CN115733838A - Vehicle networking multidimensional resource allocation method based on mobile edge calculation - Google Patents
Vehicle networking multidimensional resource allocation method based on mobile edge calculation Download PDFInfo
- Publication number
- CN115733838A CN115733838A CN202111017806.8A CN202111017806A CN115733838A CN 115733838 A CN115733838 A CN 115733838A CN 202111017806 A CN202111017806 A CN 202111017806A CN 115733838 A CN115733838 A CN 115733838A
- Authority
- CN
- China
- Prior art keywords
- task
- strategy
- vehicle
- calculation
- mec server
- 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 102
- 238000000034 method Methods 0.000 title claims abstract description 45
- 238000013468 resource allocation Methods 0.000 title claims abstract description 36
- 230000006855 networking Effects 0.000 title abstract description 16
- 239000011159 matrix material Substances 0.000 claims abstract description 39
- 230000002787 reinforcement Effects 0.000 claims description 9
- 238000013528 artificial neural network Methods 0.000 claims description 6
- 230000005540 biological transmission Effects 0.000 claims description 6
- 238000011156 evaluation Methods 0.000 claims description 6
- 230000008569 process Effects 0.000 claims description 4
- 230000009471 action Effects 0.000 claims description 3
- 230000007774 longterm Effects 0.000 claims description 3
- 239000000126 substance Substances 0.000 claims description 3
- 238000012549 training Methods 0.000 claims description 3
- 238000004220 aggregation Methods 0.000 abstract 1
- 230000002776 aggregation Effects 0.000 abstract 1
- 238000010586 diagram Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 230000007613 environmental effect Effects 0.000 description 2
- 230000006870 function Effects 0.000 description 2
- 238000012545 processing Methods 0.000 description 2
- 230000008859 change Effects 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 239000002360 explosive Substances 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000007726 management method Methods 0.000 description 1
- 229920003087 methylethyl cellulose Polymers 0.000 description 1
- 230000005012 migration Effects 0.000 description 1
- 238000013508 migration Methods 0.000 description 1
- 238000003062 neural network model Methods 0.000 description 1
- 230000008447 perception Effects 0.000 description 1
- 230000035945 sensitivity Effects 0.000 description 1
- 238000001228 spectrum Methods 0.000 description 1
- 230000003068 static effect Effects 0.000 description 1
Images
Landscapes
- Mobile Radio Communication Systems (AREA)
Abstract
The invention provides a vehicle networking multidimensional resource allocation method based on mobile edge calculation, which comprises the following steps: s1, acquiring wireless network environment data and vehicle calculation task information in a vehicle networking system in real time; s2, selecting a calculation task unloading strategy with optimal performance and a multi-dimensional resource allocation strategy of the MEC server for the vehicle user according to the parameters of the S1; s3, distributing multidimensional resources to the vehicle users according to the selected task unloading strategy and the multidimensional resource distribution strategy, and unloading the calculation tasks; s4, an MEC server with edge computing capability carried by the small base station calculates the QoE of the current strategy according to the completion delay of the computing task in the next iteration period and the current network environment parameters; s5, the MEC server updates a scheduling strategy matrix according to the fed-back QoE; s6, each time the strategy matrix is updated by the MEC servers for N times, uploading the strategy matrix to a central controller; s7, the central controller aggregates a plurality of local strategy matrixes through the Federal learning architecture and then sends back all MEC servers participating in aggregation; s8, repeating the steps, continuously iterating and updating, and finally enabling the performance of the parameter matrix of the task unloading and resource allocation strategy to be optimal; the invention can adaptively adjust the multi-dimensional resource allocation and calculation task unloading strategies in the vehicle networking system, meet the multi-dimensional resource requirements changing along with time, reduce the average time delay for completing the calculation task and improve the Qos of the user.
Description
Technical Field
The invention belongs to the technical field of car networking, and particularly relates to a car networking multidimensional resource allocation method based on mobile edge calculation.
Background
The internet of vehicles, i.e., the internet of things of vehicles, is receiving more and more attention from academic and industrial circles. The running vehicle is used as an information perception object, and network link between the vehicle and a vehicle, a person and a road service platform is realized by means of a new generation communication technology. By accessing the internet of vehicles, road safety and traffic efficiency are improved, and more vehicle applications and data services are enabled. However, applications and services requiring delay sensitivity still face many challenges due to the limited spectrum resources, on-board computing resources, and caching resources.
Mobile Edge Computing (MEC) is an effective method to solve the above problems. In the MEC framework, cloud computing capabilities are provided within the wireless access network proximate to these in-vehicle devices. Some machine learning algorithms do bring better flexibility and accuracy to mobile video bitrate selection, but the intensive computing requirements also increase the load pressure of a mobile cloud server and the operation pressure of a mobile intelligent device, so that a mode of combining the mobile intelligent device and edge computing can be selected to meet the challenges.
Compared with the traditional Mobile Cloud Computing (MCC), the MEC server is closer to the mobile user, and the mobile user can obtain the required computing resources by only one hop of wireless transmission, and has lower delay compared with the MCC. However, MEC servers have limited computing power and are less scalable. With the explosive growth of mobile users and emerging applications, simply relying on MECs cannot fully meet the computing offload needs of mobile users. Therefore, the MEC should not completely replace cloud computing or local computing of the mobile terminal, and the three should be coordinated and supplemented with each other to better meet the requirements of the mobile user.
On the other hand, the demand of multidimensional resources in the car networking system fluctuates greatly along with time, and in a traditional resource allocation mode, frequent reallocation of resources such as bandwidth and cache consumes great cost, and is very difficult to implement. Therefore, the allocation of the multidimensional resource will not change for a long period of time, and only the rough resource requirement in the period of time can be satisfied. However, in the car networking system, the resource demand fluctuates obviously along with time, and a static allocation mode causes a large amount of resources to be idle in some time periods; meanwhile, the resource shortage situation can be caused by the frequent occurrence of the sudden user request, which can greatly damage the use experience of the user. With the development of Software Defined Networking (SDN) and network function virtualization technology (NFV), a way to dynamically allocate resources over time becomes possible. Under the dynamic management method, the allocation of resources is adjusted every short period of time, so that the multi-dimensional resource requirement changing along with the time is met, and the service quality is improved.
Disclosure of Invention
The invention aims to solve the technical problem of providing an edge-computing-based multidimensional resource allocation method for the Internet of vehicles, which is based on vehicle-mounted equipment in an Internet of vehicles system, and automatically adjusts the dynamic migration of computing tasks in a vehicle-mounted equipment terminal and an edge computing node and the allocation of multidimensional resources in the Internet of vehicles system through an MEC (media independent center) server according to the processor load and the system resource condition of the vehicle-mounted equipment, thereby realizing the self-adaptive resource allocation changing along with time and improving the service quality of the system.
In order to solve the technical problems, the invention specifically adopts the following technical scheme:
the invention provides a vehicle networking multi-dimensional resource allocation method based on mobile edge calculation, which comprises the following steps of:
s1, acquiring performance parameters of a vehicle user in real time by a vehicle-mounted application, wherein the performance parameters comprise a calculation task generation condition, a network connection condition and vehicle-mounted equipment performance, and defining a calculation unloading task;
s2, selecting a calculation task unloading strategy and a multidimensional resource distribution strategy with optimal performance for a user by the MEC server according to parameter data acquired by the vehicle-mounted application;
s3, the vehicle user and the MEC server complete a calculation task according to the selected task unloading strategy and the resource allocation strategy;
s4, calculating the QoE of the current strategy by the MEC server according to the completion condition of the calculation task in the next iteration period, the resource allocation condition in the Internet of vehicles and the current network environment parameters;
s5, forming a strategy matrix by the MEC server according to the calculation task parameters fed back by the vehicle user, the network environment state and the user QoE after the MEC server selects the task mode;
s6, the MEC servers update the strategy matrix in cooperation with other MEC servers through the central controller; when the system environment changes and a vehicle generates a calculation task, the MEC server selects an unloading mode for the vehicle at the current moment according to the strategy matrix and distributes multidimensional resources in the system.
Further, in the method for allocating multidimensional resources of the internet of vehicles for mobile edge computing, provided by the present invention, in step S1, the calculation task generation condition includes the size of the calculated input data, including program codes and input files; the amount of computation required to complete this task, quantified by the number of cpu cycles; calculating the maximum waiting time of the task, namely delay constraint duration; the onboard device capabilities include the local computing capabilities of device i.
Further, in the step S2, the MEC server selects a multidimensional resource allocation policy with optimal performance and a calculation task offloading policy for the system according to parameter data acquired by the vehicle-mounted application, and the specific operations are as follows:
s21, representing the calculation task generated by the vehicle user as M according to the collected environment information i (t), the expression is as follows:
M i (t)＝{c i (ca) (t)，de i (t)，c i (co) (t)}
wherein, c i (ca) (t) is the size of the input data for the calculation, including program code, input files; c. C i (co) (t) represents the amount of computation required to complete the task, quantified by the number of cpu cycles; de i The method comprises the steps of calculating the maximum waiting time of a task, namely delay constraint duration, wherein the task is a calculation intensive task and can be divided for calculation;
s22, selecting a calculation task unloading strategy for the vehicles in the system according to the environment state and the reinforcement learning strategy matrix and distributing multidimensional resources in the system;
s23, allocating multidimensional resources: allocating bandwidth resources to vehicle users, and allocating cache resources and computing resources of the MEC server to computing tasks unloaded to the MEC server;
s24, unloading a calculation task: a calculation task offloading strategy is determined for each vehicle user within the system, i.e., a proportion of calculation tasks to be offloaded to the MEC server is selected.
Further, the method for allocating multidimensional resources in the internet of vehicles for mobile edge computing provided by the present invention is characterized in that, in step S22, the environment state and the reinforcement learning strategy matrix select a computation task offloading strategy for vehicles in the system and allocate multidimensional resources in the system, and the specific operations are as follows:
(1) And a vehicle set in the coverage range of one MEC server in the time slot t is represented as N (t), wherein N vehicles are shared. The vehicle i generates a calculation task in the time slot, and the user generates a calculation task according to an unloading strategy a i (t)∈[0，1]Selecting to unload the calculation tasks to the MEC server to execute calculation, wherein the parameter represents the proportion of the calculation tasks unloaded to the MEC server by the vehicle, and the closer to 1, the more the calculation tasks are unloaded to the MEC server;
(2) The multidimensional resource used by the small cell to handle the offload task is denoted as { C (ca) ，C (sp) ，C (co) Denotes a cache resource, a bandwidth resource and a computing resource, respectively, and a computing resource usable by the vehicle itself is denoted as C i (co) The computational resources are quantified by the number of cpu cycles an MEC server can perform per second. The multidimensional resource occupation ratio of each item allocated to the vehicle i by the small base station m can be expressed as { f } i (ca) (t)，f i (sp) (t)，f i (co) (t)}。
Further, the multidimensional resource allocation method of the internet of vehicles for mobile edge computing provided by the present invention is characterized in that the completion of the computing task in step S4 specifically comprises the following operations:
s41, determining an unloading strategy of a vehicle user, and distributing multidimensional resources by an MEC server;
s42, the vehicle user i carries out unloading strategy a i (t) uploading task data to the small base station, and starting to calculate the task, wherein the steps are as follows:
(1) When allocating to it a cache resource C (ca) f i (ca) (t)≥c i (ca) (t), the offloading policy and resource allocation are unchanged; otherwise will a i (t)，f i (ca) (t)，f i (sp) (t)，f i (co) (t) setting all the components to 0; ,
(2) The vehicle user i uploads the task to the MEC server, considering a general path loss model for the uplink between the vehicle and the small cell. Assuming that the transmission power of the vehicle is fixed to P, the power attenuation speed of the transmission signal is
d m，i Is the horizontal distance, h, between the vehicle and the small base station m Is the small base station height, and gamma is the path loss exponent; the ratio of the bandwidth resources distributed to the vehicle i by the small base station m to the total available resources is f i (sp) (t)；
According to the Shannon formula, the uplink transmission rate when the vehicle unloads the tasks to the small base station is
Wherein the content of the first and second substances,
representing the gain, σ, of the uplink channel from vehicle i to the small base station during time slot t 2 Is the received background noise.
According to the formula, the time for uploading the unloading task from the vehicle to the small base station can be calculated as
(3) The MEC server uses the computing resources C allocated to the computing task (co) (t)f i (co) (t) calculating. The computation time of a task is the number of cpu cycles required for that task divided by the cpu proportion of that task at the MEC server. The computation task completion time unloaded to the MEC server is T up，i (t)。T up，i (t) consists of three parts, which are the upload time of the task, the computation time at the small cell site and the return time of the final computation result. Compared with the former two, the return time of the calculation result is negligible, so the completion time of the unloading task can be expressed as the sum of the uploading time and the calculation time of the task,
(4) The vehicle-mounted equipment processes the local calculation task according to the unloading strategy, and the time for completing the local task is recorded as T loc，i (t), namely, calculating the time of the task for the vehicle-mounted equipment, and representing the following steps:
s43, the completion time of the calculation task Mi (T) generated by the vehicle user i in the time slot T is recorded as T i (t)，
T i (t)＝max{T loc，i (t)，T up，i (t)}
Further, the method for calculating a real-time video bitrate self-adaption for mobile edge provided by the present invention is characterized in that, in step S4, the MEC server calculates a QoE of a current policy according to an environmental parameter after policy execution and updates a policy form matrix, and specifically operations are as follows:
s51, modeling the strategy into pi θ (s, a) representing selecting action a in state s, wherein the weight parameters in the network are iteratively updated for the θ strategy;
s52, obtaining the next environment state s according to the selected strategy t+1 And immediate return r t The quadruple is divided into four groups<s t ，a t ，r t ，s t+1 >Storing the data into an experience playback buffer B;
s53, training a neural network theta for evaluating the importance of the current strategy by using data in the empirical replay buffer μ The network predicts the long-term reward expectation of (s, a), i.e., Q (s, a). Note the book
Updating the neural network θ μ ←θ μ + β δ φ (s, a), wherein
(s ', a') is sample data.
S54, updating the strategy weight parameter theta according to the strategy gradient method
Further, the multidimensional resource allocation method of the internet of vehicles for mobile edge computing provided by the present invention is characterized in that the step S6: the method comprises the following steps that a plurality of MEC servers and other MEC servers cooperatively update a strategy matrix through a central controller, and the method specifically comprises the following operations:
s61, after repeating the steps S1 to S5 for a plurality of times, all MEC servers upload the strategy matrix and the evaluation matrix to the central server;
s62, using a federal average algorithm to aggregate the parameter matrixes to obtain two global strategy matrixes and a global evaluation matrix which are marked as theta g ，
Where Dm is the number of samples in MEC server m, and D is the sum of the number of samples for all servers.
And S63, returning the global matrix to the MEC server to replace the local weight parameter.
Further, the multidimensional resource allocation method of the internet of vehicles for mobile edge computing provided by the invention is characterized by further comprising the following steps of S7: and repeating the steps S1 to S6, and continuously iterating and updating to finally enable the performance of the edge scheduling strategy to be optimal.
The invention can obtain the following advantages by adopting the technical means:
the invention provides a vehicle networking multidimensional resource allocation method for mobile edge computing, which is mainly used for uploading part of computing tasks generated by vehicle users to an MEC server platform to be executed according to the characteristic that multidimensional resource consumption fluctuates along with time in a vehicle networking system and the requirements of the users on efficient computing service and data privacy security, selecting an unloading strategy for the vehicle users by adopting a reinforcement learning method, and allocating multidimensional resources in the vehicle networking system. The method comprehensively considers the service quality of the Internet of vehicles system and the data privacy safety of vehicle users, and minimizes the average time delay of task completion in the system under the constraint of limited resources. Meanwhile, by adopting the federal learning architecture, the data privacy safety of the user is protected, the convergence rate of reinforcement learning is increased, the expandability of the Internet of vehicles is improved, more Internet of vehicles service providers are encouraged to join the system, and the system performance is improved.
Drawings
Fig. 1 is a network architecture diagram of a multidimensional resource allocation method for internet of vehicles based on mobile edge computing in an embodiment of the present invention.
Fig. 2 is a schematic diagram of a policy update method in the embodiment of the present invention.
Detailed Description
The technical scheme of the invention is further explained by combining the drawings as follows:
the invention provides a vehicle networking multi-dimensional resource allocation method based on mobile edge calculation, which specifically comprises the following steps:
a multidimensional resource allocation method of Internet of vehicles based on mobile edge calculation is characterized by comprising the following steps:
A. the method comprises the steps that a vehicle-mounted application collects performance parameters of a vehicle user in real time, wherein the performance parameters comprise calculation task generation conditions, network connection conditions and vehicle-mounted equipment performance, and defines calculation unloading tasks;
B. the MEC server selects a calculation task unloading strategy and a multidimensional resource allocation strategy with optimal performance for a user according to parameter data acquired by the vehicle-mounted application;
C. the vehicle user and the MEC server complete the calculation task according to the selected task unloading strategy and the resource allocation strategy;
D. the MEC server calculates the QoE of the current strategy according to the completion condition of the calculation task during the iteration, the resource allocation condition in the Internet of vehicles system and the current network environment parameters, so as to evaluate the importance of the current strategy and update the strategy matrix;
E. the method comprises the following steps that a plurality of MEC servers and other MEC servers cooperatively update a strategy matrix through a central controller; when the system environment changes and a vehicle generates a calculation task, the MEC server selects an unloading mode for the vehicle at the current moment according to the strategy matrix and distributes multidimensional resources in the system;
F. repeating the steps A to E, continuously iterating and updating, and finally enabling the performance of the strategy matrix to be optimal;
further, the calculation task generation condition in the step a includes a calculated size of input data, including a program code and an input file; the amount of computation required to complete this task, quantified by the number of cpu cycles; calculating the maximum waiting time of the task, namely delay constraint duration; the onboard device capabilities include the local computing capabilities of device i.
Further, in the step B, the MEC server selects a multidimensional resource allocation strategy and a calculation task unloading strategy with optimal performance for the system according to the parameter data acquired by the vehicle-mounted application, and the specific operations are as follows:
B1. representing a vehicle user-generated computing task as M from collected environmental information i (t), the expression is as follows:
M i (t)＝{c i (ca) (t)，de i (t)，c i (co) (t)}
wherein, c i (ca) (t) is the size of the input data for the calculation, including program code, input files; c. C i (co) (t) represents the amount of computation required to complete the task, quantified by the number of cpu cycles; de i The method comprises the steps of calculating the maximum waiting time of a task, namely delay constraint duration, wherein the task is a calculation intensive task and can be divided for calculation;
B2. selecting a calculation task unloading strategy for vehicles in the system according to the environment state and the reinforcement learning strategy matrix and distributing multi-dimensional resources in the system;
B3. and (3) allocating multidimensional resources: allocating bandwidth resources to vehicle users, and allocating cache resources and computing resources of the MEC server to computing tasks unloaded to the MEC server;
B4. and (3) unloading a computing task: determining a calculation task unloading strategy for each vehicle user in the system, namely selecting the proportion of calculation tasks to be unloaded to an MEC server;
further, in the step B2, the environment state and the reinforcement learning strategy matrix select a calculation task unloading strategy for the vehicle in the system and allocate multidimensional resources in the system, specifically, the operations are as follows:
(1) The vehicle set in the coverage range of one MEC server in the time slot t is represented as N (t), wherein N vehicles are shared; the vehicle i generates a calculation task in the time slot, and the user unloads the vehicle according to the taskPolicy a i (t)∈[0，1]Selecting to unload the calculation tasks to the MEC server to execute calculation, wherein the parameter represents the proportion of the calculation tasks unloaded to the MEC server by the vehicle, and the closer to 1, the more the calculation tasks are unloaded to the MEC server;
(2) The multidimensional resource used by the small cell to handle the offload task is denoted as { C (ca) ，C (sp) ，C (co) Denotes a cache resource, a bandwidth resource and a computing resource, respectively, and a computing resource usable by the vehicle itself is denoted as C i (co) The computing resources are quantified by the number of cpu cycles a MEC server can perform per second; the multidimensional resource occupation ratio of each item allocated to the vehicle i by the small base station m can be expressed as { f } i (ca) (t)，f i (sp) (t)，f i (co) (t)}。
Further, the completion of the calculation task in the step C specifically includes the following operations:
C1. determining an unloading strategy of a vehicle user, and distributing multidimensional resources by an MEC server;
C2. the vehicle user i follows the unloading strategy a i (t) uploading task data to the small base station, and starting to calculate the task, wherein the steps are as follows:
(1) When allocated to its cache resource C (ca) f i (ca) (t)≥c i (ca) (t), the offloading policy and resource allocation are unchanged; otherwise will a i (t)，f i (ca) (t)，f i (sp) (t)，(f i (co) (t) setting all the components to 0; ,
(2) A vehicle user i uploads a task to an MEC server, and a general path loss model is considered for an uplink between a vehicle and a small base station; let the transmitting power of the vehicle be fixed as P and the power attenuation speed of the transmitting signal be
d m，i Is the horizontal distance, h, between the vehicle and the small base station m Is the small base station height, and gamma is the path loss exponent; the ratio of the bandwidth resource allocated to the vehicle i by the small base station m to the total available resource is f i (sp) (t)；
According to the shannon formula, the uplink transmission rate when the vehicle unloads the tasks to the small base station is as follows:
wherein the content of the first and second substances,
representing the gain, σ, of the uplink channel from vehicle i to the small base station between time slots t 2 Is the received background noise;
according to the above formula, the time for uploading the unloading task from the vehicle to the small base station can be calculated as
(3) The MEC server uses the computing resources C allocated to the computing task (co) (t)f i (co) (t) performing an operation. The calculation time of the task is the ratio of the number of the cpu cycles needed by the task to the cpu of the MEC server; the computation task completion time unloaded to the MEC server is T up，i (t)；T up，i (t) is composed of three parts, which are the uploading time of the task, the calculation time at the small base station and the returning time of the final calculation result; compared with the former two, the return time of the calculation result is negligible, so the completion time of the unloading task can be expressed as the sum of the uploading time and the calculation time of the task, i.e. the calculation time
(4) The vehicle-mounted equipment processes the local calculation task according to the unloading strategy, and the completion time of the local task is recorded as T loc，i (t), namely, calculating the time of the task for the vehicle-mounted equipment, and representing the following steps:
C3. the completion time of the calculation task Mi (T) generated by the vehicle user i in the time slot T is recorded as T i (t)，
T i (t)＝max{T loc，i (t)，T up，i (t)}
Further, in step D, the MEC server calculates a QoE of the current policy according to the environment parameter after the policy execution and updates the policy form matrix, specifically operating as follows:
D1. modeling a policy as modeling the policy as pi θ (s, a) representing selecting action a in state s, wherein the weight parameters in the network are iteratively updated for the θ strategy;
D2. obtaining the next environment state s according to the selected strategy t+1 And immediate reporting r t The quadruple is divided into four groups<s t ，a t ，r t ，s t+1 >Storing the data into an experience playback buffer B;
D3. training a neural network θ for evaluating the importance of a current strategy using data in an empirical replay buffer μ The network predicts the long-term reward expectation of (s, a), i.e., Q (s, a), as
Updating the neural network θ μ ←θ μ + β δ φ (s, a), wherein
(s', a) is sample data;
D4. updating the policy weight parameter θ according to a policy gradient method
Further, in the step E, the MEC servers update the policy matrix in cooperation with other MEC servers through the central controller, and the specific operations are as follows:
E1. after repeating the steps S1 to S5 for a plurality of times, all MEC servers upload the strategy matrix and the evaluation matrix to the central server;
E2. using a federal average algorithm to aggregate the parameter matrixes to obtain two global strategy matrixes and a global evaluation matrix which are marked as theta g ，
Where Dm is the number of samples in the MEC server m, and D is the sum of the number of samples of all servers;
E3. and returning the global matrix to the MEC server to replace the local weight parameters.
Further, the step F: and repeating the step A to the step E, and continuously iterating and updating to finally enable the performance of the edge scheduling strategy to be optimal.
The system architecture of the present invention is shown in fig. 1, and the detailed functions are as follows:
the terminal equipment in the system architecture is vehicle-mounted equipment on a mobile vehicle. The vehicle-mounted equipment collects the information of the vehicle, namely local processing capacity and generated calculation task information are uploaded to an MEC server carried by the small base station; and simultaneously, the vehicle-mounted equipment processes a part of calculation tasks according to the unloading strategy.
On one hand, the MEC server collects system network environment information and information uploaded by vehicles, trains a reinforcement learning model, and determines a calculation task unloading strategy and a multidimensional resource allocation strategy; and on the other hand, processing the calculation tasks unloaded to the server side.
The central controller is connected with each MEC server and aggregates the neural network models of the MEC servers.
Each small base station and the vehicles in the coverage area are mutually independent vehicle networking systems and are connected only through a central controller.
The embodiments of the present invention have been described in detail with reference to the drawings, but the present invention is not limited to the above embodiments, and various changes can be made within the knowledge of those skilled in the art without departing from the gist of the present invention.
Claims (8)
1. A multidimensional resource allocation method of Internet of vehicles based on mobile edge calculation is characterized by comprising the following steps:
s1, acquiring performance parameters of a vehicle user in real time by a vehicle-mounted application, wherein the performance parameters comprise a calculation task generation condition, a network connection condition and vehicle-mounted equipment performance, and defining a calculation unloading task;
s2, selecting a calculation task unloading strategy and a multidimensional resource allocation strategy with optimal performance for a user by the MEC server according to parameter data acquired by the vehicle-mounted application;
s3, the vehicle user and the MEC server complete a calculation task according to the selected task unloading strategy and the selected resource allocation strategy;
s4, calculating the QoE of the current strategy by the MEC server according to the completion condition of the calculation task in the next iteration period, the resource allocation condition in the Internet of vehicles and the current network environment parameters;
s5, an MEC server forms a strategy matrix according to the calculation task parameters fed back by the vehicle user, the network environment state and the user QoE after the task mode is selected;
s6, the MEC servers update the strategy matrix in cooperation with other MEC servers through the central controller; when the system environment changes and a vehicle generates a calculation task, the MEC server selects an unloading mode for the vehicle at the current moment according to the strategy matrix and distributes multidimensional resources in the system;
and S7, repeating the steps S1 to S6, and continuously iterating and updating to finally enable the performance of the strategy matrix to be optimal.
2. The method for allocating multidimensional resources in the internet of vehicles for mobile edge computing as claimed in claim 1, wherein in step S1, the computing task generation condition includes the size of the computed input data, including program codes and input files; the amount of computation required to complete this task, quantified by the number of cpu cycles; calculating the maximum waiting time of the task, namely delay constraint duration; the onboard device capabilities include the local computing capabilities of device i.
3. The method for allocating multidimensional resources in the internet of vehicles for mobile edge computing according to claim 1, wherein in step S2, the MEC server selects a multidimensional resource allocation policy with optimal performance and a calculation task offloading policy for the system according to parameter data collected by the vehicle-mounted application, and specifically:
s21, representing the calculation task generated by the vehicle user as M according to the acquired environment information i (t), the expression is as follows:
wherein, c i (ca) (t) is the size of the input data for the calculation, including program code, input files; c. C i (co) (t) represents the amount of computation required to complete the task, quantified by the number of cpu cycles; de i The maximum waiting time of a calculation task, namely delay constraint duration, is calculated, and the task is a calculation intensive task and can be divided for calculation;
s22, selecting a calculation task unloading strategy for the vehicles in the system according to the environment state and the reinforcement learning strategy matrix and distributing multidimensional resources in the system;
s23, distributing multidimensional resources: allocating bandwidth resources to vehicle users, and allocating cache resources and computing resources of the MEC server to computing tasks unloaded to the MEC server;
s24, unloading a calculation task: a calculation task offloading strategy is determined for each vehicle user within the system, i.e., a proportion of calculation tasks to be offloaded to the MEC server is selected.
4. The method as claimed in claim 3, wherein in step S22, the environment state and reinforcement learning strategy matrix selects a calculation task unloading strategy for the vehicle in the system and allocates the multidimensional resource in the system, specifically as follows:
(1) The vehicle set in the coverage range of one MEC server in the time slot t is represented as N (t), wherein N vehicles are shared; the vehicle i generates a calculation task in the time slot, and the user generates a calculation task according to an unloading strategy a i (t)∈[0，1]Selecting to unload the calculation tasks to the MEC server to execute calculation, wherein the parameter represents the proportion of the calculation tasks unloaded to the MEC server by the vehicle, and the closer to 1, the more the calculation tasks are unloaded to the MEC server;
(2) The multidimensional resource used by the small cell to handle the offload task is denoted as { C (ca) ，C (sp) ，C (co) Respectively representing cache resources, bandwidth resources and computing resources, and the computing resources available to the vehicle are represented as
The computational resources are quantified by the number of cpu cycles an MEC server can perform per second. The multi-dimensional resource occupation ratio allocated to the vehicle i by the small base station m can be expressed as
5. The method for allocating multidimensional resources in the internet of vehicles for mobile edge computing according to claim 1, wherein the completion of the computing task in step S4 is specifically performed as follows:
s41, determining an unloading strategy of a vehicle user, and distributing multidimensional resources by an MEC server;
s42, the vehicle user i carries out unloading strategy a i (t) uploading task data to the small cell and starting to calculate the task,the method comprises the following steps:
(1) When allocated to its cache resources
When the load is not changed, the load-off strategy and the resource allocation are not changed; otherwise will a i (t)，f i (ca) (t)，f i (sp) (t)，f i (co) (t) setting all the components to 0; ,
(2) The vehicle user i uploads the task to the MEC server, considering a general path loss model for the uplink between the vehicle and the small cell. Let the transmitting power of the vehicle be fixed as P and the power attenuation speed of the transmitting signal be
d m，i Is the horizontal distance, h, between the vehicle and the small base station m Is the small base station height, and gamma is the path loss exponent; the ratio of the bandwidth resource allocated to the vehicle i by the small base station m to the total available resource is f i (sp) (t); according to the shannon formula, the uplink transmission rate when the vehicle unloads the tasks to the small base station is as follows:
wherein the content of the first and second substances,
representing the gain, σ, of the uplink channel from vehicle i to the small base station between time slots t 2 Is the received background noise;
according to the formula, the time for uploading the unloading task from the vehicle to the small base station can be calculated as follows:
(3) The MEC server uses the computing resources C allocated to the computing task (co) (t)f i (co) (t) performing an operation; the calculation time of the task is the ratio of the number of the cpu cycles needed by the task to the cpu of the MEC server; the computation task completion time unloaded to the MEC server is T up，i (t)；T up，i (t) consists of three parts, which are the upload time of the task, the computation time at the small cell site and the return time of the final computation result. Compared with the former two, the return time of the calculation result is negligible, so the completion time of the unloading task can be expressed as the sum of the uploading time and the calculation time of the task;
(4) The vehicle-mounted equipment processes the local calculation task according to the unloading strategy, and the completion time of the local task is recorded as T loc，i (t), namely, calculating the time of the task for the vehicle-mounted equipment, and representing the following steps:
s43, the completion time of the calculation task Mi (T) generated by the vehicle user i in the time slot T is recorded as T i (t)，
T i (t)＝max{T loc，i (t)，T up，i (t)} 。
6. The method of claim 1, wherein in step S4, the MEC server calculates QoE of the current policy according to the environment parameters after the policy execution and updates the policy-like matrix, and the method specifically includes the following operations:
s51, modeling the strategy into pi θ (s, a) representing the selection of action a in state s, wherein the weight parameters in the network are iteratively updated for the θ strategy;
s52, obtaining the next environment state s according to the selected strategy t+1 And immediate return r t The four isTuple<s t ，a t ，r t ，s t+1 >Storing the data into an experience playback buffer B;
s53, training a neural network theta for evaluating the importance of the current strategy by using data in the empirical replay buffer μ The network predicts the long-term reward expectation of (s, a), i.e., Q (s, a), as
Updating the neural network θ μ ←θ μ + β δ φ (s, a), wherein
(s ', a') is sample data;
s54, updating the strategy weight parameter theta according to the strategy gradient method
7. The multidimensional resource allocation method for the internet of vehicles for the mobile edge computing as claimed in claim 1, wherein the step S6: the method comprises the following steps that a plurality of MEC servers and other MEC servers cooperatively update a strategy matrix through a central controller, and the method specifically comprises the following operations:
s61, after repeating the steps S1 to S5 for a plurality of times, all MEC servers upload the strategy matrix and the evaluation matrix to the central server;
s62, using a federal average algorithm to aggregate the parameter matrixes to obtain two global strategy matrixes and a global evaluation matrix which are marked as theta g ，
Where Dm is the number of samples in MEC server m, and D is the sum of the number of samples for all servers.
And S63, returning the global matrix to the MEC server to replace the local weight parameter.
8. The multidimensional resource allocation method for the internet of vehicles for the mobile edge computing as claimed in claim 1, wherein the step S7: and repeating the steps S1 to S6, and continuously iterating and updating to finally lead the performance of the edge scheduling strategy to be optimal.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111017806.8A CN115733838A (en) | 2021-08-31 | 2021-08-31 | Vehicle networking multidimensional resource allocation method based on mobile edge calculation |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111017806.8A CN115733838A (en) | 2021-08-31 | 2021-08-31 | Vehicle networking multidimensional resource allocation method based on mobile edge calculation |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115733838A true CN115733838A (en) | 2023-03-03 |
Family
ID=85291867
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111017806.8A Pending CN115733838A (en) | 2021-08-31 | 2021-08-31 | Vehicle networking multidimensional resource allocation method based on mobile edge calculation |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115733838A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115988462A (en) * | 2023-03-17 | 2023-04-18 | 中电建市政建设集团山东工程有限公司 | Debugging method of edge computing module based on vehicle-road cooperation |
-
2021
- 2021-08-31 CN CN202111017806.8A patent/CN115733838A/en active Pending
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115988462A (en) * | 2023-03-17 | 2023-04-18 | 中电建市政建设集团山东工程有限公司 | Debugging method of edge computing module based on vehicle-road cooperation |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110035410B (en) | Method for joint resource allocation and computational offloading in software-defined vehicle-mounted edge network | |
| CN110996393B (en) | Single-edge computing server and multi-user cooperative computing unloading and resource allocation method | |
| CN111586696B (en) | Resource allocation and unloading decision method based on multi-agent architecture reinforcement learning | |
| CN111431941B (en) | Real-time video code rate self-adaption method based on mobile edge calculation | |
| CN111818168A (en) | Self-adaptive joint calculation unloading and resource allocation method in Internet of vehicles | |
| CN111277437A (en) | Network slice resource allocation method for smart power grid | |
| CN107708152B (en) | Task unloading method of heterogeneous cellular network | |
| CN112188627B (en) | Dynamic resource allocation strategy based on state prediction | |
| CN112422644A (en) | Method and system for unloading computing tasks, electronic device and storage medium | |
| CN115209426B (en) | Dynamic deployment method for digital twin servers in edge car networking | |
| CN113452566A (en) | Cloud edge side cooperative resource management method and system | |
| CN116541106B (en) | Computing task unloading method, computing device and storage medium | |
| CN114885420A (en) | User grouping and resource allocation method and device in NOMA-MEC system | |
| CN112153728B (en) | Optimization method for base station association and module dormancy | |
| CN111526526B (en) | Task unloading method in mobile edge calculation based on service mashup | |
| Lan et al. | Deep reinforcement learning for computation offloading and caching in fog-based vehicular networks | |
| CN114422349A (en) | Cloud-edge-end-collaboration-based deep learning model training and reasoning architecture deployment method | |
| CN113271221B (en) | Network capacity opening method and system and electronic equipment | |
| CN115733838A (en) | Vehicle networking multidimensional resource allocation method based on mobile edge calculation | |
| CN115835294A (en) | RAN slice and task unloading joint optimization method assisted by deep reinforcement learning in Internet of vehicles | |
| US9253781B2 (en) | Scheduling in consideration of terminal groups in a mobile communication system | |
| CN116781788B (en) | Service decision method and service decision device | |
| CN113821346A (en) | Computation uninstalling and resource management method in edge computation based on deep reinforcement learning | |
| CN117580063A (en) | Multi-dimensional resource collaborative management method in vehicle-to-vehicle network | |
| KR20210147240A (en) | Energy Optimization Scheme of Mobile Devices for Mobile Augmented Reality Applications in Mobile Edge Computing |
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 |