CN116634401A - Task unloading method for maximizing satisfaction of vehicle-mounted user under edge calculation - Google Patents
Task unloading method for maximizing satisfaction of vehicle-mounted user under edge calculation Download PDFInfo
- Publication number
- CN116634401A CN116634401A CN202310586317.7A CN202310586317A CN116634401A CN 116634401 A CN116634401 A CN 116634401A CN 202310586317 A CN202310586317 A CN 202310586317A CN 116634401 A CN116634401 A CN 116634401A
- Authority
- CN
- China
- Prior art keywords
- task
- vehicle
- unloading
- particle
- rsu
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000004364 calculation method Methods 0.000 title claims abstract description 28
- 238000000034 method Methods 0.000 title claims abstract description 21
- 239000002245 particle Substances 0.000 claims abstract description 94
- 238000005457 optimization Methods 0.000 claims abstract description 14
- 238000013461 design Methods 0.000 claims abstract description 6
- 238000010276 construction Methods 0.000 claims abstract description 4
- 238000004891 communication Methods 0.000 claims description 49
- 238000012545 processing Methods 0.000 claims description 44
- 230000006870 function Effects 0.000 claims description 12
- 238000013468 resource allocation Methods 0.000 claims description 5
- 239000004973 liquid crystal related substance Substances 0.000 description 14
- 230000005540 biological transmission Effects 0.000 description 12
- 238000009825 accumulation Methods 0.000 description 2
- 230000001174 ascending effect Effects 0.000 description 2
- 230000001186 cumulative effect Effects 0.000 description 2
- 230000007547 defect Effects 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 230000002708 enhancing effect Effects 0.000 description 2
- 238000011144 upstream manufacturing Methods 0.000 description 2
- 206010063385 Intellectualisation Diseases 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000003247 decreasing effect Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000006855 networking Effects 0.000 description 1
- 230000035945 sensitivity Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04W—WIRELESS COMMUNICATION NETWORKS
- H04W4/00—Services specially adapted for wireless communication networks; Facilities therefor
- H04W4/30—Services specially adapted for particular environments, situations or purposes
- H04W4/40—Services specially adapted for particular environments, situations or purposes for vehicles, e.g. vehicle-to-pedestrians [V2P]
- H04W4/44—Services specially adapted for particular environments, situations or purposes for vehicles, e.g. vehicle-to-pedestrians [V2P] for communication between vehicles and infrastructures, e.g. vehicle-to-cloud [V2C] or vehicle-to-home [V2H]
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04W—WIRELESS COMMUNICATION NETWORKS
- H04W24/00—Supervisory, monitoring or testing arrangements
- H04W24/02—Arrangements for optimising operational condition
Landscapes
- Engineering & Computer Science (AREA)
- Computer Networks & Wireless Communication (AREA)
- Signal Processing (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a task unloading method for maximizing vehicle-mounted user satisfaction under edge calculation, belonging to the field of edge calculation and optimization algorithms, wherein the method comprises 3 stages: model construction stage, target optimal design stage and task unloading stage. The model construction stage is to build a three-layer model framework with resource constraint for vehicle edge calculation, wherein the three-layer model framework comprises a cloud, roadside units and a vehicle, and the vehicle carries a plurality of independent tasks to be unloaded; the target optimization design stage deeply considers the deadline of the task, the resource demand, the resource constraint of each node and the task deadline constraint, and takes the maximized user satisfaction degree as an optimization target; the task unloading stage provides a task unloading algorithm based on improved particle swarms, which aims to reduce task execution delay and improve the rate of completion within the maximum deadline of the task, and maximize user satisfaction.
Description
Technical Field
The invention relates to a task unloading method for maximizing satisfaction of vehicle-mounted users under edge calculation, belonging to the field of edge calculation and optimization algorithms.
Background
With the rapid development of wireless communication and artificial intelligence, in the car networking era, people are required to obtain a safe and rapid experience in the middle of driving, which increases the demands for interconnection and intellectualization between vehicles, and in order to solve the demands, delay-sensitive automobile applications such as auxiliary driving, automatic driving, live navigation, road condition recognition and the like are designed and developed. These powerful applications typically create a large number of tasks, requiring a large amount of computing resources. However, since the computing resources of a single vehicle are very limited, the quality of service (QoS) requirements of the application at low latency cannot be guaranteed, and the application becomes a bottleneck for the development of the internet of vehicles.
Vehicle Edge Computing (VEC) provides a practical and efficient solution to guaranteeing quality of service for large-scale vehicle applications, largely solving the problem of computational resource limitations. The information of nodes and tasks in the area is counted periodically by a central controller (RSU-C) of a roadside unit (RSU), a part of computationally intensive tasks are offloaded to the roadside unit (RSU) provided with an edge server (RES) or to a remote cloud, so that more powerful processing capacity is provided for the vehicle, and meanwhile, the processing delay of the tasks is reduced, so that mass data of delay-sensitive applications are processed more efficiently, and user satisfaction is improved. However, it should be noted that the RSU and cloud serve multiple vehicles simultaneously, and how to efficiently utilize limited node computing resources and achieve maximum all user satisfaction is a critical issue, both as resource requirers and as resource providers.
In view of the above, the present invention proposes a task offloading method for maximizing vehicle user satisfaction under edge computing to solve the above-mentioned problems, aiming at the defects of weak local computing resources of the vehicle and the delay sensitivity of the task.
Disclosure of Invention
The invention aims to provide a task unloading method for maximizing vehicle-mounted user satisfaction under edge calculation, which is used for reducing task delay as much as possible and increasing the rate of task deadline completion and improving user satisfaction by reasonably distributing task execution nodes.
In order to achieve the above object, the present invention provides a task offloading method for maximizing satisfaction of a vehicle-mounted user under edge calculation, the method comprising the steps of: A. model construction stage: according to the vehicle-mounted task unloading requirement under the edge computing, a three-layer vehicle-edge-cloud (VEC) frame model with communication and unloading functions is established, wherein the three-layer model comprises: cloud, roadside units (edge devices), and vehicles; periodically updating the computing resource residual quantity (namely the processing frequency of each device) of each node by a central controller (RSU-C) of a roadside unit (RSU), wherein the computing resource residual quantity comprises the computing resource residual quantity of a remote cloud (C), the computing resource residual quantity of the roadside unit (RSU) and the computing resource residual quantity of all vehicles (V) in the coverage area of the RSU, and simultaneously, the RSU-C also counts the task (T) characteristic parameter information which needs to be unloaded by each vehicle, wherein the task characteristic parameter information comprises the task size, the ratio of the size of the input data to the size of the output data of the task, the task processing density, the task unit deadline, the task deadline limit coefficient, the task category and the task priority;
B. target design stage: defining an unloading decision variable and a user satisfaction calculating method, and designing a target optimization function according to constraint conditions such as the residual quantity of calculation resources, the deadline of a task, the task priority and the like;
C. task unloading stage: after the RSU-C counts the residual computing resource information of the equipment and the vehicle task information at the starting moment of each time slot, the unloading decision, the communication resource, the computing resource, the task deadline, the priority and the like are comprehensively considered, an unloading scheme is executed on the task by adopting an unloading algorithm based on improved particle swarm optimization, the task delay is reduced as much as possible, the completion rate in the task deadline is improved, and the user satisfaction is maximized.
The model building stage specifically comprises the following steps:
A1. the method comprises the steps of establishing a system model for unloading a vehicle edge calculation task, wherein the system model comprises three parts: the maximum processing capacity of the remote cloud C is F c Maximum processing capacity of roadside unit R is F e When (when)The vehicle set covered by the front roadside units is V= { V 1 ,V 2 ,…,V K Vehicle V n The maximum processing capacity of (1.ltoreq.n.ltoreq.K) isWherein the vehicle V n Is expressed as a set of tasks of (1)M n Representing vehicle V n Is the total number of tasks of vehicle V n Task of (1)>Represented by a seven-tuple: /> Wherein->Respectively representing the size of a task, the ratio of the size of input data to the size of output data, the processing density, the deadline, the delay limit coefficient, the category and the priority;
A2. constructing a communication and calculation model in vehicle edge calculation, wherein the communication comprises three parts: communication model between vehiclesAnd->Communication model between vehicle and RSU->And->Communication model R of RSU and remote cloud r2c And R is c2r ；
Communication modelAnd->For the demand vehicle V n And a target vehicle V o Wireless communication rate of communication model between:
wherein W is n,o Is a vehicle V n And V o Channel bandwidth, mu n Is V n Transmission power, mu o Is V o Zeta of transmission power of (c) 2 Is background noise power, A 0 = -17.8dB is a constant coefficient, L n,o Is V n And V o Is used for the distance of the Europe type (R),is V n And V o Channel gain of>
Communication modelUplink wireless communication rate for a communication model between a vehicle and a roadside unit:
wherein W is up Is the upstream channel bandwidth, mu n Is V n Zeta of transmission power of (c) 2 Is background noise power ρ n Due to the radio interference caused by a plurality of vehicles,is V n And a channel gain between the RSU, wherein +.>The communication module->Downlink wireless communication rate for a communication model between a vehicle and a roadside unit:
wherein W is down Is the downlink channel bandwidth, mu r Is the transmission power of RSU, take
The communication model R r2c And R is c2r Wireless communication rate for the communication model between RSU and remote cloud:
wherein W is r2c Channel bandwidth, μ, of RSU and cloud r Is the transmission power, mu, of the RSU c Is the transmission power of the cloud ζ 2 Is background noise power g r2c Is the channel gain of RSU and cloud, taking g r2c ＝g c2r ；
A3. The computing model consists of four parts, including tasks local to the vehicleProcessing timeProcessing time for unloading to target vehicle->Processing time offloaded to RSU->Offloading processing time to cloud->
Processing time of tasks locally on the vehicle:
wherein, the liquid crystal display device comprises a liquid crystal display device,representing that a local embedded processor of a vehicle is assigned to a local task +.>I.e. number of CPU cycles per second);
processing time for task offloading to target vehicle:wherein (1)>Indicating the assignment of the target vehicle to the mission>Is a processing power of the (a);
processing time for task offloading to roadside units:wherein (1)>Indicating allocation of RSU to task->Is a processing power of the (a);
processing time for task offloading to remote cloud:wherein (1)>Representing remote cloud allocation to tasks>Is a processing power of the (a);
B. the specific steps of the target design stage are as follows:
B1. definition of decision variablesAnd user satisfaction->
B2. Considering constraint of task delay and priority, designing an objective function to maximize user satisfaction, and distributing limiting resources according to computing resources and task constraint; the method comprises the following steps:
defining task offloading decision variables:representing vehicle V p Task of (1)>Unloading to node q, where q= -1 represents a remote cloud, q = 0 represents a roadside unit, q= {1,2, …, K } represents a vehicle V q ；
Defining a user satisfaction calculation formula:otherwiseWherein [ x ]] + ＝max{x,0}，/> The smaller the priority, the higher the priority, the value range is [1, p max ]；
Wherein, the liquid crystal display device comprises a liquid crystal display device,representing tasks calculated according to the communication model>Completion time of->Representing task->Is a maximum deadline of (2); defining an objective function: />Because of the limited computational resources of the vehicle, the resource constraints are: /> Wherein (1)>Representing vehicle V q Assigned to task->Is->Representing vehicle V q Left amount of computing resources, +.>Representing vehicle V q Is the maximum amount of computational resources;
because the RSU has limited computational resources, the resource constraints are: wherein (1)>Indicating allocation of RSU to task->F e Representing the amount of remaining computing resources of the RSU, F e Representing the maximum amount of computing resources of the roadside unit R;
because the computing resources of the remote cloud are limited, the resource constraints are: wherein (1)>Representation cloud allocation to tasks>F c Representing the amount of remaining computing resources of the cloud, F c Representing a maximum amount of computing resources of the remote cloud C;
the task allocation constraint of a vehicle indicates that any task can be allocated to only one node: wherein (1)>
C. Task unloading stage: the task unloading stage comprises the following specific steps:
the RSU-C uses an improved particle swarm optimization algorithm to find an unloading decision scheme, abstracts tasks to be unloaded and residual computing resources of all nodes in all vehicles into particles, defines particle representation of a population, initializes population parameters, and comprises setting population scale, population history optimal record number, particle dimension, maximum iteration number, inertia weight, individual learning factors, population learning factors, particle position boundary range and particle speed range;
C2. randomly initializing the position and speed of each particle, recording the individual history optimal position, the population history optimal position, the individual history optimal target value and the population history optimal target value of the particle, and initializing the individual history optimal position, the population history optimal position, the individual history optimal target value and the population history optimal target value by combining a roulette strategy;
C3. judging whether an ending condition is met, wherein the ending condition comprises that the maximum iteration times are reached, if yes, ending the iteration, outputting an optimal solution, unloading a task to a target node by the RSU-C according to an unloading decision and resource allocation represented by particles of the optimal solution, and processing, otherwise executing the step C4;
C4. updating the position and the speed of each particle, namely updating the unloading decision of each particle and the change degree of the optimizing direction of each particle, and affecting the unloading decision and the resource allocation through the speed change of the particle so as to further optimize in the search space;
C5. carrying task characteristic information and particle unloading decisions into a user satisfaction formula by combining a communication model to calculate the fitness value of each particle;
C6. updating the individual historical optimal position and optimal fitness value of each particle, and recording the current optimal unloading decision and user satisfaction of each particle;
C7. updating the historical optimal position and optimal fitness value of the population by combining with a roulette strategy, and recording the optimal unloading decision and user satisfaction of the population;
C8. boundary condition processing, namely checking whether particles exceed a position range defined by constraint, ensuring that unloading decisions represented by the particles are within a reasonable node range and meet computational resource constraint conditions, if so, reinitializing according to the step C2, otherwise, executing the step C9;
C9. updating other parameters in the particle swarm algorithm, including inertia weight, learning factors, iteration times and the like, searching the performance of the optimal decision according to the particle swarm algorithm at different stages, updating the speed change of particles affected by the change degree of the weight and the learning factors so as to iteratively optimize the unloading decision, and returning to the step C3 to continue iterative optimization.
Wherein, the step C1 comprises the following steps:
step C11 particles of the population are denoted S i (g)＝{λ,f q ,f v ,f e ,f c λ for each particle represents the offloading decision for all tasks, f q Representing the amount of computing resources allocated to a task by node q, which is essentially an array, with element subscripts being the task number, the former element value being the target node for task offloading and the latter element value being the allocated resources, f v 、f e 、f c Representing the residual computing resources of each node, and searching a solution vector maximizing user satisfaction in a vector space by each particle;
step c12. Population size s, particle dimension d=2×Σ p∈{V} M p +K+2, population history optimum number The maximum iteration number is g max Inertia weight is omega, individual learning factor is sigma 1 And population learning factor sigma 2 The position range of the particles is x min And x max The velocity range of the particles is v min And v max ；
Wherein the front d of the particle 1 ＝∑ p∈{V} M p Dimension represents offloading decisions for tasks, middle d 2 ＝∑ p∈{V} M p Dimension represents the amount of computing resources allocated to a task, d 3 The dimension =k+2 represents the amount of remaining computing resources for each node, and is used to determine the resource constraint for the current slot, where the location range and the speed range of the particles are used to constrain the previous d 1 Dimensional change, middle d 2 The value of the dimension needs to meet the resource constraint condition, and d is the later 3 The dimension change does not affect the experimental result, and is only used for counting the residual resources of the nodes by the RSU-C at the beginning of the time slot;
wherein, step C2 comprises the following steps:
C21. initializing the velocity of each particle to V i ＝{v i,1 ,v i,2 ,…,v i,d Position of each particle is S i ＝{s i,1 ,s i,2 ,…,s i,d }；
C22. The individual historical optimal positions of the particles are a setThe optimal position of the population history is set->The optimal objective of individual history is set-> Population history optimal goal is set->Particles S to be initialized i (g) Is assigned to P by a value of (2) indv And from P indv Selecting s based on roulette strategy-1 glb Assignment of individual elements to P glb ，O indv And O glb The initial assignment is 0, the objective function is user satisfaction, and the larger and better the value range is [0, theta-1]The method comprises the steps of carrying out a first treatment on the surface of the C23. The roulette strategy-1 comprises the steps of:
C231. calculating the probability that the element is selected:here take f (x) i ) Is the fitness;
C232. element accumulation selected probabilities:
C233. randomly generating an array sel, and selecting element value range [0,1 ]]And are arranged in ascending order, if the cumulative probability q (x i ) Greater than element sel i in the array]Individual x i Is selected to be less than sel [ i ]]Then the next individual x is compared i+1 The size of the array sel determines the number of times, here s, until an individual is selected glb ；
If the algorithm iteration number in the step C3 reaches the maximum iteration number g max Will be assembled P glb The optimal solution particles in the (a) represent the variables mapped to the objective function, the RSU-C downloads the task to the objective node according to the write-in decision and resource allocation represented by the particles of the optimal solution, and updates the amount of the left computing resources of the node, otherwise, the step C4 is continuously executed;
wherein, the step C4 comprises the following steps:
C41. from the slaveAnd P glb A solution is randomly selected from +.> And->A representation;
C42. for each particle, the current position is S i (g)＝{s i,1 (g),s i,2 (g),…,s i,d (g) Current speed is V i (g)＝{v i,1 (g),v i,2 (g),…,v i,d (g)}；
C43. The speed update formula is:
wherein rand is 1 And rand 2 Is [0,1 ]]Random numbers in between;
C44. the location update formula is:
the fitness value formula of each particle calculated in the step C5 is as follows:
if S in said step C6 i (g) The fitness value is higher thanDescription particle S i (g) Is better than the individual history optimal solution, S is i (g) Added to->In the collection, and target value +.>Added to->In the set, recording the current optimal unloading scheme and the optimal target value;
if S in said step C7 i (g) The fitness value is higher than P glb Any solution of (3)Explaining the decision of the current particle due to one of the solutions in the history optimal set, the roulette strategy-2 is adopted at P glb Selecting an element from the set to be S i (g) Replacement and target value +.>Added to O glb In the set, recording a historical optimal unloading decision and a corresponding target value of the population, and updating the historical optimal position to reduce the probability of the particle swarm algorithm falling into local optimal by using a roulette strategy;
in the roulette strategy-2, the probability calculation formula of the selected particles is:here the sel array size takes a value of 1, the rest of the operations are the same as roulette strategy-1;
in said step C8 for each particle S i (g) Judgment element s i,ε (g) Whether the constraint definition is exceeded, i.e. whether the target node represented by the particle offload decision is in the model scope, and whether the resource constraint and allocation constraint are met, ensures that RSU-C can successfully execute the decision, if the position scope is exceeded, reinitialization is performed according to step C2, after which the population is represented as S "(g) = { S" 1 (g),S″ 2 (g),…,S″ s (g) -else, executing step C9;
the step C9 comprises the following steps:
C91. more, theNew inertial parameters:
C92. updating the learning factor:
control ω is continuously decreasing, σ 1 Gradually decrease, sigma 1 Gradually increasing, namely controlling particles to pay attention to self-cognition in the early stage of iteration, enhancing the ergodic property of the particles, enhancing the communication among the particles in the later stage of iteration, and approaching to the global optimal solution of the population;
C93. updating iteration parameters: g=g+1;
C94. and C3, continuing to iterate and optimizing.
Compared with the prior art, the invention has the following advantages: according to the invention, task unloading in the VEC environment is analyzed in detail, and the idle or underutilized computing resources of the vehicle, RSU and cloud are considered, so that the computing intensive tasks of the vehicle are allocated with proper computing resources for unloading in combination with the demand characteristics of different tasks, the processing time delay of the tasks is reduced, the defect of limited computing capacity of the vehicle is overcome, and the service satisfaction of vehicle-mounted users is improved.
Drawings
FIG. 1 is a schematic diagram of a model of task offloading based on vehicle edge calculation of the present invention.
FIG. 2 is a task offloading flow framework of a system model.
Fig. 3 is a flow chart of an improved particle swarm optimization algorithm applied to the model of the present invention.
Detailed Description
The invention will be described in further detail with reference to the drawings and the specific embodiments.
Examples: as shown in fig. 1, the invention builds a system model for vehicle edge computing task offloading, consisting of three parts: the maximum processing capacity of the remote cloud C is F c Maximum processing capacity of roadside unit R is F e The vehicle set covered by the current roadside unit is V= { V 1 ,V 2 ,…,V K Vehicle V n The maximum processing capacity of (1.ltoreq.n.ltoreq.K) isWherein the vehicle V n Is expressed as +.>M n Representing vehicle V n Is the total number of tasks of vehicle V n Task of (1)>Represented by a seven-tuple: />Wherein-> The task size, the ratio of the input data to the output data size, the processing density, the deadline, the delay limit coefficient, the category and the priority are respectively represented.
It is assumed here that the node and task information is as follows:
node table:
| node numbering | Node representation | Task number | Computing resources |
| -1 | C | - | 50000 |
| 0 | R | - | 10000 |
| 1 | V 1 | 3 | 500 |
| 2 | V 2 | 2 | 1400 |
| 3 | V 3 | 4 | 300 |
Task table:
the particle codes correspond to target nodes for unloading decision according to task numbers;
as shown in fig. 2, the general flow of task offloading of the system model is divided into the following steps:
and periodically updating the computing resource information of each node by a central control unit (RSU-C) of the roadside unit, and counting the task characteristic parameter information.
Abstracting a communication model and a calculation model of task unloading, wherein the abstracting comprises the following steps:
constructing a communication and calculation model in vehicle edge calculation, wherein the communication comprises three parts: communication model between vehiclesAnd->Communication model between vehicle and RSU->And->Communication model R of RSU and remote cloud r2c And R is c2r The method comprises the steps of carrying out a first treatment on the surface of the Communication model->And->For the demand vehicle V n And a target vehicle V o Wireless communication rate of communication model between:
wherein W is n,o Is a vehicle V n And V o Channel band therebetweenWide, mu n Is V n Transmission power, mu o Is V o Zeta of transmission power of (c) 2 Is background noise power, A 0 = -17.8dB is a constant coefficient, L n,o Is V n And V o Is used for the distance of the Europe type (R),is V n And V o Channel gain of>
Communication modelUplink wireless communication rate for a communication model between a vehicle and a roadside unit:
wherein W is up Is the upstream channel bandwidth, mu n Is V n Zeta of transmission power of (c) 2 Is background noise power ρ n Due to the wireless interference of the multiple vehicle bands,is V n And the channel gain between the RSU, the wireless interference calculation formula is +.>
The communication modelDownlink wireless communication rate for a communication model between a vehicle and a roadside unit:
wherein W is down Is the downlink channel bandwidth, mu r Is the transmission power of RSU, take
The communication model R r2c And R is c2r Wireless communication rate for the communication model between RSU and remote cloud:
wherein W is r2c Channel bandwidth, μ, of RSU and cloud r Is the transmission power, mu, of the RSU c Is the transmission power of the cloud ζ 2 Is background noise power g r2c Is the channel gain of RSU and cloud, taking g r2c ＝g c2r ；
The calculation model consists of four parts including the processing time of the task on the vehicleProcessing time for unloading to target vehicle->Processing time offloaded to RSU->Offloading processing time to cloud->The local processing time of the task at the vehicle: />
Wherein, the liquid crystal display device comprises a liquid crystal display device,representing that a local embedded processor of a vehicle is assigned to a local task +.>I.e., number of CPU cycles per second);
processing time for task offloading to target vehicle:
wherein, the liquid crystal display device comprises a liquid crystal display device,representing a target vehicle V o Assigned to task->Is used for calculating the resource quantity;
processing time for task offloading to roadside units:
wherein, the liquid crystal display device comprises a liquid crystal display device,indicating allocation of RSU to task->Is used for calculating the resource quantity;
processing time for task offloading to remote cloud:
wherein, the liquid crystal display device comprises a liquid crystal display device,representing remote cloud allocation to tasks>Is used for calculating the resource quantity;
defining task offloading decision variables:representing vehicle V p Task of (1)>Unloading to node q, where q= -1 represents a remote cloud, q = 0 represents a roadside unit, q= {1,2, …, K } represents a vehicle V q ；
Defining a user satisfaction calculation formula:otherwiseWherein [ x ]] + ＝max{x,0}，/> The smaller the priority, the higher the value range is [1, p max ]；
Defining an objective function:
because of the limited computational resources of the vehicle, the resource constraints are: wherein (1)>Representing vehicle V q Assigned to task->Is->Representing vehicle V q Left amount of computing resources, +.>Representing vehicle V q Is the maximum amount of computational resources;
because the RSU has limited computational resources, the resource constraints are: wherein (1)>Indicating allocation of RSU to task->F e Representing the amount of remaining computing resources of the RSU, F e Representing the maximum amount of computing resources of the roadside unit R;
because the computing resources of the remote cloud are limited, the resource constraints are: wherein (1)>Representation cloud allocation to tasks>F c Representing the amount of remaining computing resources of the cloud, F c Representing a maximum amount of computing resources of the remote cloud C;
the task allocation constraint of a vehicle indicates that any task can be allocated to only one node: wherein (1)>
As shown in fig. 3, the task offloading scheme is obtained by adopting an improved particle swarm optimization algorithm, and the steps are as follows:
the particles defining the population are denoted S i (g)＝{λ,f q ,f v ,f e ,f c }，
Where λ is the offload decision variable, the node to which the decision task is offloaded, f q Representing the amount of computing resources allocated to a task by node q, f v Representing remaining computing resources of the vehicle, f e Representing the remaining computing resources of the roadside units, f c The cloud residual computing resources are represented, and the units of the computing resources are represented as CPU cycles/s;
population size s, particle dimension d=2 Σ p∈{V} M p +K+2, population history optimum numberThe maximum iteration number is g max Inertia weight is omega, individual learning factor is sigma 1 And population learning factor sigma 2 The position range of the particles is x min And x max The velocity range of the particles is v min And v max ；
Wherein, the s initialization range is [20,200 ]]Front d of particle 1 ＝∑ p∈{V} M p The dimension represents the allocation decision of all task offloading, the middle d 2 ＝∑ p∈{V} M p Dimension represents the computing resources allocated to the task by each node, d 3 The =k+2 dimension represents the remaining computing resources of each node (vehicle, roadside unit, cloud);
C2. randomly initializing the position and the speed of each particle, setting four memory units, recording the individual history optimal position, the population history optimal position, the individual history optimal target value and the population history optimal target value of the particle, and initializing the memory units of the population by combining a roulette strategy, wherein the method comprises the following steps:
C21. initializing the velocity of each particle to V i ＝{v i,1 ,v i,2 ,…,v i,d Position of each particle is S i ＝{s i,1 ,s i,2 ,…,s i,d }；
Assume here that 3 particles are initialized, respectively:
wherein S is i The array subscript of (1) indicates the task number, and the element value indicates the target node of task offloading, e.g. S in λ 1 (0)[1]=1 denotes a task with a task number of 1 (here) Is offloaded to node 1 (here V 1 ) The allocated computing resource is +.>And so on;
C22. recording individual historical optimal positions of particles as a setThe optimal position of the population history is set->The optimal objective of individual history is set-> Population history optimal goal is set->
C23. Particles S to be initialized i (g) Is assigned to P by a value of (2) indv And from P indv Selecting s based on roulette strategy-1 glb Assignment of individual elements to P glb ，O indv And O glb The initial value is 0, the objective function is user satisfaction, the larger the better, the value range is [0, theta-1]；
Wherein the roulette strategy-1 comprises the steps of:
calculating the probability that the element is selected:here take f (x) i ) Is the fitness;
element accumulation selected probabilities:
randomly generating an array sel, and selecting element value range [0,1 ]]And are arranged in ascending order, if the cumulative probability q (x i ) Greater than element sel i in the array]Individual x i Is selected to be less than sel [ i ]]Then the next individual x is compared i+1 The size of the array sel determines the number of times until an individual is selected, where sel size takes s glb ；
C3. If the algorithm iteration number reaches the maximum iteration number g max Will be assembled P glb The optimal solution particle in (a) represents the variable mapped to the objective function, otherwise, executing the step C4;
assuming that the iteration is complete, the resulting optimal particles are represented as follows:
task offloading is performed according to the particle representation;
C4. updating the position and velocity of each particle, comprising the steps of:
C41. from the slaveAnd P glb A solution is randomly selected from +.>/>And->A representation;
C42. for each particle, the current position is S i (g)＝{s i,1 (g),s i,2 (g),…,s i,d (g) Current speed is V i (g)＝{v i,1 (g),v i,2 (g),…,v i,d (g)}；
C43. The formula of the speed update is as follows:
wherein rand is 1 And rand 2 Is [0,1 ]]Random numbers in between;
C44. and (3) position updating:
C5. calculating the fitness value of each particle:
C6. updating individual historical optimal fitness values and optimal locations for each particle if S i (g) The fitness value is higher thanWill S i (g) Added to->In the collection, and target value +.>Added to->In the collection;
C7. updating the historical optimal fitness value and optimal position of the population in combination with the roulette strategy if S i (g) The fitness value is higher than P glb Any solution of (3)Then use roulette strategy-2 at P glb Selecting an element from the set to be S i (g) Replacement and target value +.>Added to O glb In the collection;
in the roulette strategy-2, the probability calculation formula of the selected particles is:here the sel array size takes 1, the rest of the operations are the same as roulette strategy-1;
C8. boundary condition processing for each particle S i (g) Judgment element s i,ε (g) If the constraint definition is exceeded, the location is reinitialized according to step C2, after which the population is represented as S "(g) = { S" 1 (g),S″ 2 (g),…,S″ s (g) -else, executing step C9;
C9. updating other parameters, including the following:
C91. updating inertia parameters:
wherein w is max ＝0.95，w min ＝0.4；
C92. Updating the learning factor:
wherein, the liquid crystal display device comprises a liquid crystal display device,
C93. updating iteration parameters: g=g+1;
C94. and C3, continuing to iterate and optimizing.
It should be noted that the above-mentioned embodiments are not intended to limit the scope of the present invention, and equivalent changes or substitutions made on the basis of the above-mentioned technical solutions fall within the scope of the present invention as defined in the claims.
Claims (4)
1. A method for task offloading to maximize vehicle user satisfaction with edge computation, the method comprising the steps of:
A. model construction stage: according to the vehicle-mounted task unloading requirement under the edge computing, a three-layer vehicle-edge-cloud (VEC) frame model with communication and unloading functions is established, wherein the three-layer model comprises: cloud, roadside units (edge devices), and vehicles; periodically updating the computing resource residual quantity (namely the processing frequency of each device) of each node by a central controller (RSU-C) of a roadside unit (RSU), wherein the computing resource residual quantity comprises the computing resource residual quantity of a remote cloud (C), the computing resource residual quantity of the roadside unit (RSU) and the computing resource residual quantity of all vehicles (V) in the coverage area of the RSU, and simultaneously, the RSU-C also counts the task (T) characteristic parameter information which needs to be unloaded by each vehicle, wherein the task characteristic parameter information comprises the task size, the ratio of the size of the input data to the size of the output data of the task, the task processing density, the task unit deadline, the task deadline limit coefficient, the task category and the task priority;
B. target design stage: defining an unloading decision variable and a user satisfaction calculating method, and designing a target optimization function according to constraint conditions such as the residual quantity of calculation resources, the deadline of a task, the task priority and the like;
C. task unloading stage: after the RSU-C counts the residual computing resource information of the equipment and the vehicle task information at the starting moment of each time slot, the unloading decision, the communication resource, the computing resource, the task deadline, the priority and the like are comprehensively considered, an unloading scheme is executed on the task by adopting an unloading algorithm based on improved particle swarm optimization, the task delay is reduced as much as possible, the completion rate in the task deadline is improved, and the user satisfaction is maximized.
2. The task offloading method of maximizing vehicle user satisfaction under edge computing of claim 1, wherein: A. the specific steps of the model establishment stage are as follows:
A1. establishing a vehicle-mounted task unloading system model under edge calculation, wherein the vehicle-mounted task unloading system model consists of three parts: the processing capacity of the remote cloud C is F c The processing capacity of the roadside unit R is F e The vehicle set covered by the roadside units is V= { V 1 ,V 2 ,…,V K Vehicle V n The treatment capacity of (1. Ltoreq.n.ltoreq.K) isWherein the vehicle V n Is expressed as +.>M n Representing vehicle V n Is the total number of tasks of vehicle V n Task of (1)>Represented by a seven-tuple: />Wherein->Respectively representing the size of a task, the ratio of the size of input data to the size of output data, the processing density, the deadline, the delay limit coefficient, the category and the priority;
A2. constructing a communication and calculation model in vehicle edge calculation, the communication being composed of three parts, namely the vehicle V n And V is equal to o Communication model betweenAnd->Communication model between vehicle and RSU->And->Communication model R of RSU and remote cloud r2c And R is c2r ；
A3. The calculation model consists of four parts including local processing time of tasks on the vehicleProcessing time for unloading to other vehicles->Processing time offloaded to RSU->Unloading toProcessing time of remote cloud->
3. The task offloading method of maximizing vehicle user satisfaction under edge computing of claim 1, wherein:
B. the specific steps of the target design stage are as follows:
B1. definition of decision variablesAnd user satisfaction->
B2. The objective function is designed to maximize user satisfaction in view of task delay and priority constraints, and defined resources are allocated according to computing resources and task constraints.
4. The task offloading method of maximizing vehicle user satisfaction under edge computing of claim 1, wherein: C. the task unloading stage comprises the following specific steps:
using an improved particle swarm optimization algorithm to find an unloading decision scheme by the RSU-C, abstracting all tasks to be unloaded and the residual computing resources of each node into particles, and defining the particles of the population to be expressed as S i (g)＝{λ,f q ,f v ,f e ,f c λ represents the offloading decision of all tasks, the target node that decides the offloading of the task, f q Representing the amount of computing resources allocated to a task by node q, f v 、f e 、f c Representing the residual computing resources of each node, initializing population parameters including setting population scale s and population history optimal record number s glb Particle dimension d, maximum number of iterations g max Inertial weight omega, individual learning factor sigma 1 And population learning factor sigma 2 Boundary x of particle position min And x max Particle velocity range v min And v max ；
C2. Randomly initializing the position S of each particle i ＝{s i,1 ,s i,2 ,…,s i,d Sum of velocity V i ＝{v i,1 ,v i,2 ,…,v i,d Recording the historic optimal position of the particlePopulation history optimal positionIndividual history optimal target value-> Population history optimal target value->Initializing P in conjunction with roulette strategy indv 、p glb 、O indv 、O glb ；
C3. Judging whether the algorithm meets the ending condition, namely ending iteration if the maximum iteration number is reached, outputting an optimal solution, unloading the task to a target node for processing by the RSU-C according to an unloading decision and computing resource allocation represented by particles of the optimal solution, and executing the step C4 otherwise;
C4. updating the position and the speed of each particle, namely updating the unloading decision of each task and the computing resources allocated to the task, and affecting the unloading decision and the resource allocation through the speed change of the particle to further optimize in the search space;
C5. carrying task characteristic information and particle unloading decisions into a user satisfaction formula by combining a communication model to calculate the fitness value of each particle;
C6. updating the individual historical optimal position and optimal fitness value of each particle, and recording the current optimal unloading decision and user satisfaction of each particle;
C7. updating the historical optimal position and optimal fitness value of the population by combining with a roulette strategy, and recording the current optimal unloading decision and user satisfaction of the population;
C8. boundary condition processing, namely checking whether the particles exceed the position range defined by the constraint, ensuring that the unloading decision represented by the particles is within a reasonable node range and meets the constraint of computing resources, if so, reinitializing according to the step C2, otherwise, executing the step C9;
C9. updating other parameters of the particle swarm algorithm, including inertia weight, learning factors, iteration times and the like, searching the performance of the optimal decision according to the particle swarm algorithm at different stages, updating the speed change of particles affected by the change degree of the weight and the learning factors to iteratively optimize the unloading decision, and returning to the step C3 to continue iterative optimization.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310586317.7A CN116634401A (en) | 2023-05-23 | 2023-05-23 | Task unloading method for maximizing satisfaction of vehicle-mounted user under edge calculation |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310586317.7A CN116634401A (en) | 2023-05-23 | 2023-05-23 | Task unloading method for maximizing satisfaction of vehicle-mounted user under edge calculation |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116634401A true CN116634401A (en) | 2023-08-22 |
Family
ID=87609326
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310586317.7A Pending CN116634401A (en) | 2023-05-23 | 2023-05-23 | Task unloading method for maximizing satisfaction of vehicle-mounted user under edge calculation |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116634401A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117648172A (en) * | 2024-01-26 | 2024-03-05 | 南京邮电大学 | Vehicle-mounted edge calculation scheduling optimization method and system |
-
2023
- 2023-05-23 CN CN202310586317.7A patent/CN116634401A/en active Pending
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117648172A (en) * | 2024-01-26 | 2024-03-05 | 南京邮电大学 | Vehicle-mounted edge calculation scheduling optimization method and system |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111240701B (en) | Task unloading optimization method for end-side-cloud collaborative computing | |
| CN109656703B (en) | Method for assisting vehicle task unloading through mobile edge calculation | |
| CN112188442A (en) | Vehicle networking data-driven task unloading system and method based on mobile edge calculation | |
| CN111711666B (en) | Internet of vehicles cloud computing resource optimization method based on reinforcement learning | |
| CN111401744B (en) | Dynamic task unloading method in uncertainty environment in mobile edge calculation | |
| CN112911648A (en) | Air-ground combined mobile edge calculation unloading optimization method | |
| CN112995289B (en) | Internet of vehicles multi-target computing task unloading scheduling method based on non-dominated sorting genetic strategy | |
| CN111757361B (en) | Task unloading method based on unmanned aerial vehicle assistance in fog network | |
| CN111813506A (en) | Resource sensing calculation migration method, device and medium based on particle swarm algorithm | |
| CN115297171B (en) | Edge computing and unloading method and system for hierarchical decision of cellular Internet of vehicles | |
| CN113867843B (en) | Mobile edge computing task unloading method based on deep reinforcement learning | |
| CN116634401A (en) | Task unloading method for maximizing satisfaction of vehicle-mounted user under edge calculation | |
| CN113645273B (en) | Internet of vehicles task unloading method based on service priority | |
| CN113766037B (en) | Task unloading control method and system for large-scale edge computing system | |
| CN114172558B (en) | Task unloading method based on edge calculation and unmanned aerial vehicle cluster cooperation in vehicle network | |
| CN115629873A (en) | System and method for controlling unloading of vehicle-road cloud cooperative tasks and stability of task queue | |
| CN114641041A (en) | Edge-intelligent-oriented Internet of vehicles slicing method and device | |
| CN112040512A (en) | Fair-based fog computing task unloading method and system | |
| CN113709249B (en) | Safe balanced unloading method and system for driving assisting service | |
| CN112596910B (en) | Cloud computing resource scheduling method in multi-user MEC system | |
| CN116916386A (en) | Large model auxiliary edge task unloading method considering user competition and load | |
| CN115499875B (en) | Satellite internet task unloading method, system and readable storage medium | |
| CN117528649A (en) | Method for establishing end-edge cloud system architecture, task unloading and resource allocation optimization method and end-edge cloud system architecture | |
| CN115514769B (en) | Satellite elastic Internet resource scheduling method, system, computer equipment and medium | |
| CN115134370B (en) | Multi-unmanned aerial vehicle assisted mobile edge computing and unloading method |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |