present invention for solving an optimal resource allocation strategy using gradient iterations;

FIG. 5 shows the arrangement of different C's in the example of the invention _i Comparing the difference schematic diagram between the algorithm and the optimal solution;

FIG. 6 is a schematic diagram of the difference between the running time of the algorithm proposed by the present invention and other algorithms in the example of the present invention;

FIG. 7 is a schematic illustration of setting C in an example of the invention _i ＝1000*10 ⁶ Schematic diagram of the influence of 10 groups of different vehicle numbers on the system utility during cycle;

FIG. 8 is an illustration of setting C in an example of the invention _i ＝1500*10 ⁶ Schematic diagram of the influence of 10 groups of different vehicle numbers on the system utility during cycle;

FIG. 9 is an illustration of setting C in an example of the invention _i ＝2000*10 ⁶ Schematic diagram of the influence of 10 groups of different vehicle numbers on the system utility during cycle;

FIG. 10 is a schematic diagram illustrating the effect of setting 5 different task input data sizes on system utility in an example of the present invention;

FIG. 11 is a schematic diagram showing the effect of setting 5 different task execution cycle sizes on the system utility in an example of the present invention;

FIG. 12 is a schematic diagram of studying the effect of a vehicle user on the utility of a system on time delay preference in an example of the invention;

FIG. 13 is a schematic diagram of studying the impact of a vehicle user on energy consumption preferences on system utility in an example of the invention.

Detailed Description

The present invention is further illustrated in the following drawings and detailed description, which are to be understood as being merely illustrative of the invention and not limiting the scope of the invention.

As shown in the figure, the QoE-based internet of vehicles task offloading and resource allocation method of the present invention includes the following steps:

the method comprises the following steps that (1), a vehicle side calculates and communicates capacity selection to offload tasks to an RSU for execution or to perform tasks locally; each vehicle user selects whether to offload tasks to the RSU; describing relevant characteristics of the RSU through a tuple, wherein the relevant characteristics comprise communication range, RSU communication bandwidth and computing capacity; the tuples may be represented by quaternions: { W, F ^max R, D }, where W represents the total bandwidth available to the RSU; f (F) ^max Representing the maximum calculation frequency of the RSU; r represents the communication radius of the RSU; d represents the vertical distance between the RSU and the road; defining a set of vehicles generating a task asWherein N is the total number of vehicles, i is the number of each vehicle; wherein the vehicle i has a speed v _i Assuming that the vehicle is kept traveling at a constant speed within the RSU range; finally, describing the characteristics of each vehicle computing task through a tuple, wherein the characteristics comprise a task execution period and task input data; the tuple can be a two-tuple { C _i ,L _i Described by }, wherein C _i A task execution cycle representing the vehicle i; l (L) _i The task input data size of the vehicle i is indicated.

Step (2) of defining d in consideration of mobility between vehicles _i The distance between the vehicle i and the RSU is set; the vehicle starts to follow a constant speed at the intersection with the RSU coverage edge, d _i =r; at this time, the communication time of the vehicle in the range of the RSU is defined asThe vehicle i and RSU transmission rates are expressed by shannon's theorem as:

S _i ＝w _i Wlog ₂ (1+SNR _i ) #(1)

wherein w is _i Denoted as the portion of bandwidth allocated to vehicle i; w is the total bandwidth available. SNR between vehicle i and RSU _i The formula is:

wherein p is _i Is the signal transmission power of vehicle i;a distance path loss based on delta coefficient, denoted as vehicle i; sigma (sigma) ² Is the noise variance;

step (3) of introducing a decision variable alpha _i To indicate at which end the task is performed: when the vehicle i performs the task locally, α _i =0; if unloading to RSU execution, alpha _i =1. Thus, the total execution time and total energy consumption of the task of the vehicle i are expressed as:

wherein the method comprises the steps ofTime delay and energy consumption generated by unloading execution of the vehicle i are represented; />Represented by the time delay and energy consumption that the vehicle i generates when executing locally.

Specifically, the vehicle-carried task is performed in the local section:

the local CPU computing power of the vehicle is expressed as:the task carried by the vehicle i is executed in the local section for the following time:

processor energy consumption is modeled as:wherein->Representing the energy consumption coefficient of the vehicle processor. By multiplying the above equation with equation (5), the energy consumed to process the ith task is obtained:

when the RSU end uninstalls and executes:

the time required for task offloading typically includes three parts: and uploading the data carried by the task to the RSU at the roadside by the vehicle, wherein the RSU receives the data and then processes the data to obtain the execution time and the output time required by the RSU for returning the data to the vehicle after the RSU processes the data. The transmission delay required by the vehicle to transmit a task to the RSU is expressed as:

upon receiving the offloaded task from the vehicle, the RSU will perform the task on behalf of the vehicle and return the output to the user after completion. The task execution time delay is expressed as:

in summary, the total time delay of task execution at the RSU is:

the vehicle generates energy consumption due to task transmission to the RSUI.e. the product of the transmission power of the vehicle i and the time required for the task transmission, expressed by the formula:

defining the system utility as the experience quality of a vehicle end user, namely QoE; whereas in VEC systems, qoE appears to be improved in terms of time and energy consumption compared to vehicle-local end calculations, described in particular as:andthe formula is:

when the vehicle i performs a task locally, namely:the vehicle user QoE is 0; meanwhile, if the task is unloaded to the RSU for execution, because the RSU resources are abundant, when the task execution time delay and the energy consumption are lower: />At this timePositive, vehicle user QoE positive; however, with excessive offloading tasks, since the RSU computing communication resources are limited, offloading to the RSU computing may result in longer latency and energy consumption compared to the local computing when the latency and energy consumption of the remote computing are more, i.e.: />The vehicle user QoE is negative; in addition, a->And is also provided withThe concrete explanation is as follows: preference of vehicle end users for time delay and energy consumption; in practical use, the vehicle side is set by different power saving modes>Is a value of (2);

from equation (11), the task offload decision variable a, i.e., a= { i|α _i Vehicle-side task transmission power =1 }Bandwidth allocation +.>Frequency scaling +.>The magnitude of the vehicle QoE may be affected; in order to illustrate the QoE-based internet of vehicles task offloading and resource allocation method, the following constraints need to be introduced:

(1) Task offloading constraints: regardless of the task migration technique, the tasks of vehicle i are completely offloaded to the RSU or are performed at the vehicle's local end. There is therefore a need to introduce the following constraints in terms of task offloading:

(2) Bandwidth allocation constraints: due to the limited bandwidth resources of the vehicle to RSU communication, it must be ensured that the bandwidth of the total offloaded vehicle does not exceed the total available bandwidth. The following constraints are thus introduced:

(3) Transmission power constraint: when the vehicle transmits tasks to the RSU, the transmission power needs to be ensured not to exceed the maximum transmission power of the vehicleThe following constraints are thus introduced:

(4) Calculating a frequency constraint: the RSU side has limited computational resources, and it must be ensured that the total computational resources required for the task transmitted to the RSU should not exceed the available computational resources of the RSU. The following constraints are thus introduced:

(5) Task real-time constraints: the vehicles can communicate only within the range of the RSU, and when the vehicles leave the range, the vehicles are disconnected with the RSU, so that task unloading failure is caused, and therefore, the task unloading calculation time must be ensuredIn vehicle communication time->Within a range of (2). The following constraints are thus introduced:

to improve system QoE, the task offloading problem can be expressed as the following model:

s.t.(12)-(18)

and (4) decomposing the problem model established in the step (3). The method comprises the steps of decomposing a resource allocation problem into a fixed task unloading decision and optimizing a task unloading problem corresponding to the resource allocation problem.

Resource allocation problem for fixed task offloading:

the objective function model is to solve the most available resource allocation strategy to maximize the system QoE, if the task offloading decision is known, namely:

s.t.(13)-(18)

because for a given offloading decision-making,to be fixed, question->The adaptation is to minimize the overhead of unloading the vehicle user, namely: />According to the above formula, question->The equivalent is:

s.t.(13)-(18)

wherein the method comprises the steps ofIn order to find the most power and bandwidth allocation, the problem +.>The conversion is to solve the Lagrange dual function, which is specifically expressed as:

s.t.χ _p,w,f ≥0

wherein the Lagrangian function is expressed as:

wherein,represented as non-negative lagrangian multipliers; f (F) ^max Representing the maximum computing power of the RSU, f _i Representing a frequency scaling factor; p is p _i Is the signal transmission power of vehicle i; />Is the maximum signal transmission power of vehicle i; x-shaped articles _p,w,f And the equal to or greater than 0 is Lagrange multiplier constraint. The specific solving steps are as follows:

first, according to the Karush-Kuhn-Tucker (KKT) condition, the Lagrangian function is relative to p _i ,w _i ,f _i The derivatives of (2) are as follows:

/>

because of the difficulty in finding the closed form expression from (24), the dichotomy is used to find p _i Is a root of (2). The algorithm solving process is shown in fig. 3.

1) Initializing an interval: setting p _i Is a range of values of (a)Wherein->So that

2) The mid-point of the interval is calculated at each step:

3) Checking convergence: if it isWhere ε is a given small positive number, then the iteration is stopped and p _i As p _i Is a root of (2);

4) Determining the next subinterval: if it isThen say root is located in interval->In, thus let->If->Then say root is located in interval->In, thus let->

5) Repeating the steps until the convergence condition is satisfied.

Second, to obtain the best bandwidth allocation, the Lagrangian function is calculated relative to w _i The result is shown in equations (27) (28).

The problem is obtained by solving the algorithm of FIG. 3 and equations (27), (28)After the solution of (2), the lagrangian multiplier is updated as:

/>

wherein s is ₁ ，s ₂ ，s ₃ ，s ₄ ，s ₅ Sum s ₆ Is a positive step size. A gradient descent method for optimizing resource allocation is presented in fig. 4. The optimal power and bandwidth allocation is sought through a gradient iteration method, and the process is as follows:

1) Initializing: setting the transmission power p of each vehicle _i Bandwidth and computing resource allocation omega _i ,f _i The method comprises the steps of carrying out a first treatment on the surface of the Initializing Lagrangian multiplier χ _p,w,f The method comprises the steps of carrying out a first treatment on the surface of the Setting the iteration times t=0; setting an initial objective function value:

2) Iterative optimization: each timeIn the iteration, first, resource allocation is performed to determine transmission power according to the dichotomy, and bandwidth and resource allocation ω are calculated by the formulas (27), (28) _i ,f _i . The lagrangian multiplier χ is then updated by equations (29) - (34) _p,w,f ；

3) Convergence check: after each iteration is completed, the objective function value Θt is recalculated. If the difference between Θt and Θt-1 is less than a predetermined threshold, the algorithm is considered to have reached convergence and the iterative process ends.

For a given offloading decision, the allocation of computational and communication resources has been solved in the above steps, in particular: according to the formula and the algorithm, the system utility is obtained as follows:

solving the optimal resource and communication allocation strategy under the current unloading decision A according to a gradient iteration methodAt this time, problem->Can be reduced to the question->Namely:

s.t.|A|≤N

the problem is solved by selecting a simulated annealing algorithm, and the specific steps are as follows:

the first step: let t=t ₀ Indicated as the temperature at which annealing was started. An initial solution is randomly generated, namely: unloading decision set A ₀ 。

And a second step of: the most in the case of obtaining the current solution according to the gradient iteration methodOptimal resource allocation strategy, and combining (35) to calculate corresponding QoE value J (A ₀ )。

And a third step of: randomly perturbing the current solution, wherein the perturbation rule is as follows: randomly modifying the unloading decision of the vehicle to generate a new solution and calculating the QoE value of the current solution according to step 2).

Fourth step: assume that the current solution is J (A _i ) The newly solved QoE value is J (a _i+1 ). By comparing the sizes of the current solution and the new solution, if the current solution is larger than the current solution, the unloading strategy is accepted, the current solution of the function is updated, otherwise, the current solution is according to the formulaAnd obtaining the cooling probability p, generating a random number which is uniformly distributed, and accepting the unloading strategy if the cooling probability p is larger than the random number, otherwise, giving up the unloading strategy.

Fifth step: according to the cooling coefficient and the temperature updating rule: t (T) _k ＝T ₀ γ ^k ，γ<1, wherein γ is the annealing rate. And (3) adjusting the current temperature, and repeatedly executing the steps 3) -4).

Sixth step: and when the termination temperature or the specified iteration times are reached, terminating annealing, and solving the maximum unloading effect of the vehicle user.

Example 2:

in order to measure the reliability of the method, experimental tests are carried out on the method, the experimental environment of the method is Matlab R2021b, and experimental parameters are selected and referred to the settings in the vehicle-mounted mobile network in IEEE802.11 p; FIG. 1 is a flow chart of the steps of the method of the invention; FIGS. 5-12 are graphs showing the experimental results of the present invention.

Fig. 5-6 are comparisons of the difference between the SAJTORA algorithm proposed in example 2 of the present invention and the exhaustive approach to find the optimal solution, and the algorithm run time. It can be seen from fig. 5 that the solution obtained by the SAJTORA algorithm proposed by the present invention is very close to the optimal solution of the exhaustive method and significantly better than the other algorithms. It can be seen in connection with fig. 5,6 that the solution found by our proposed algorithm is close to the optimal solution and has less run time.

FIGS. 7-9 illustrate the arrangement of example 2 of the present inventionDifferent mission execution cycles study the relationship between different vehicle numbers and system utility. It can be seen from the figure that the SAJTORA algorithm always performs best and that the performance of all schemes improves significantly as the workload of the task increases. This is because of task workload c _i The increase results in more computing resources being required to perform the task, and the vehicle user offloads its task to the resource-rich RSU server, thereby reducing latency and energy consumption, with a consequent increase in system utility QoE. As can be seen from the figure, when the number of users is small, the system utility increases with the increase of the number of users; however, when the number of users exceeds a certain threshold, the system utility will begin to decrease. This is because when there are many vehicle users competing for communication with the RSU end and computing resources to offload tasks, the overhead of sending tasks and executing tasks on the MEC server will be higher, thereby reducing the offload utility. For the IOJRA algorithm, only the calculation resource size of the vehicle is considered, most of tasks are selected to be executed on a local end, and the tasks are only considered to be unloaded to the RSU end for execution under the condition that the task quantity is too large to exceed a certain threshold value, so that the IOJRA algorithm obtains the least system utility compared with other algorithms for unloading the tasks to the RSU end with rich calculation resources. The ROJRA algorithm randomly selects whether the task is unloaded or not, so that the system utility is increased to a certain extent; however, when the number of vehicles is large, the task of random offloading to the RSU is calculated in a large amount and the allocated computing resources are small, which results in a decrease in the utility of the system.

The hJTORA algorithm selects heuristic algorithm to perform task removal and update operation, so that the system utility can be better increased, and the system is easy to fall into a local optimal solution when the number of vehicles is increased. The algorithm proposed in this chapter can jump out of the local optimal solution, thus better increasing the system utility.

FIG. 10 is a schematic diagram of the effect of setting 5 different task input data sizes on system utility in example 2 of the present invention; it can be seen that data L is input with a task _i But the system utility decreases. It can be seen from this that a higher system utility can be achieved when tasks with a smaller amount of task input data are offloaded to the RSU for execution.

FIG. 11 is a schematic illustration of the arrangement of 5 sets in example 2 of the present inventionSchematic diagrams of the influence of different task input data sizes on the system utility; it can be seen that with task execution period C _i And the system utility increases. It can be seen from this that a higher system utility can be obtained when tasks with longer task execution periods are offloaded to the RSU for execution.

FIGS. 12 and 13 are schematic diagrams of the effect of studying vehicle users on time delay and energy consumption preferences on system utility in example 2 of the present invention; it can be seen that whenWhen increased, the average time consumption decreases, at the cost of increased energy consumption. In addition, as there are more vehicles in the VEC system, the vehicle user will get a greater average time and energy consumption. This is because the chances of the vehicle end getting low latency and energy consumption from the RSU are reduced when there are more vehicles competing for limited resources.

It should be noted that the foregoing merely illustrates the technical idea of the present invention and is not intended to limit the scope of the present invention, and that a person skilled in the art may make several improvements and modifications without departing from the principles of the present invention, which fall within the scope of the claims of the present invention.

Claims (5)

1. The QoE-based vehicle networking task unloading and resource allocation method is characterized by comprising the following steps of:

(1) Building a vehicle networking edge calculation VEC system model according to the vehicle end calculation and communication capability;

(2) Based on the VEC system, the vehicle communicates with the road side unit RSU through the wireless communication technology V2I technology, so that the calculation task is unloaded to the RSU for execution;

(3) By introducing a task unloading decision, vehicle transmission power, vehicle communication bandwidth and RSU computing resource constraint, the QoE problem of the maximized system is constructed;

(4) According to the problem structure, constraint conditions are mutually independent, so that the problem is resolved, the problem of resource allocation is solved by using a gradient iteration method, the system QoE is maximized by using a simulated annealing algorithm, and the problem solving time is shortened.

2. The QoE-based internet of vehicles task offloading and resource allocation method of claim 1, wherein in step (1), the vehicle side computing and communication capability selects to offload tasks to the RSU for execution or to execute locally; each vehicle user selects whether to offload tasks to the RSU; describing relevant characteristics of the RSU through a tuple, wherein the relevant characteristics comprise communication range, RSU communication bandwidth and computing capacity; the tuples may be represented by quaternions: { W, F ^max R, D }, where W represents the total bandwidth available to the RSU; f (F) ^max Representing the maximum calculation frequency of the RSU; r represents the communication radius of the RSU; d represents the vertical distance between the RSU and the road; defining a set of vehicles generating a task asWherein N is the total number of vehicles, i is the number of each vehicle; wherein the vehicle i has a speed v _i Assuming that the vehicle is kept traveling at a constant speed within the RSU range; finally, describing the characteristics of each vehicle computing task through a tuple, wherein the characteristics comprise a task execution period and task input data; the tuple can be a two-tuple { C _i ,L _i Described by }, wherein C _i A task execution cycle representing the vehicle i; l (L) _i The task input data size of the vehicle i is indicated.

3. The QoE-based internet of vehicles task offloading and resource allocation method of claim 1, wherein in step (2), d is defined considering mobility between vehicles _i The distance between the vehicle i and the RSU is set; the vehicle starts to follow a constant speed at the intersection with the RSU coverage edge, d _i =r; at this time, the communication time of the vehicle in the range of the RSU is defined asThe vehicle i and RSU transmission rates are expressed by shannon's theorem as:

S _i ＝w _i Wlog ₂ (1+SNR _i )#(1)

wherein w is _i Denoted as the portion of bandwidth allocated to vehicle i; w is the total bandwidth available; SNR between vehicle i and RSU _i The formula is:

wherein p is _i Is the signal transmission power of vehicle i;a distance path loss based on delta coefficient, denoted as vehicle i; sigma (sigma) ² Is the noise variance.

4. The QoE-based internet of vehicles task offloading and resource allocation method of claim 1, wherein in step (3), the decision variable α is introduced _i To indicate at which end the task is performed: when the vehicle i performs the task locally, α _i =0; if unloading to RSU execution, alpha _i =1; thus, the total execution time and total energy consumption of the task of the vehicle i are expressed as:

wherein the method comprises the steps ofTime delay and energy consumption generated by unloading execution of the vehicle i are represented; />Represented as vehicle i when executing locallyThe time delay and energy consumption are generated;

specifically, the vehicle-carried task is performed in the local section:

the local CPU computing power of the vehicle is expressed as: f (f) _i ^loc The task carried by the vehicle i is executed in the local section for the following time:

processor energy consumption is modeled as:wherein->Representing an energy consumption coefficient of a vehicle processor; by multiplying the above equation with equation (5), the energy consumed to process the ith task is obtained:

when the RSU end uninstalls and executes:

the time required for task offloading typically includes three parts: the vehicle uploads the data carried by the task to the transmission time required by the roadside RSU, the execution time required by the RSU for processing after receiving the data and the output time required by the RSU for returning the data to the vehicle after finishing the processing; the transmission delay required by the vehicle to transmit a task to the RSU is expressed as:

after receiving the offloaded task from the vehicle, the RSU will perform the task on behalf of the vehicle and return the output to the user after completion; the task execution time delay is expressed as:

in summary, the total time delay of task execution at the RSU is:

the vehicle generates energy consumption due to task transmission to the RSUI.e. the product of the transmission power of the vehicle i and the time required for the task transmission, expressed by the formula:

defining the system utility as the experience quality of a vehicle end user, namely QoE; whereas in VEC systems, qoE appears to be improved in terms of time and energy consumption compared to vehicle-local end calculations, described in particular as:and->The formula is:

when the vehicle i performs a task locally, namely:the vehicle user QoE is 0; meanwhile, if the task is unloaded to the RSU for execution, the RSU has abundant resources, and the time delay and the energy consumption for executing the task are lowerWhen (1): />At this time->Positive, vehicle user QoE positive; however, with excessive offloading tasks, since the RSU computing communication resources are limited, offloading to the RSU computing may result in longer latency and energy consumption compared to the local computing when the latency and energy consumption of the remote computing are more, i.e.: />The vehicle user QoE is negative; in addition, a->And is also provided withThe concrete explanation is as follows: preference of vehicle end users for time delay and energy consumption; in practical use, the vehicle side is set by different power saving modes>Is a value of (2);

from equation (11), the task offload decision variable a, i.e., a= { i|α _i =1 }, vehicle-side task transmission power P, bandwidth allocationFrequency scaling +.>The magnitude of the vehicle QoE may be affected; in order to illustrate the QoE-based internet of vehicles task offloading and resource allocation method, the following constraints need to be introduced:

(1) Task offloading constraints: irrespective of the task migration technique, the task of the vehicle i is completely offloaded to the RSU or is executed at the vehicle local end; there is therefore a need to introduce the following constraints in terms of task offloading:

(2) Bandwidth allocation constraints: because the communication bandwidth resources of the vehicles and the RSU are limited, the bandwidth of the total unloading vehicles must be ensured not to exceed the available total bandwidth; the following constraints are thus introduced:

(3) Transmission power constraint: when the vehicle transmits tasks to the RSU, the transmission power needs to be ensured not to exceed the maximum transmission power of the vehicleThe following constraints are thus introduced:

(4) Calculating a frequency constraint: the RSU end has limited computing resources, and the total computing resources required by the task transmitted to the RSU must not exceed the available computing resources of the RSU; the following constraints are thus introduced:

(5) Task real-time constraints: the vehicles can communicate only within the range of the RSU, and when the vehicles leave the range, the vehicles are disconnected with the RSU, so that task unloading failure is caused, and therefore, the task unloading calculation time must be ensuredIn vehicle communication time->Is within the range of (2); the following constraints are thus introduced:

to improve system QoE, the task offloading problem is expressed as the following model:

s.t.(12)-(18)。

5. the QoE-based internet of vehicles task offloading and resource allocation method of claim 1, wherein in step (4), decomposing the problem model established in step (3); decomposing the resource allocation problem into a fixed task unloading decision and optimizing a task unloading problem corresponding to the resource allocation problem;

resource allocation problem for fixed task offloading:

the objective function model is to solve the most available resource allocation strategy to maximize the system QoE, if the task offloading decision is known, namely:

s.t.(13)-(18)

because for a given offloading decision-making,to be fixed, question->The adaptation is to minimize the overhead of unloading the vehicle user, namely: />According to the above formula, the problem is solvedThe equivalent is:

s.t.(13)-(18)

wherein the method comprises the steps ofIn order to find the most power and bandwidth allocation, the problem +.>The conversion is to solve the Lagrange dual function, which is specifically expressed as:

s.t.χ _p,w,f ≥0

wherein the Lagrangian function is expressed as:

wherein,represented as non-negative lagrangian multipliers; f (F) ^max Representing the maximum computing power of the RSU, f _i Representing a frequency scaling factor; p is p _i Is the signal transmission power of vehicle i; />Is the maximum signal transmission power of vehicle i; x-shaped articles _p,w,f 0 is Lagrange multiplier constraint; the specific solving steps are as follows:

first, according to the KKT condition, the Lagrangian function is relative to p _i ,w _i ,f _i The derivatives of (2) are as follows:

because of the difficulty in finding the closed form expression from (24), the dichotomy is used to find p _i Is a root of (2); the method comprises the following specific steps:

1) Initializing an interval: setting p _i Is a range of values of (a)Wherein->So that

2) The mid-point of the interval is calculated at each step:

3) Checking convergence: if it isWhere ε is a given small positive number, then the iteration is stopped and p _i As p _i Is a root of (2);

4) Determining the next subinterval: if it isThen say root is located in interval->In the middle, thus letIf->Then say root is located in interval->In, thus let->

5) Repeating the steps until convergence conditions are met;

second, to obtain the best bandwidth allocation, the Lagrangian function is calculated relative to w _i Results are shown as formula [ (first derivative)27 (28);

after solving the problem by dichotomy and equations (27), (28)After the solution of (2), the lagrangian multiplier is updated as:

wherein s is ₁ ，s ₂ ，s ₃ ，s ₄ ，s ₅ Sum s ₆ Is a positive step size; the optimal power and bandwidth allocation is sought through a gradient iteration method, and the process is as follows:

1) Initializing: setting the transmission power p of each vehicle _i Bandwidth and computing resource allocation omega _i ,f _i The method comprises the steps of carrying out a first treatment on the surface of the Initializing Lagrangian multiplier χ _p,w,f The method comprises the steps of carrying out a first treatment on the surface of the Setting the iteration times t=0; setting an initial objective function value:

2) Iterative optimization: in each iteration, the resource allocation is first performed to determine the transmission power according to the dichotomy, and the bandwidth and the calculation resource allocation ω are calculated by the formulas (27), (28) _i ,f _i The method comprises the steps of carrying out a first treatment on the surface of the The lagrangian multiplier χ is then updated by equations (29) - (34) _p,w,f ；

3) Convergence check: after each iteration is finished, the objective function value Θt is recalculated; if the difference between Θ [ t ] and Θ [ t-1] is less than a predetermined threshold, the algorithm is considered to have reached convergence, and the iterative process is ended;

for a given offloading decision, the allocation of computational and communication resources has been solved in the above steps, in particular: according to the formula and the algorithm, the system utility is obtained as follows:

solving the optimal resource and communication allocation strategy under the current unloading decision A according to a gradient iteration methodAt this time, problem->Is simplified as question->Namely:

the problem is solved by selecting a simulated annealing algorithm, and the specific steps are as follows:

the first step: let a=a ₀ Expressed as the temperature at which annealing begins; an initial solution is randomly generated, namely: unloading decision set A ₀ ；

And a second step of: obtaining an optimal resource allocation strategy under the condition of the current solution according to a gradient iteration method, and calculating a corresponding QoE value J (A) by combining the optimal resource allocation strategy with the QoE value J (35) ₀ )；

And a third step of: randomly perturbing the current solution, wherein the perturbation rule is as follows: randomly modifying the unloading decision of the vehicle to generate a new solution and calculating the QoE value of the current solution according to the step 2);

fourth step: assume that the current solution is J (A _i ) The newly solved QoE value is J (a _i+1 ) The method comprises the steps of carrying out a first treatment on the surface of the By comparing the sizes of the current solution and the new solution, if the current solution is larger than the current solution, the unloading strategy is accepted, the current solution of the function is updated, otherwise, the current solution is according to the formulaObtaining cooling probability p, generating a random number which is uniformly distributed, and receiving an unloading strategy if p is larger than the random number, otherwise, giving up the unloading strategy;

fifth step: according to the cooling coefficient and the temperature updating rule: t (T) _k ＝T ₀ γ ^k ，γ<1, wherein gamma is the annealing rate; adjusting the current temperature, and repeatedly executing the steps 3) -4);

sixth step: and when the termination temperature or the specified iteration times are reached, terminating annealing, and solving the maximum unloading effect of the vehicle user.