CN112995289B - Internet of vehicles multi-target computing task unloading scheduling method based on non-dominated sorting genetic strategy - Google Patents

Internet of vehicles multi-target computing task unloading scheduling method based on non-dominated sorting genetic strategy Download PDF

Info

Publication number
CN112995289B
CN112995289B CN202110156669.XA CN202110156669A CN112995289B CN 112995289 B CN112995289 B CN 112995289B CN 202110156669 A CN202110156669 A CN 202110156669A CN 112995289 B CN112995289 B CN 112995289B
Authority
CN
China
Prior art keywords
vehicle
node
task
subtask
energy consumption
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.)
Expired - Fee Related
Application number
CN202110156669.XA
Other languages
Chinese (zh)
Other versions
CN112995289A (en
Inventor
张德干
董文淼
朴铭杰
张婷
张捷
宋金杰
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Tianjin University of Technology
Original Assignee
Tianjin University of Technology
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Tianjin University of Technology filed Critical Tianjin University of Technology
Priority to CN202110156669.XA priority Critical patent/CN112995289B/en
Publication of CN112995289A publication Critical patent/CN112995289A/en
Application granted granted Critical
Publication of CN112995289B publication Critical patent/CN112995289B/en
Expired - Fee Related legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L67/00—Network arrangements or protocols for supporting network services or applications
    • H04L67/01—Protocols
    • H04L67/10—Protocols in which an application is distributed across nodes in the network
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F9/00—Arrangements for program control, e.g. control units
    • G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
    • G06F9/46—Multiprogramming arrangements
    • G06F9/50—Allocation of resources, e.g. of the central processing unit [CPU]
    • G06F9/5061—Partitioning or combining of resources
    • G06F9/5072—Grid computing
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/004—Artificial life, i.e. computing arrangements simulating life
    • G06N3/006—Artificial life, i.e. computing arrangements simulating life based on simulated virtual individual or collective life forms, e.g. social simulations or particle swarm optimisation [PSO]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/12—Computing arrangements based on biological models using genetic models
    • G06N3/126—Evolutionary algorithms, e.g. genetic algorithms or genetic programming
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L41/00—Arrangements for maintenance, administration or management of data switching networks, e.g. of packet switching networks
    • H04L41/14—Network analysis or design
    • H04L41/145—Network analysis or design involving simulating, designing, planning or modelling of a network
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L67/00—Network arrangements or protocols for supporting network services or applications
    • H04L67/01—Protocols
    • H04L67/10—Protocols in which an application is distributed across nodes in the network
    • H04L67/1001—Protocols in which an application is distributed across nodes in the network for accessing one among a plurality of replicated servers
    • H04L67/1004—Server selection for load balancing
    • H04L67/1008—Server selection for load balancing based on parameters of servers, e.g. available memory or workload
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L67/00—Network arrangements or protocols for supporting network services or applications
    • H04L67/01—Protocols
    • H04L67/12—Protocols specially adapted for proprietary or special-purpose networking environments, e.g. medical networks, sensor networks, networks in vehicles or remote metering networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2209/00—Indexing scheme relating to G06F9/00
    • G06F2209/50—Indexing scheme relating to G06F9/50
    • G06F2209/509—Offload
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
    • Y02D10/00—Energy efficient computing, e.g. low power processors, power management or thermal management

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Health & Medical Sciences (AREA)
  • Signal Processing (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Biophysics (AREA)
  • Software Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • General Health & Medical Sciences (AREA)
  • Mathematical Physics (AREA)
  • General Physics & Mathematics (AREA)
  • Computing Systems (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Evolutionary Computation (AREA)
  • Computational Linguistics (AREA)
  • Molecular Biology (AREA)
  • Biomedical Technology (AREA)
  • Artificial Intelligence (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Biology (AREA)
  • Physiology (AREA)
  • Genetics & Genomics (AREA)
  • Computer Hardware Design (AREA)
  • Medical Informatics (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

A multi-objective calculation task unloading scheduling method based on a non-dominated sorting genetic strategy belongs to the field of internet of vehicles, divides a calculation task into small subtasks which have a dependency relationship and can be processed in parallel, and provides a corresponding time delay and energy consumption model. A constraint multi-objective optimization model for calculating task unloading in the Internet of vehicles is built, a non-dominated sorting genetic strategy (NSGS) is used for optimizing an objective function, and new non-dominated relations and constraints are provided for the problem of calculating task unloading in the Internet of vehicles. In addition, a series of experiments were performed and compared with other unloading methods. The experimental result proves that the method solves the problems of time delay and energy consumption of calculation task unloading by using the non-dominated sorting genetic strategy, has better performance and certain practical value compared with other unloading methods.

Description

Internet of vehicles multi-target computing task unloading scheduling method based on non-dominated sorting genetic strategy
Technical Field
The invention belongs to the field of car networking, and particularly relates to a multi-target calculation task unloading scheduling method based on a non-dominated sorting genetic strategy.
Background
In recent years, intelligent transportation systems and automated driving techniques have been continuously developed, and various attractive applications have emerged. The internet of vehicles is widely concerned by people. The large number of compute-intensive and delay-sensitive applications increases the demand for vehicle computing and storage capacity, but due to the constraints of physical space and economic cost, vehicle computing resources and energy consumption are limited and difficult to meet the demands of the applications.
The Mobile Edge Computing (MEC) technology is a very promising technology developed in recent years, in which a vehicle network can use a base station or a Road Side Unit (RSU) to serve a vehicle, and an edge server can be deployed at the base station or the RSU closer to the vehicle, so that the distance between the vehicle and the server becomes short, and the vehicle can offload computing tasks to the edge server, thereby reducing the time delay and energy consumption for processing the computing tasks. Computation task offloading, one of the key technologies for edge computation, can implement more computation-intensive and delay-sensitive applications by offloading computation tasks to vehicles in a vehicle MEC network.
Disclosure of Invention
In order to solve the problem that the unloading calculation task is limited by MEC server resources, the vehicle calculation task is divided into small subtasks with dependency relationships, the divided subtasks can be processed in parallel, a constrained multi-objective optimization model for unloading the vehicle networking calculation task is constructed, a non-dominated ordering genetic strategy (NSGS) is provided to optimize an objective function, a new non-dominated relationship and new constraints are provided specific to the problem of unloading the calculation task in the vehicle networking, the vehicle calculation task can be processed on a vehicle and an edge server at the same time, and time and energy consumption required by calculation task processing are reduced. Furthermore, the effectiveness of the proposed algorithm was demonstrated through a series of experiments and corresponding comparisons with other unloading methods. The experimental results prove that the algorithm provided by the invention has better performance.
The invention discloses a vehicle networking multi-objective computing task unloading scheduling method based on a non-dominated sorting genetic strategy, which mainly comprises the following steps:
1, constructing a vehicle networking system model:
1.1, defining a general noun of the Internet of vehicles system;
1.2, establishing a communication model;
1.3, establishing a calculation time delay model;
1.4, establishing a calculation energy consumption model;
and 2, constructing a task unloading strategy of the Internet of vehicles:
2.1, calculating the scheduling constraint of the task node;
2.2, constructing a problem model according to the objective function;
design of NSGS algorithm:
3.1, designing a vehicle population and initializing according to a genetic algorithm;
3.2, setting constraint conditions;
3.3, sorting individuals by using a rapid non-dominated sorting method;
3.4, calculating the congestion degree;
3.5, designing a crossover and mutation strategy;
and 3.6, designing a population updating strategy.
Further, in step 1.1, a general term of the car networking system is defined, that is, M ═ 1,2, 3.., M } vehicles, N ═ 1,2, 3.., N } channels, and K ═ 1,2, 3.., K } edge servers are located on one road. Vehicles can only be connected with one channel at most in one time period, and each vehicle is restricted to select one MEC server to execute tasks. The computational task of each vehicle is divided into several interdependent subtasks that can be offloaded to the edge server for processing, but some tasks must be performed locally, such as vehicle GPS to obtain the vehicle's position. The first and last subtasks of the vehicle are constrained to be performed locally. The invention proposes to use a directed acyclic graph Dm＝(Vm,Rm) Represents a computational task of the vehicle, wherein
Figure BDA0002934969650000021
Set of all subtasks representing vehicle m, use
Figure BDA0002934969650000022
To represent the subtasks of the vehicle m
Figure BDA0002934969650000023
And
Figure BDA0002934969650000024
the two subtasks are neighbor nodes, and the current subtask
Figure BDA0002934969650000025
Is that
Figure BDA0002934969650000026
In the case of the immediate parent node of (c),
Figure BDA0002934969650000027
must be in
Figure BDA0002934969650000028
And is done before. Additional communication costs are incurred when the two subtasks are offloaded locally and the other is offloaded at the edge server. For subtask node of each vehicle m
Figure BDA0002934969650000029
And (4) showing.
Figure BDA00029349696500000210
The data size of the subtask node i representing the vehicle m,
Figure BDA00029349696500000211
represents the CPU resources consumed by the subtask node i of the vehicle m when performing task offloading locally,
Figure BDA00029349696500000212
and the CPU resource consumed by the subtask node i of the vehicle m when the edge server carries out task unloading is shown.
Figure BDA00029349696500000213
Representing a decision factor, when the decision factor is equal to 1, representing that the subtask i is unloaded at the edge server, and when the decision factor is equal to 0, representing that the subtask i is unloaded at the local server;
the method for establishing the communication model in the step 1.2 comprises the following steps of dividing the vehicle calculation task into subtasks, improving the task unloading parallel capability, and nodularizing the task of the vehicle into nodes
Figure BDA00029349696500000214
And
Figure BDA00029349696500000215
two disjoint sets represent the set of tasks offloaded locally and offloaded at the edge server, respectively. Use of
Figure BDA00029349696500000216
Representing a set of divided edges in a directed acyclic graph. The transmission rate can be obtained according to the Shannon formula
Figure BDA0002934969650000031
Figure BDA0002934969650000032
In the above formula, parameter BnRepresenting the bandwidth of the upload channel n, parameter PmTransmission power, parameter representing vehicle-mounted device m
Figure BDA0002934969650000033
Represents the path loss between the vehicle and the RSU, parameter delta represents the path loss factor, parameter h represents the channel fading factor of the uplink link, and parameter N0Representing white gaussian noiseThe acoustic power.
Thereby obtaining an additional transmission delay of vehicle m through channel n
Figure BDA0002934969650000034
Figure BDA0002934969650000035
In the above formula, the first and second carbon atoms are,
Figure BDA0002934969650000036
representing local offload subtask nodes
Figure BDA0002934969650000037
Offloading streaming nodes to edge server
Figure BDA0002934969650000038
The data size of (d);
the method for establishing the model for calculating the time delay in the step 1.3 comprises the following steps: using TmRepresents the time delay, T, of the vehicle m when the calculation task is unloadedmConsists of three parts: 1) delay of task node in task unloading of local server
Figure BDA0002934969650000039
2) Delay of task node in task unloading of edge server
Figure BDA00029349696500000310
3) Two task nodes with dependency relationship
Figure BDA00029349696500000311
And
Figure BDA00029349696500000312
additional transmission delay therebetween
Figure BDA00029349696500000313
Delay of vehicle m subtask node i in local task unloading
Figure BDA00029349696500000314
Can be expressed as:
Figure BDA00029349696500000315
in the above formula
Figure BDA00029349696500000316
Representing the local computing power (number of turns per second CPU) of the vehicle m.
Delay of vehicle m subtask node i in task unloading of edge server
Figure BDA00029349696500000317
Can be expressed as:
Figure BDA00029349696500000318
in the above formula fkRepresenting the computing power of the edge server k.
When two subtask neighbor nodes of a vehicle are unloaded at two different places, i.e.
Figure BDA00029349696500000319
Additional transmission delays may result.
Since the invention allows for parallel processing of computational tasks, the total delay TmNot a simple addition of the three parts. Offloading a portion of the vehicle's computational tasks to an edge processor for processing can achieve parallel processing tasks, thereby reducing latency and energy consumption.
Use of
Figure BDA0002934969650000041
Which represents the start time of the node i,
Figure BDA0002934969650000042
indicating the end time of the node i,
Figure BDA0002934969650000043
representing the execution time of the subtask node i. Thus, the formula is obtained:
Figure BDA0002934969650000044
since the subtask node i of vehicle m can be executed either locally or at the edge server, we get:
Figure BDA0002934969650000045
start time of node i of vehicle m
Figure BDA0002934969650000046
Depending mainly on the completion time of its predecessor.
Figure BDA0002934969650000047
The calculation formula of (a) is as follows:
Figure BDA0002934969650000048
when the subtask node i of the vehicle m is the first unload task, then the start time
Figure BDA0002934969650000049
In the above formula
Figure BDA00029349696500000410
Is a node
Figure BDA00029349696500000412
The direct front set of nodes.
Figure BDA00029349696500000413
Can be represented by the following formula:
Figure BDA00029349696500000414
from the above, the total calculated off-load task time for vehicle m is the end time of the last task, where v is the last task node. Namely:
Figure BDA00029349696500000415
the method for establishing the energy consumption calculation model in the step 1.4 comprises the following steps: emEnergy consumption of the incoming vehicle m when the calculation task is unloaded, EmIt consists of two parts: 1) compute task offload energy consumption for all subtask nodes of vehicle m
Figure BDA00029349696500000416
2) Energy consumption resulting from data transmission between cut edges in directed acyclic graphs
Figure BDA00029349696500000417
The energy consumption of the data transmission back to the vehicle is not considered.
Energy consumption for unloading computing tasks of all subtask nodes of vehicle m
Figure BDA00029349696500000418
Can be expressed as:
Figure BDA00029349696500000419
in the above equation, μ represents the power consumption coefficient of each CPU in the local server. i represents any one subtask node among all subtasks of the vehicle.
Only considering the energy consumption of the task unloaded from the vehicle to the edge server on the problem of the data transmission energy consumption of the associated node, the data transmission energy consumption of the associated node
Figure BDA0002934969650000051
Can be expressed as:
Figure BDA0002934969650000052
in summary, the total energy consumption of the vehicle m can be obtained as follows:
Figure BDA0002934969650000053
further, in step 2.1, the scheduling constraint of the task node is calculated, and the implementation method is as follows, and according to the objective functions given in (9) and (12), the subtask node needs to satisfy the following constraints:
1) executing the priority constraint:
subtask node of vehicle m
Figure BDA0002934969650000054
Is composed of
Figure BDA0002934969650000055
Direct father node of the task
Figure BDA0002934969650000056
Is higher than the execution priority of
Figure BDA0002934969650000057
The priority of the subtask node is calculated from the last subtask node of the vehicle m
Figure BDA0002934969650000058
Traversing the directed acyclic graph to recursively calculate the priority of each subtask node;
2) unload deadline constraint:
the unloading deadline is the time that the completion time of the last subtask node of the vehicle m cannot exceed the unloading time of the whole calculation task;
3) and (4) completing unloading constraint:
each subtask node of the vehicle m must unload its own task after all its predecessor subtask nodes are completed, and the start unloading time of the subtask node cannot be earlier than the end time of its predecessor node. When the subtask node and the predecessor node are not in the same position for unloading, the transmission time of the two nodes is calculated.
The method for establishing the objective function optimization problem model in the step 2.2 comprises the following steps: the unloading delay of the vehicle m mainly relates to the unloading time delay of the vehicle calculation task and the vehicle transmission communication delay, and the average delay of the vehicle can be expressed as:
Figure BDA0002934969650000059
in the above formula etam,n,kE {0, 1}, η if vehicle m offloads the computation task to edge server k over channel n m,n,k1, if not, ηm,n,k＝0。
The energy consumption of the vehicle m mainly relates to the energy consumption when unloading locally and the energy consumption through the channel, and the average consumption of the vehicle can be expressed as:
Figure BDA0002934969650000061
the problem of optimization delay and energy consumption is regarded as a constraint multi-objective optimization problem (CMOP), and an optimization function mainly comprises a time delay function and an energy consumption function. The optimization objective is to minimize the average delay and the average energy consumption, and the optimization strategy can be expressed as following six constraints when the optimization objective is satisfied: constraint C1 is that the task node for vehicle m can only be offloaded at one place, either locally or by an edge server; constraint C2 indicates whether vehicle m is connected to edge server k via channel n; constraint C3 is that a vehicle can only connect to one channel at a time; constrainingC4 is current node
Figure BDA0002934969650000062
Is a node
Figure BDA0002934969650000063
The direct parent node of (1); constraint C5 is that the completion time of the last subtask node of vehicle m cannot exceed the time to unload the entire computing task; constraint C6 is that the start unload time of a subtask node cannot be earlier than the end time of its predecessor node.
Further, the method for initializing the population in step 3.1 comprises the following steps: the genes in the chromosome correspond to one vehicle, and the gene factor represents the total number of vehicles. Vehicle m is composed of v computational tasks, and the gene values in the chromosomes can be expressed as value e {0,1,2v-1}. The value of the gene can convert value into binary representation of the vehicle m's computational task unloading decision;
step 3.2 the set constraints are as follows:
CON represents the degree of constraint violation and,
Figure BDA0002934969650000064
is the execution priority constraint violation for task node i,
Figure BDA0002934969650000065
is the degree of the unload deadline constraint violation for task node i,
Figure BDA0002934969650000066
is the degree of constraint violation for completion offload for task node i.
Step 3.3 Using the fast non-dominated ranking algorithm, two parameters n for each individual P in the population P are calculatedp(number of individuals in the population that dominate individual p) and Sp(set of individuals within the population that are dominated by individual p).
1) Finding out all n in the populationpThe 0 individuals are stored in the set F1 (i.e., the first tier).
2) For each individual p in F1, its nameThe dominant set of individuals is SpGo through SpIn each individual l, nl＝nl-1, if nlL is saved in the set F2 (second layer) at 0.
3) Repeating the step 2) by taking F2 as a current set until the whole population is layered;
the method for calculating the congestion degree in step 3.4 is described as follows:
Figure BDA0002934969650000067
in the above formula nd,iIndicating the degree of congestion of the solution i,
Figure BDA0002934969650000068
which represents the average delay function of the delay,
Figure BDA0002934969650000069
representing the average energy consumption function.
Figure BDA00029349696500000610
And
Figure BDA00029349696500000611
representing the maximum and minimum values in the mean delay function,
Figure BDA00029349696500000612
and
Figure BDA00029349696500000613
representing the maximum and minimum values in the average energy consumption function. Meanwhile, the degree of congestion for the two sorted boundaries is ∞. When the domination levels of the two solutions are the same, selecting the solution with high crowdedness;
the crossover and mutation algorithm described in step 3.5 is described as follows:
the crossover probability CP is typically set to 1.5; the mutation probability MP is calculated as follows:
Figure BDA0002934969650000071
in the above equation, mp represents a variation probability value given at initialization, N represents the current iteration number, and N represents the maximum iteration number.
The crossover operation was performed as follows:
Figure BDA0002934969650000072
in the above formula, x1And x2Representing p in the parent chromosome1And p2Cross gene value of (1), x1'and x'2Represents c in the offspring chromosome1And c2The crossed gene value and the gene position are the same as the corresponding parent, and alpha belongs to [0,1 ]]Is a random variable.
The mutation operation is carried out according to the following formula:
x'm,i＝2v-1-xm,i (18)
in the above formula, xm,iDenotes the ith gene value, x 'on chromosome m'm,iRepresents the filial generation after mutation;
the population update method described in step 3.6 is described as follows:
obtaining a new offspring population after self-adaptive crossing and mutation operations, then calculating an objective function value and a constraint violation degree of the new offspring population, combining the parent population and the offspring population into a new population by adopting a championship algorithm, sequencing the new population by using a rapid non-dominated sequencing method, calculating the crowding degree of each individual, and sequencing by comparing the non-dominated level and the crowding degree to obtain the optimal chromosome to be reserved and generate the next generation. After iteration is terminated, the solution of the Pareto optimal leading edge is converted into a binary representation to be an optimal calculation unloading strategy. Otherwise, returning to the operation of initializing the population, and continuously iterating until the end condition of iteration is met.
The invention has the advantages and positive effects that:
the invention mainly designs a multi-objective calculation task unloading scheduling method based on a non-dominated sorting genetic strategy, and mainly researches the problems that the unloading of vehicle calculation tasks is limited by limited resources of an MEC server, the time delay and the energy consumption of the calculation tasks are influenced and the like. Due to the limited resources of the MEC server, the vehicle computing task is split into sub-tasks that can be executed in parallel, partly processed locally and partly offloaded to the MEC server. The method solves the problems of time delay and energy consumption of the calculation task unloading by using a non-dominated sorting genetic strategy. Compared with other unloading methods, the algorithm provided by the invention has better performance and certain practical value.
Drawings
FIG. 1 is a schematic diagram of parallel processing;
FIG. 2 is a graph of vehicle density versus average time delay;
FIG. 3 is a graph of vehicle density versus average energy consumption;
FIG. 4 is a graph of the number of subtasks versus the average delay;
FIG. 5 is a flow diagram of a method for multi-objective computational task offload scheduling that is a non-dominated ranking genetic strategy.
Detailed Description
Referring to the attached figure 5, the method for unloading and scheduling the multi-objective calculation task in the internet of vehicles based on the non-dominated sorting genetic strategy mainly comprises the following steps:
1, constructing a vehicle networking system model:
1.1, defining a general noun of the Internet of vehicles system;
1.2, establishing a communication model;
1.3, establishing a calculation time delay model;
1.4, establishing a calculation energy consumption model;
and 2, constructing a task unloading strategy of the Internet of vehicles:
2.1, calculating the scheduling constraint of the task node;
2.2, constructing a problem model according to the objective function;
design of NSGS algorithm:
3.1, designing a vehicle population according to a genetic algorithm and initializing;
3.2, setting constraint conditions;
3.3, sorting individuals by using a rapid non-dominated sorting method;
3.4, calculating the congestion degree;
3.5, designing a crossover and mutation strategy;
and 3.6, designing a population updating strategy.
In step 1.1, a general term of the car networking system is defined, that is, M ═ 1,2, 3.., M } vehicles, N ═ 1,2, 3.., N } channels, and K ═ 1,2, 3.., K } edge servers are located on one road. Vehicles can only be connected with one channel at most in one time period, and each vehicle is restricted to select one MEC server to execute tasks. The computational task of each vehicle is divided into several interdependent subtasks that can be offloaded to the edge server for processing, but some tasks must be performed locally, such as vehicle GPS to obtain the vehicle's position. The first and last subtasks of the vehicle are constrained to be performed locally. The invention proposes to use a directed acyclic graph Dm＝(Vm,Rm) Represents a computational task of the vehicle, wherein
Figure BDA0002934969650000091
Set of all subtasks representing vehicle m, use
Figure BDA0002934969650000092
To represent the subtasks of the vehicle m
Figure BDA0002934969650000093
And
Figure BDA0002934969650000094
the two subtasks are neighbor nodes, and the current subtask
Figure BDA0002934969650000095
Is that
Figure BDA0002934969650000096
In the case of the immediate parent node of (c),
Figure BDA0002934969650000097
must be in
Figure BDA0002934969650000098
And is done before. Additional communication costs are incurred when the two subtasks are offloaded locally and the other is offloaded at the edge server. For subtask node of each vehicle m
Figure BDA0002934969650000099
And (4) showing.
Figure BDA00029349696500000910
The data size of the subtask node i representing the vehicle m,
Figure BDA00029349696500000911
represents the CPU resources consumed by the subtask node i of the vehicle m when performing task offloading locally,
Figure BDA00029349696500000912
and the CPU resource consumed by the subtask node i of the vehicle m when the edge server carries out task unloading is shown.
Figure BDA00029349696500000913
Representing a decision factor, when the decision factor is equal to 1, representing that the subtask i is unloaded at the edge server, and when the decision factor is equal to 0, representing that the subtask i is unloaded at the local server;
the method for establishing the communication model in the step 1.2 comprises the following steps of dividing the vehicle calculation task into subtasks, improving the task unloading parallel capability, and nodularizing the task of the vehicle into nodes
Figure BDA00029349696500000914
And
Figure BDA00029349696500000915
two disjoint sets represent the set of tasks offloaded locally and offloaded at the edge server, respectively. The invention uses
Figure BDA00029349696500000916
Representing a set of divided edges in a directed acyclic graph. The transmission rate can be obtained according to the Shannon formula
Figure BDA00029349696500000917
Figure BDA00029349696500000918
In the above formula, parameter BnRepresenting the bandwidth of the upload channel n, parameter PmTransmission power, parameter representing vehicle-mounted device m
Figure BDA00029349696500000919
Represents the path loss between the vehicle and the RSU, parameter delta represents the path loss factor, parameter h represents the channel fading factor of the uplink link, and parameter N0Representing gaussian white noise power.
So that an additional transmission delay of the vehicle m through the channel n can be obtained
Figure BDA00029349696500000920
Figure BDA00029349696500000921
In the above formula, the first and second carbon atoms are,
Figure BDA00029349696500000922
representing local offload subtask nodes
Figure BDA00029349696500000923
Offloading multitask nodes to edge servers
Figure BDA00029349696500000924
The data size of (d);
the method for establishing the model for calculating the time delay in the step 1.3 comprises the following steps: in the present invention, T is usedmRepresents the time delay, T, of the vehicle m when the calculation task is unloadedmConsists of three parts: 1) delay of task node in task unloading of local server
Figure BDA00029349696500000925
2) Delay of task node in task unloading of edge server
Figure BDA00029349696500000926
3) Two task nodes with dependency relationship
Figure BDA00029349696500000927
And
Figure BDA0002934969650000101
additional transmission delay therebetween
Figure BDA0002934969650000102
Delay of vehicle m subtask node i in local task unloading
Figure BDA0002934969650000103
Can be expressed as:
Figure BDA0002934969650000104
in the above formula
Figure BDA0002934969650000105
Representing the local computing power (number of turns per second CPU) of the vehicle m.
Delay of vehicle m subtask node i in task unloading of edge server
Figure BDA0002934969650000106
Can be expressed as:
Figure BDA0002934969650000107
in the above formula fkRepresenting the computing power of the edge server k.
When two subtask neighbor nodes of a vehicle are unloaded at two different places, i.e.
Figure BDA0002934969650000108
Additional transmission delays may result.
Since the invention allows for parallel processing of computational tasks, the total delay TmNot a simple addition of the three parts. Offloading a portion of the vehicle's computational tasks to an edge processor for processing can achieve parallel processing tasks, thereby reducing latency and energy consumption.
Used in the invention
Figure BDA0002934969650000109
Which represents the start time of the node i,
Figure BDA00029349696500001010
indicating the end time of the node i,
Figure BDA00029349696500001011
representing the execution time of the subtask node i. From this, the formula can be derived:
Figure BDA00029349696500001012
since the subtask node i of vehicle m can be executed either locally or at the edge server, we get:
Figure BDA00029349696500001013
section of vehicle mStarting time of point i
Figure BDA00029349696500001014
Depending mainly on the completion time of its predecessor.
Figure BDA00029349696500001015
The calculation formula of (a) is as follows:
Figure BDA00029349696500001016
when the subtask node i of the vehicle m is the first unload task, then the start time
Figure BDA00029349696500001017
In the above formula
Figure BDA00029349696500001018
Is a node
Figure BDA00029349696500001019
The direct front set of nodes.
Figure BDA00029349696500001020
Can be represented by the following formula:
Figure BDA0002934969650000111
from the above, the total calculated off-load task time for vehicle m is the end time of the last task, where v is the last task node. Namely:
Figure BDA0002934969650000112
the method for establishing the energy consumption calculation model in the step 1.4 comprises the following steps: emEnergy consumption of the incoming vehicle m when the calculation task is unloaded, EmIt consists of two parts: 1) of all subtask nodes of vehicle mCompute task offload energy consumption
Figure BDA0002934969650000113
2) Energy consumption resulting from data transmission between cut edges in directed acyclic graphs
Figure BDA0002934969650000114
The energy consumption of the data transmission back to the vehicle is not taken into account in the present invention.
Energy consumption for unloading computing tasks of all subtask nodes of vehicle m
Figure BDA0002934969650000115
Can be expressed as:
Figure BDA0002934969650000116
in the above equation, μ represents the power consumption coefficient of each CPU in the local server. i represents any one of all subtask nodes of the vehicle.
In the invention, only the energy consumption of unloading the task from the vehicle to the edge server is considered on the problem of the data transmission energy consumption of the associated node, so that the data transmission energy consumption of the associated node
Figure BDA0002934969650000117
Can be expressed as:
Figure BDA0002934969650000118
in summary, the total energy consumption of the vehicle m can be obtained as follows:
Figure BDA0002934969650000119
in step 2.1 of the present invention, the scheduling constraint of the task node is calculated, and the implementation method is as follows, according to the objective functions given in (9) and (12), the subtask node needs to satisfy the following constraints:
1) executing the priority constraint:
subtask node of vehicle m
Figure BDA00029349696500001110
Is composed of
Figure BDA00029349696500001111
Direct father node of the task
Figure BDA00029349696500001112
Is higher than the execution priority of
Figure BDA00029349696500001113
The priority of the subtask node is calculated from the last subtask node of the vehicle m
Figure BDA00029349696500001114
Traversing the directed acyclic graph to recursively calculate the priority of each subtask node;
2) unload deadline constraint:
the unloading deadline is the time that the completion time of the last subtask node of the vehicle m cannot exceed the unloading time of the whole calculation task;
3) and (4) completing unloading constraint:
each subtask node of the vehicle m must unload its own task after all its predecessor subtask nodes are completed, and the start unloading time of the subtask node cannot be earlier than the end time of its predecessor node. When the subtask node and the predecessor node are not in the same position for unloading, the transmission time of the two nodes is calculated.
The method for establishing the objective function optimization problem model in the step 2.2 comprises the following steps: the unloading delay of the vehicle m in the invention mainly relates to the unloading time delay of the vehicle calculation task and the vehicle transmission communication delay, and the average delay of the vehicle can be expressed as:
Figure BDA0002934969650000121
in the above formula etam,n,kE {0, 1}, η if vehicle m offloads the computation task to edge server k over channel n m,n,k1, if not, ηm,n,k＝0，
Figure BDA0002934969650000122
Is the set of task nodes preceding the last subtask node of vehicle m, p is
Figure BDA0002934969650000123
Any one of the task nodes;
the energy consumption of the vehicle m in the present invention mainly relates to the energy consumption when unloading is performed locally and the energy consumption through the channel, and the average consumption of the vehicle can be expressed as:
Figure BDA0002934969650000124
the problem of optimization delay and energy consumption is regarded as a constraint multi-objective optimization problem (CMOP), and an optimization function mainly comprises a time delay function and an energy consumption function. The optimization objective is to minimize the average delay and the average energy consumption, and the optimization strategy can be expressed as following six constraints when the optimization objective is satisfied: constraint C1 is that the task node for vehicle m can only be offloaded at one place, either locally or by an edge server; constraint C2 indicates whether vehicle m is connected to edge server k via channel n; constraint C3 is that a vehicle can only connect to one channel at a time; constraint C4 is current node
Figure BDA0002934969650000125
Is a node
Figure BDA0002934969650000126
The direct parent node of (1); constraint C5 is that the completion time of the last subtask node of vehicle m cannot exceed the time to unload the entire computing task; constrainingC6 is the subtask node's start unload time cannot be earlier than the end time of its predecessor node.
The method for initializing the population in the step 3.1 of the invention comprises the following steps: the genes in the chromosome correspond to one vehicle, and the gene factor represents the total number of vehicles. Vehicle m is composed of v computational tasks, and the gene values in the chromosomes can be expressed as value e {0,1,2v-1}. The value of the gene can convert value into binary representation of the vehicle m's computational task unloading decision;
step 3.2 the set constraints are as follows:
CON represents the degree of constraint violation and,
Figure BDA0002934969650000131
is the execution priority constraint violation for task node i,
Figure BDA0002934969650000132
is the degree of the unload deadline constraint violation for task node i,
Figure BDA0002934969650000133
is the degree of constraint violation for completion offload for task node i.
Step 3.3 Using the fast non-dominated ranking algorithm, two parameters n for each individual P in the population P are calculatedp(number of individuals in the population that dominate individual p) and Sp(set of individuals within the population that are dominated by individual p).
1) Finding out all n in the populationpThe 0 individuals are stored in the set F1 (i.e., the first tier).
2) For each individual p in F1, the set of individuals dominated by it is SpGo through SpIn each individual l, nl＝nl-1, if nlL is saved in the set F2 (second layer) at 0.
3) Repeating the step 2) by taking F2 as a current set until the whole population is layered;
the method for calculating the congestion degree in step 3.4 is described as follows:
Figure BDA0002934969650000134
in the above formula nd,iIndicating the degree of congestion of the solution i,
Figure BDA0002934969650000135
which represents the average delay function of the delay,
Figure BDA0002934969650000136
representing the average energy consumption function.
Figure BDA0002934969650000137
And
Figure BDA0002934969650000138
representing the maximum and minimum values in the mean delay function,
Figure BDA0002934969650000139
and
Figure BDA00029349696500001310
representing the maximum and minimum values in the average energy consumption function. And setting the congestion degree of the two sorted boundaries to be infinity. When the domination levels of the two solutions are the same, selecting the solution with high crowdedness;
the crossover and mutation algorithm described in step 3.5 is described as follows:
the crossover probability CP is typically set to 1.5; the mutation probability MP is calculated as follows:
Figure BDA00029349696500001311
in the above formula, mp represents a variation probability value given at initialization, N represents the current iteration number, and N represents the maximum iteration number.
The crossover operation was performed as follows:
Figure BDA00029349696500001312
in the above formula, x1And x2Representing p in the parent chromosome1And p2Cross gene value of (1), x1'and x'2Represents c in the offspring chromosome1And c2The crossed gene value and the gene position are the same as the corresponding parent, and alpha belongs to [0,1 ]]Is a random variable.
The mutation operation is carried out according to the following formula:
x'm,i＝2v-1-xm,i (18)
in the above formula, xm,iDenotes the ith gene value, x 'on chromosome m'm,iRepresents the filial generation after mutation; the description of the population update method in step 3.6 is as follows: obtaining a new offspring population after self-adaptive crossing and mutation operations, then calculating an objective function value and a constraint violation degree of the new offspring population, combining the parent population and the offspring population into a new population by adopting a championship algorithm, sequencing the new population by using a rapid non-dominated sequencing method, calculating the crowding degree of each individual, and sequencing by comparing the non-dominated level and the crowding degree to obtain the optimal chromosome to be reserved and generate the next generation. After iteration is terminated, the solution of the Pareto optimal leading edge is converted into a binary representation to be an optimal calculation unloading strategy. Otherwise, returning to the operation of initializing the population, and continuously iterating until the termination condition of iteration is met.
Example 1:
the method designed by the embodiment is to verify the NSGS offload algorithm based on Matlab 2018 a. The primary goal of validation is to determine the impact of non-dominated ranking genetic strategies on the offloading of vehicle computing tasks. In addition to this, it is desirable to examine the benefits of using the proposed computational task offload scheduling method in urban scenarios of large traffic volumes with different numbers of vehicles.
Different parameters were used in experiments to verify the performance of the non-dominated ranking genetic strategy algorithm. The implementation operation mainly involved comprises the implementation of a non-dominated sorting genetic algorithm, the construction of a simulation scene and a specific algorithm calculation process.
In the example, a simulation scene is constructed, on urban roads with large traffic flow, four bidirectional lanes are provided with an MEC server every 300 meters, and the speed is 40-60 kilometers per hour.
The speed of each lane follows normal distribution, all vehicles run at a constant speed along the road, the distribution of the vehicles on the road is distributed according to the Poisson theorem, and each vehicle is independent. The settings of the respective parameters are as shown in tables 1 and 2:
TABLE 1 Experimental parameters
Figure BDA0002934969650000141
Figure BDA0002934969650000151
TABLE 2 NSGS parameters
Parameter Value
Number of population [40,80]
Maximum number of iterations 100
ε 1.5
This simulation experiment will consider two performance indicators, which are:
1. average time delay
Figure BDA0002934969650000152
The average time delay is the average of the time delay for the vehicle to compute the task offload and the vehicle transmission communication delay.
2. Average energy consumption
Figure BDA0002934969650000153
The average energy consumption is the average of the energy consumption when locally offloaded and the energy consumption through the channel.
The results of the simulation experiments for this example are as follows:
1. effect of different vehicle densities on two Performance indicators
1) Relationship between vehicle density and average time delay
Fig. 2 is a graph of vehicle density versus average time delay. It can be seen from the figure that as the vehicle density increases, the average delay of the five algorithms increases, while the average delay of the NSGS algorithm proposed by the present invention is the lowest among the five algorithms. The method mainly considers the environmental factors in the sub-task dividing process after the calculation and unloading task segmentation, such as the calculation capacity of the edge server, the transmission rate of a channel and the like, so that the method is more accurate and efficient in task dividing. Because the dynamic perception mechanism combining transmission delay and energy consumption is to divide the unloading task nodes and then perform channel selection and server matching, the node division is ambiguous, and more time and energy are needed, so that the optimal energy-saving effect cannot be achieved. The traditional NSGA-II algorithm does not introduce methods such as a fast non-dominant method, an elite strategy and the like, so that the traditional NSGA-II algorithm has a gap from the algorithm provided by the invention in the aspect of constraint processing. Compared with the traditional NSGA-II algorithm, the MOEA/D algorithm can be optimized according to the information of the adjacent subproblems, so that the MOEA/D algorithm is low in complexity and time delay. The offloading position of the randomly assigned (random) node does not use an optimization algorithm, so that the performance is not good in terms of time delay and energy consumption.
2) Relationship between vehicle density and average energy consumption
Figure 3 shows the relationship between vehicle density and average energy consumption. It can be seen from the figure that as the vehicle density increases, the average energy consumption of the 5 algorithms increases, but the NSGS works best. The analysis of the relationship between vehicle density and average energy consumption is similar to the analysis of the relationship between vehicle density and average time delay.
2. Effect of different numbers of subtasks on average delay
Fig. 4 is a relationship between the number of subtasks and the average delay. The relationship between the number of subtasks and the average time delay is similar to the analysis of the relationship between the vehicle density and the average time delay.
Simulation results show that the NSGS can achieve the best effect in all tested algorithms.

Claims (1)

1. A multi-objective calculation task unloading scheduling method for a vehicle networking based on a non-dominated sorting genetic strategy is characterized by comprising the following steps:
1, constructing a vehicle networking system model:
1.1, defining a general noun of the Internet of vehicles system;
1.2, establishing a communication model;
1.3, establishing a calculation time delay model;
1.4, establishing a calculation energy consumption model;
and 2, constructing a task unloading strategy of the Internet of vehicles:
2.1, calculating the scheduling constraint of the task node;
2.2, constructing a problem model according to the objective function;
3, designing an NSGS algorithm:
3.1, designing a vehicle population according to a genetic algorithm and initializing;
3.2, setting constraint conditions;
3.3, sorting individuals by using a rapid non-dominated sorting method;
3.4, calculating the congestion degree;
3.5, designing a crossover and mutation strategy;
3.6, designing a population updating strategy;
in step 1.1, a general term of the car networking system is defined, that is, M ═ 1,2, 3.., M } vehicles, N ═ 1,2, 3.., N } channels, and K ═ 1,2, 3.., K } edge servers are located on one road; the vehicles can only be connected with one channel at most in one time period, and simultaneously each vehicle is restricted to only select one MEC server to execute tasks; dividing the computing task of each vehicle into a plurality of sub-tasks which are mutually dependent, wherein the sub-tasks are unloaded to an edge server for processing, but some tasks must be executed locally, and the first sub-task and the last sub-task of the vehicle are restricted to be executed locally; using directed acyclic graphs Dm＝(Vm,Rm) Represents a computational task of the vehicle, wherein
Figure FDA0003590930640000011
Set of all subtasks representing vehicle m, use
Figure FDA0003590930640000012
To represent the subtasks of the vehicle m
Figure FDA0003590930640000013
And
Figure FDA0003590930640000014
the two subtasks are neighbor nodes, and the current subtask
Figure FDA0003590930640000015
Is that
Figure FDA0003590930640000016
In the case of the immediate parent node of (c),
Figure FDA0003590930640000017
must be in
Figure FDA0003590930640000018
The previous completion; when the two subtasks are detached locallyThe other will generate extra communication cost when the edge server is unloaded, and the subtask node of each vehicle m is used
Figure FDA0003590930640000019
It is shown that,
Figure FDA00035909306400000110
the data size of the subtask node i representing the vehicle m,
Figure FDA00035909306400000111
represents the CPU resources consumed by the subtask node i of the vehicle m when performing task offloading locally,
Figure FDA00035909306400000112
represents the CPU resources consumed by the subtask node i of the vehicle m when the edge server performs task offloading,
Figure FDA0003590930640000021
representing a decision factor, when the decision factor is equal to 1, representing that the subtask i is unloaded at the edge server, and when the decision factor is equal to 0, representing that the subtask i is unloaded at the local server;
the method for establishing the communication model in the step 1.2 comprises the following steps of dividing the vehicle calculation task into subtasks to improve the parallel capability of task unloading, and nodularizing the task of the vehicle into nodes
Figure FDA0003590930640000022
And
Figure FDA0003590930640000023
two disjoint sets, representing the set of tasks offloaded locally and offloaded at the edge server,
Figure FDA0003590930640000024
representing a set of divided edges in a directed acyclic graph; obtained according to the Shannon formulaTransmission rate
Figure FDA0003590930640000025
Figure FDA0003590930640000026
In the above formula, parameter BnRepresenting the bandwidth of the upload channel n, parameter PmTransmission power, parameter representing vehicle-mounted device m
Figure FDA0003590930640000027
Represents the path loss between the vehicle and the RSU, parameter delta represents the path loss factor, parameter h represents the channel fading factor of the uplink link, and parameter N0Representing a gaussian white noise power;
thereby obtaining an additional transmission delay of vehicle m through channel n
Figure FDA0003590930640000028
Figure FDA0003590930640000029
In the above formula, the first and second carbon atoms are,
Figure FDA00035909306400000210
representing local offload subtask nodes
Figure FDA00035909306400000211
Offloading streaming nodes to edge server
Figure FDA00035909306400000212
The data size of (d);
the method for establishing the model for calculating the time delay in the step 1.3 comprises the following steps: using TmRepresents the time delay, T, of the vehicle m when the calculation task is unloadedmIs composed of three partsConsists of the following components: 1) delay of task node in task unloading of local server
Figure FDA00035909306400000213
2) Delay of task node in task unloading of edge server
Figure FDA00035909306400000214
3) Two task nodes with dependency relationship
Figure FDA00035909306400000215
And
Figure FDA00035909306400000216
additional transmission delay therebetween
Figure FDA00035909306400000217
Delay of vehicle m subtask node i in local task unloading
Figure FDA00035909306400000218
Expressed as:
Figure FDA00035909306400000219
in the above formula
Figure FDA00035909306400000220
Representing the local computing power of vehicle m, number of turns of CPU per second;
delay of vehicle m subtask node i in task unloading of edge server
Figure FDA00035909306400000221
Expressed as:
Figure FDA0003590930640000031
in the above formula fkRepresenting the computing power of the edge server k;
when two subtask neighbor nodes of a vehicle are unloaded at two different places, i.e.
Figure FDA0003590930640000032
Additional transmission delays may occur;
use of
Figure FDA0003590930640000033
Which represents the start time of the node i,
Figure FDA0003590930640000034
indicating the end time of the node i,
Figure FDA0003590930640000035
representing the execution time of the subtask node i, thereby obtaining the formula:
Figure FDA0003590930640000036
since the subtask node i of vehicle m is executed both locally and at the edge server, we get:
Figure FDA0003590930640000037
start time of node i of vehicle m
Figure FDA0003590930640000038
Depending on the completion time of its predecessor node,
Figure FDA0003590930640000039
the calculation formula of (a) is as follows:
Figure FDA00035909306400000310
when the subtask node i of the vehicle m is the first unload task, then the start time
Figure FDA00035909306400000311
In the above formula
Figure FDA00035909306400000312
Is a node
Figure FDA00035909306400000313
The set of direct front-end nodes of (c),
Figure FDA00035909306400000314
represented by the following formula:
Figure FDA00035909306400000315
from the above, the total calculated off-load task time for vehicle m is the end time of the last task, where v is the last task node, i.e.:
Figure FDA00035909306400000316
the method for establishing the energy consumption calculation model in the step 1.4 comprises the following steps: emEnergy consumption of the incoming vehicle m when the calculation task is unloaded, EmIt consists of two parts: 1) compute task offload energy consumption for all subtask nodes of vehicle m
Figure FDA00035909306400000317
2) Energy consumption resulting from data transmission between cut edges in directed acyclic graphs
Figure FDA00035909306400000318
Energy consumption for data transfer back to the vehicle is not considered;
energy consumption for unloading computing tasks of all subtask nodes of vehicle m
Figure FDA00035909306400000319
Expressed as:
Figure FDA00035909306400000320
in the formula, mu represents the energy consumption coefficient of each CPU in the local server, and i represents any subtask node in all subtasks of the vehicle;
only considering the energy consumption of the task unloaded from the vehicle to the edge server on the problem of the data transmission energy consumption of the associated node, the data transmission energy consumption of the associated node
Figure FDA0003590930640000041
Expressed as:
Figure FDA0003590930640000042
in summary, the total energy consumption of the vehicle m is obtained as:
Figure FDA0003590930640000043
in step 2.1, the scheduling constraint of the task node is calculated, and the implementation method is as follows, and according to the objective functions given by the formulas (9) and (12), the subtask node needs to satisfy the following constraint:
1) executing the priority constraint:
subtask node of vehicle m
Figure FDA0003590930640000044
Is composed of
Figure FDA0003590930640000045
Direct father node of the task
Figure FDA0003590930640000046
Is higher than the execution priority of
Figure FDA0003590930640000047
The priority of the subtask node is calculated from the last subtask node of the vehicle m
Figure FDA0003590930640000048
Traversing the directed acyclic graph to recursively calculate the priority of each subtask node;
2) unload deadline constraint:
the unloading deadline is the time that the completion time of the last subtask node of the vehicle m cannot exceed the unloading time of the whole calculation task;
3) and (4) completing unloading constraint:
each subtask node of the vehicle m can unload the task after all the predecessor subtask nodes of the vehicle m are completely finished, the unloading starting time of the subtask node cannot be earlier than the ending time of the predecessor node, and when the subtask node and the predecessor node are not in the same position for unloading, the transmission time of the two nodes also needs to be calculated;
the method for establishing the objective function optimization problem model in the step 2.2 comprises the following steps: the unloading delay of the vehicle m relates to the unloading time delay of the vehicle calculation task and the vehicle transmission communication delay, and the average delay of the vehicle is represented as:
Figure FDA0003590930640000049
in the above formula etam,n,kE {0, 1}, if vehicle m offloads the computing task over channel n toOn the edge server k, then ηm,n,k1, if not, ηm,n,k＝0；
The energy consumption of the vehicle m relates to the energy consumption when unloading locally and the energy consumption through the channel, and the average consumption of the vehicle is expressed as:
Figure FDA0003590930640000051
the problem of optimization delay and energy consumption is regarded as a constraint multi-objective optimization problem CMOP, the optimization function is composed of a time delay function and an energy consumption function, the optimization objective is to minimize average time delay and average energy consumption, and the optimization strategy is expressed to meet the following six constraints under the condition of meeting the optimization objective: constraint C1 is that the task node for vehicle m can only be offloaded at one place, either locally or by an edge server; constraint C2 indicates whether vehicle m is connected to edge server k via channel n; constraint C3 is that a vehicle only connects to one channel at a time; constraint C4 is current node
Figure FDA0003590930640000052
Is a node
Figure FDA0003590930640000053
The direct parent node of (1); constraint C5 is that the completion time of the last subtask node of vehicle m cannot exceed the time to unload the entire computing task; constraint C6 is that the start unload time of a subtask node cannot be earlier than the end time of its predecessor node;
the method for initializing the population in the step 3.1 comprises the following steps: the genes in the chromosome correspond to one vehicle, the base factor represents the total number of the vehicles, the vehicle m consists of v calculation tasks, and the values of the genes in the chromosome are represented as value e {0,1,2v-1} the value of the gene converts value into binary representation of the vehicle m's computational task offloading decision;
step 3.2 the set constraints are as follows:
CON represents the degree of constraint violation and,
Figure FDA0003590930640000054
is the execution priority constraint violation for task node i,
Figure FDA0003590930640000055
is the degree of the unload deadline constraint violation for task node i,
Figure FDA0003590930640000056
the degree of violation of unloading constraint completion of the task node i;
step 3.3 Using the fast non-dominated ranking algorithm, two parameters n for each individual P in the population P are calculatedpNumber of individuals in the population that dominate individual p and SpA set of individuals within the population that are dominated by individual p;
1) finding out all n in the populationpIndividuals of 0, saved in set F1, i.e. the first tier;
2) for each individual p in F1, the set of individuals dominated by it is SpGo through SpIn each individual l, nl＝nl-1, if nlWhen 0, store l in the set F2, i.e. the second layer;
3) repeating the step 2) by taking F2 as a current set until the whole population is layered;
the method for calculating the congestion degree in step 3.4 is described as follows:
Figure FDA0003590930640000057
in the above formula nd,iIndicating the degree of congestion of the solution i,
Figure FDA0003590930640000058
which represents the average delay function of the delay,
Figure FDA0003590930640000059
the average energy consumption function is represented as,
Figure FDA00035909306400000510
and
Figure FDA00035909306400000511
representing the maximum and minimum values in the mean delay function,
Figure FDA0003590930640000061
and
Figure FDA0003590930640000062
representing the maximum value and the minimum value in the average energy consumption function, setting the congestion degrees of the two sorted boundaries to be infinite, and selecting a solution with a high congestion degree when the domination levels of the two solutions are the same;
the crossover and mutation algorithm described in step 3.5 is described as follows:
the crossover probability CP is set to 1.5; the mutation probability MP is calculated as follows:
Figure FDA0003590930640000063
in the above formula, mp represents a given variation probability value during initialization, N represents the current iteration number, and N represents the maximum iteration number;
the crossover operation was performed as follows:
Figure FDA0003590930640000064
in the above formula, x1And x2Representing p in the parent chromosome1And p2X 'of the cross gene value of'1And x'2Represents c in the offspring chromosome1And c2The crossed gene value and the gene position are the same as the corresponding parent, and alpha belongs to [0,1 ]]Is a random variable;
the mutation operation is carried out according to the following formula:
x′m,i＝2v-1-xm,i (18)
in the above formula, xm,iDenotes the ith gene value, x 'on chromosome m'm,iRepresents the filial generation after mutation;
the population update method described in step 3.6 is described as follows:
obtaining a new offspring population after self-adaptive crossing and mutation operations, then calculating an objective function value and a constraint violation degree of the new offspring population, synthesizing a new population from a parent population and the offspring population by adopting a championship algorithm, sequencing the new population by using a rapid non-dominated sequencing method, calculating the crowding degree of each individual, sequencing by comparing the non-dominated level and the crowding degree to obtain the optimal chromosome to be reserved to generate the next generation, converting the solution of the Pareto optimal front edge into a binary system to be represented as an optimal calculation unloading strategy after iteration is terminated, otherwise, returning to the operation of initializing the population, and continuously iterating until the termination condition of the iteration is met.
CN202110156669.XA 2021-02-04 2021-02-04 Internet of vehicles multi-target computing task unloading scheduling method based on non-dominated sorting genetic strategy Expired - Fee Related CN112995289B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110156669.XA CN112995289B (en) 2021-02-04 2021-02-04 Internet of vehicles multi-target computing task unloading scheduling method based on non-dominated sorting genetic strategy

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110156669.XA CN112995289B (en) 2021-02-04 2021-02-04 Internet of vehicles multi-target computing task unloading scheduling method based on non-dominated sorting genetic strategy

Publications (2)

Publication Number Publication Date
CN112995289A CN112995289A (en) 2021-06-18
CN112995289B true CN112995289B (en) 2022-05-17

Family

ID=76347103

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110156669.XA Expired - Fee Related CN112995289B (en) 2021-02-04 2021-02-04 Internet of vehicles multi-target computing task unloading scheduling method based on non-dominated sorting genetic strategy

Country Status (1)

Country Link
CN (1) CN112995289B (en)

Families Citing this family (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113542376A (en) * 2021-07-01 2021-10-22 广东工业大学 Task unloading method based on energy consumption and time delay weighting
CN113742077B (en) * 2021-09-08 2024-03-01 吉林大学 Computing migration method based on 5G Internet of vehicles
CN114296814B (en) * 2021-12-10 2024-06-21 中国科学院深圳先进技术研究院 Side cloud computing task unloading method, system, terminal and storage medium
CN114327526B (en) * 2022-01-05 2024-05-28 安徽大学 Task unloading method in mobile edge computing environment and application thereof
CN114637552B (en) * 2022-03-09 2024-07-30 华电重工机械有限公司 Fog computing task unloading method based on fuzzy logic strategy
CN114980029B (en) * 2022-05-20 2024-10-01 上饶丫牛软件中心(个人独资) Task relevance-based unloading method in Internet of vehicles
CN114980216B (en) * 2022-06-01 2024-03-22 江南大学 Dependency task unloading system and method based on mobile edge calculation
CN116708581B (en) * 2023-06-02 2024-03-29 暨南大学 High-reliability function scheduling method for server-free edge computing
CN117971348B (en) * 2024-04-01 2024-07-16 西安邮电大学 Internet of things edge computing unloading method and system based on film computing

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111818168A (en) * 2020-06-19 2020-10-23 重庆邮电大学 Self-adaptive joint calculation unloading and resource allocation method in Internet of vehicles
CN111836283A (en) * 2020-07-03 2020-10-27 重庆邮电大学 Internet of vehicles resource allocation method based on MEC multi-server
US10866836B1 (en) * 2019-08-20 2020-12-15 Deke Guo Method, apparatus, device and storage medium for request scheduling of hybrid edge computing

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US10866836B1 (en) * 2019-08-20 2020-12-15 Deke Guo Method, apparatus, device and storage medium for request scheduling of hybrid edge computing
CN111818168A (en) * 2020-06-19 2020-10-23 重庆邮电大学 Self-adaptive joint calculation unloading and resource allocation method in Internet of vehicles
CN111836283A (en) * 2020-07-03 2020-10-27 重庆邮电大学 Internet of vehicles resource allocation method based on MEC multi-server

Non-Patent Citations (5)

* Cited by examiner, † Cited by third party
Title
A Time Domain Passivity Approach for Asymmetric Multilateral Teleoperation System;Usman Ahmad;《IEEE》;20171212;全文 *
一种面向边缘计算节点能量优化的QoS约束路由算法;张德干等;《北京邮电大学学报》;20200831;全文 *
车联网中一种基于软件定义网络与移动边缘计算的卸载策略;张海波等;《电子与信息学报》;20200315(第03期);全文 *
面向物联网的移动边缘计算任务卸载方法研究;王登辉等;《内江师范学院学报》;20201025(第10期);全文 *
面向车联网边缘计算的智能计算迁移研究;莫瑞超等;《应用科学学报》;20200930(第05期);全文 *

Also Published As

Publication number Publication date
CN112995289A (en) 2021-06-18

Similar Documents

Publication Publication Date Title
CN112995289B (en) Internet of vehicles multi-target computing task unloading scheduling method based on non-dominated sorting genetic strategy
CN112380008B (en) Multi-user fine-grained task unloading scheduling method for mobile edge computing application
CN113242568B (en) Task unloading and resource allocation method in uncertain network environment
CN111445111B (en) Electric power Internet of things task allocation method based on edge cooperation
CN112286677B (en) Resource-constrained edge cloud-oriented Internet of things application optimization deployment method
CN110418356A (en) A kind of calculating task discharging method, device and computer readable storage medium
CN113821318B (en) Internet of things cross-domain subtask combination collaborative computing method and system
CN111813506A (en) Resource sensing calculation migration method, device and medium based on particle swarm algorithm
CN115134242B (en) Vehicle-mounted computing task unloading method based on deep reinforcement learning strategy
CN112784362A (en) Hybrid optimization method and system for unmanned aerial vehicle-assisted edge calculation
CN113709249B (en) Safe balanced unloading method and system for driving assisting service
CN112214301B (en) Smart city-oriented dynamic calculation migration method and device based on user preference
CN115297171B (en) Edge computing and unloading method and system for hierarchical decision of cellular Internet of vehicles
CN115065683B (en) Vehicle edge network task allocation and unloading method based on vehicle clustering
CN106209618A (en) A kind of communication mixed networking method and system improving intelligence adapted electric energy effect
CN113918240A (en) Task unloading method and device
CN105323166A (en) Cloud computing-oriented routing method based on network efficiency priority
CN105677447A (en) Clustering-based delay bandwidth minimization virtual machine deployment method in distributed cloud
CN113590279A (en) Task scheduling and resource allocation method for multi-core edge computing server
CN114980216B (en) Dependency task unloading system and method based on mobile edge calculation
CN116634401A (en) Task unloading method for maximizing satisfaction of vehicle-mounted user under edge calculation
CN114615705B (en) Single-user resource allocation strategy method based on 5G network
CN113157344B (en) DRL-based energy consumption perception task unloading method in mobile edge computing environment
CN112910716A (en) Mobile fog calculation loss joint optimization system and method based on distributed DNN
CN118113484B (en) Resource scheduling method, system, storage medium and vehicle

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
GR01 Patent grant
GR01 Patent grant
CF01 Termination of patent right due to non-payment of annual fee

Granted publication date: 20220517

CF01 Termination of patent right due to non-payment of annual fee