CN115065683B - Vehicle edge network task allocation and unloading method based on vehicle clustering - Google Patents
Vehicle edge network task allocation and unloading method based on vehicle clustering Download PDFInfo
- Publication number
- CN115065683B CN115065683B CN202210888963.4A CN202210888963A CN115065683B CN 115065683 B CN115065683 B CN 115065683B CN 202210888963 A CN202210888963 A CN 202210888963A CN 115065683 B CN115065683 B CN 115065683B
- Authority
- CN
- China
- Prior art keywords
- task
- vehicle
- cluster
- unloading
- clustering
- 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.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 41
- 238000004364 calculation method Methods 0.000 claims abstract description 48
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 18
- 238000013528 artificial neural network Methods 0.000 claims abstract description 13
- 238000012549 training Methods 0.000 claims abstract description 10
- 230000006870 function Effects 0.000 claims description 31
- 230000009471 action Effects 0.000 claims description 17
- 230000005540 biological transmission Effects 0.000 claims description 13
- 238000005265 energy consumption Methods 0.000 claims description 6
- 230000007613 environmental effect Effects 0.000 claims description 6
- 230000004913 activation Effects 0.000 claims description 3
- 230000008569 process Effects 0.000 abstract description 7
- 238000013135 deep learning Methods 0.000 description 4
- 238000012545 processing Methods 0.000 description 4
- 239000003795 chemical substances by application Substances 0.000 description 3
- 238000004891 communication Methods 0.000 description 3
- 238000003064 k means clustering Methods 0.000 description 3
- 238000011161 development Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 230000006855 networking Effects 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000000593 degrading effect Effects 0.000 description 1
- 230000001934 delay Effects 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 229920003087 methylethyl cellulose Polymers 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000007781 pre-processing Methods 0.000 description 1
- 238000004088 simulation Methods 0.000 description 1
- 238000001228 spectrum Methods 0.000 description 1
- 230000007704 transition Effects 0.000 description 1
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
-
- 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
- H04L67/104—Peer-to-peer [P2P] networks
- H04L67/1044—Group management mechanisms
-
- 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
- H04L67/104—Peer-to-peer [P2P] networks
- H04L67/1074—Peer-to-peer [P2P] networks for supporting data block transmission mechanisms
-
- 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
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04W—WIRELESS COMMUNICATION NETWORKS
- H04W4/00—Services specially adapted for wireless communication networks; Facilities therefor
- H04W4/30—Services specially adapted for particular environments, situations or purposes
- H04W4/40—Services specially adapted for particular environments, situations or purposes for vehicles, e.g. vehicle-to-pedestrians [V2P]
- H04W4/46—Services specially adapted for particular environments, situations or purposes for vehicles, e.g. vehicle-to-pedestrians [V2P] for vehicle-to-vehicle communication [V2V]
-
- 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
- Y02D30/00—Reducing energy consumption in communication networks
- Y02D30/70—Reducing energy consumption in communication networks in wireless communication networks
Landscapes
- Engineering & Computer Science (AREA)
- Computer Networks & Wireless Communication (AREA)
- Signal Processing (AREA)
- Computing Systems (AREA)
- Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Theoretical Computer Science (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Medical Informatics (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a vehicle edge network task allocation and unloading method based on vehicle clustering, which mainly comprises the following steps: receiving the position information of the vehicle terminal in the coverage area of the RSU and the task attribute information carried by the vehicle terminal in real time, calculating task unloading probability, and defining task priority of the vehicle terminal; dividing a vehicle terminal into H clustering clusters by adopting a K-means user clustering method, and obtaining task priority of each clustering cluster; the cluster with the highest task priority is distributed to edge calculation, and the cluster with the lowest task priority is distributed to local calculation; for other cluster clusters, obtaining an optimal distributed computing unloading strategy through iterative training of a deep neural network learning algorithm; and sending the strategy to each vehicle cluster. The invention provides more reasonable basis for the calculation resource scheduling process of the MEC server, not only improves the unloading execution success rate of the vehicle terminal calculation task, but also improves the execution success rate of the calculation task and the unloading decision accuracy rate, and effectively reduces the execution time delay of the calculation task.
Description
Technical Field
The invention relates to the technical field of Internet of vehicles communication, in particular to a vehicle edge network task allocation and unloading method based on vehicle clustering.
Background
The internet of things (IoV) is an important branch of the internet of things, and automobiles are introduced into the internet to realize various vehicle-oriented applications, so that the development of the internet of things technology also directly promotes the development of an Intelligent Transportation System (ITS). As a network dedicated to vehicle communication, ioV can realize efficient and reliable vehicle communication by transmitting traffic information, improving the safety of the intelligent traffic system. Compared with devices such as a sensor, a smart phone, a notebook computer, a wearable device and the like, the vehicle has more powerful computing capability, and can be equipped with more advanced hardware and software due to the large size and continuous energy supply. Therefore, the vehicle can locally support complex task processing to meet the high requirements of the internet of things application program, such as autopilot, driving assistance, intelligent interaction and the like.
However, many in-vehicle applications require significant computing resources and have latency limitations. For example, vehicles perceive the surrounding environment through cameras and identify potentially dangerous vehicles through deep learning-based methods to help drivers drive safely, which is both resource-consuming and time-consuming. When a vehicle itself has a large number of tasks to be completed, the computational resources of the vehicle are limited, which can create significant delays that cannot accommodate these computationally intensive tasks. Previous studies have mainly utilized cloud computing methods to handle computing-consuming tasks. Resources of cloud computing are sufficient, and the performance of computing consumption services can be improved theoretically. However, since the cloud computing server is typically far from the vehicle, long task transmission time, high link load, and even link congestion may occur, which may result in long task completion time, thereby severely degrading the quality of service. Even if the vehicle terminals have more powerful computing power, serious resource scarcity problems, such as limited battery power and CPU computing resources, are faced.
Currently, moving Edge Computing (MEC) has been an effective paradigm to solve the above problems. As a promising emerging technology, it was normalized and standardized in 2016. MECs can provide users with low transmission latency and high reliability services by moving cloud servers close to the network edge of the device. In the MEC computing mode, the edge server can execute computing tasks carried by some vehicle terminals, and the computing capacity of the edge server is far greater than that of the vehicle terminals, so that the service quality of the task unloading is improved, and the computing capacity of the vehicle terminals is effectively expanded. However, in view of the complex application environment of the internet of vehicles, the spectrum resources are limited, and the edge devices cannot support a large number of vehicle terminals to offload computing tasks together.
Disclosure of Invention
Based on the above, it is necessary to provide a vehicle edge network task allocation and offloading method based on vehicle clustering, which can improve the offloading success rate of the vehicle networking task offloading method, and can prioritize the vehicle edge network task based on the vehicle terminal or the carrying task thereof, and process the vehicle edge network task according to different demands of different sets of vehicle terminals on the computational load.
The invention discloses a vehicle edge network task allocation and unloading method based on vehicle clustering, which mainly comprises the following steps:
step S1: receiving the position information of a vehicle terminal in the coverage area of an RSU and task attribute information carried by the vehicle terminal in real time, calculating task unloading probability according to the task attribute information, and defining task priority of the vehicle terminal based on the position information and the task unloading probability;
step S2: dividing the vehicle terminal into H clustering clusters by adopting a K-means user clustering method based on the task priority of the vehicle terminal, and obtaining the task priority of each clustering cluster;
step S3: the cluster with the highest task priority is distributed to edge calculation, and the cluster with the lowest task priority is distributed to local calculation;
step S4: for other cluster groups between the highest and lowest task priorities, obtaining an optimal distributed computing and unloading strategy through iterative training of a deep neural network learning algorithm, wherein the optimal distributed computing and unloading strategy has the lowest system cost;
step S5: and sending the optimal distributed computing unloading strategy to each cluster in the coverage area of the RSU, so that the vehicle terminals in the cluster clusters perform data transmission of task unloading through the RSU according to the optimal distributed computing unloading strategy.
In the step S1, the location information of the vehicle terminal is a distance from the vehicle terminal to the MEC server; the task attribute information comprises a task execution time delay requirement, a task size and an actual task execution time delay;
the priority characteristics of the vehicle terminals are represented by a tuple p i Definition, p i ＝＜d i ，Pr i ＞，d i Representing the distance, pr, of the ith vehicle terminal to the MEC server i Representing a vehicle task offloading probability of an i-th vehicle terminal;
the vehicle task unloading probability Pr i From probability function Pr (L i,th ＜L i,e ) Obtaining, wherein L th ＜L e Representing actual task execution time delay L e Exceeding task execution delay requirement L th Requires unloading, probability function Pr (L i,th ＜L i,e ) The ratio of the number of tasks to be offloaded in the tasks carried by the ith vehicle terminal is 1 at the maximum;
the actual task execution time delay calculation formula is as follows:
L i,e ＝m i /F i ；
wherein m is i Task size for ith vehicle terminal, F i The computing power of the i-th vehicle terminal.
The step S2 specifically includes:
s21: k vehicle terminals are selected as initial clustering centers and k clustering clusters are formed;
s22: the Euclidean distance between the rest of the vehicle terminals and each cluster center is calculated, and the rest of the vehicle terminals are distributed to the cluster closest to the rest of the vehicle terminals based on the Euclidean distance;
s23: re-computing the centroid of the cluster after each allocation;
s24: returning to the step S22, and finally determining to divide the vehicle terminal into K clustering clusters until the error square sum SEE is 1; the centroid of the cluster reflects the task priority of the cluster, and the smaller the numerical value is, the higher the task priority is;
the calculation formula of the error square sum SEE is as follows:
wherein p is i Representing the priority characteristics of the ith vehicle terminal, c h The centroid of the H clustering cluster is represented, H represents the current clustering quantity, and the optimal clustering quantity is not reached at the moment;
the centroid calculation formula is as follows:
wherein,
wherein D is h Representing the cluster h, h.epsilon.1, K]U represents the number of vehicle terminals to be clustered, h i And (5) indicating the cluster index of the ith vehicle terminal.
The step S4 specifically includes:
during the iteration time t, the current environmental state s t Representing current exploration information from the environment, a t Representing cluster clusters D h Unloading action of medium-terminal vehicle and clustering cluster D h The rewards of the terminal vehicles are the system cost of the terminal vehicles;
using Q valuesThe network is used as a policy judgment standard, and the Q value represents < s t ,a t The value of > i.e. at the current environmental state s t The cluster of the lower remaining vehicles executes the unloading action a t The system cost actually generated;
by training the Q value network to iterate the unloading action, the update strategy is as follows:
wherein Q(s) t ,a t ) Representing the system cost at time t, C e,t Represents the energy consumption cost of the system by adopting edge calculation, alpha is the learning rate, gamma is the discount factor,minimum system cost for the next moment;
meeting Q value infinitely approximates to target Q value Q m (s t ,a t The method comprises the steps of carrying out a first treatment on the surface of the θ) and the corresponding unloading action strategy when the loss function value is the minimum is the optimal distributed calculation unloading strategy;
the Loss function value Loss (θ) is calculated as follows:
Loss(θ)＝E[(R e,t +Q(s t ,a t )-Q m (s t ,a t ；θ))] 2
wherein R is e,t Representing the current cost function, θ is the weight vector.
In the local computing mode, the system cost is the sum of energy consumption and actual task execution time delay; in the edge computing mode, the system cost is the sum of the actual task execution delay of the edge server, the transmission delay, the computing energy of the edge server and the transmission energy.
Wherein the channel gain g is used t Task queue f stored on a terminal vehicle t And its remaining calculated capacity ratio r t To represent the current exploration information from the environment, i.e. s t ＝{g t ,f t ,r t }。
The deep neural network learning algorithm adopts a three-layer DNN architecture, an activation function is a sigmoid function, an output function is a relu function, and an optimal distributed computing unloading strategy is obtained by setting a reward value function.
The invention has the following beneficial effects:
the invention preprocesses the priority of the vehicle cluster through the K-means clustering algorithm, provides more reasonable basis for the calculation resource scheduling process of the MEC server, not only improves the unloading execution success rate of the vehicle terminal calculation task, is more suitable for the vehicle self-organizing network environment, but also improves the execution success rate and the unloading decision accuracy rate of the calculation task, and effectively reduces the execution time delay of the calculation task.
According to the invention, the optimal distributed computing and unloading strategy is obtained through iterative training of the deep neural network learning algorithm, and each vehicle terminal performs data transmission of task unloading through the RSU according to the optimal distributed computing and unloading strategy, so that the probability of successful execution of computing tasks is improved, and the total execution time delay of system tasks is reduced.
Drawings
FIG. 1 is an application scenario diagram of a vehicle edge network task allocation offloading method based on vehicle clustering in an embodiment;
FIG. 2 is a flow chart of a vehicle edge network task allocation offloading method based on vehicle clustering in an embodiment;
fig. 3 is a schematic diagram of state, action, and prize settings of a deep learning algorithm.
Detailed Description
The present invention will be described in further detail with reference to the drawings and examples, in order to make the objects, technical solutions and advantages of the present invention more apparent. It should be understood that the specific embodiments described herein are for purposes of illustration only and are not intended to limit the scope of the invention.
The vehicle edge network task allocation and unloading method based on the vehicle clustering can be applied to the vehicle self-organizing network environment shown in the figure 1. Wherein the MEC server communicates with an RSU (road side unit) via a network; the RSU communicates with vehicle terminals within signal coverage of the RSU over a network. Each vehicle terminal within the RSU signal coverage area is considered a vehicle node. The specific vehicle terminal can offload the calculation tasks carried by the vehicle terminal to an MEC server connected with the RSU through a wireless network for processing, the MEC server obtains task priority through calculation, then determines an offloading strategy of the calculation tasks according to a deep neural network algorithm, and the vehicle terminal then follows the received offloading strategy to offload the calculation tasks acceptable by the MEC server to the MEC server for processing. The MEC server may be implemented as a stand-alone server or as a server cluster formed by a plurality of servers.
As shown in fig. 2, in an embodiment, a vehicle edge network task allocation and offloading method based on vehicle clustering is provided, which mainly includes the following steps:
step S1, an RSU receives position information of a vehicle terminal in a coverage area of the RSU and carried task attribute information in real time, calculates task unloading probability according to the task attribute information, and defines task priority of the vehicle terminal based on the position information and the task unloading probability.
In the scene, the vehicles in the coverage area of the RSU all carry a plurality of calculation tasks, and when the vehicle terminal enters the coverage area of one RSU, the attribute information of the calculation tasks carried by the vehicles is uploaded to the MEC server through the RSU, so that the calculation of the task priority is carried out after a period of time in the MEC task queue is occupied.
Specifically, the information received in real time by the RSU includes position information and task attribute information of the vehicle terminal. Wherein the position information of the vehicle terminal is embodied as a distance from the vehicle terminal to the MEC server; the task attribute information includes task execution delay requirement, task size and actual task execution delay.
The MEC server connected to the RSU will first calculate the vehicle task offloading probabilities from the task attribute information, assuming that the tasks generated are independent, and will be improved and updated at each time stage. The i-th vehicle terminal has a vehicle task offloading probability Pr i When generating each task, the ith vehicle terminal generates the task execution delay requirementIs L i,th Task size m i The actual task execution time delay calculation formula is as follows:
L i,e ＝m i /F i ，(1)
wherein F is i The computing power of the i-th vehicle terminal.
Thus, the vehicle task offloading probability Pr of the i-th vehicle terminal i From probability function Pr (L i,th ＜L i,e ) Obtaining the product. Wherein L is th ＜L e Representing the actual task execution delay L of a task on a vehicle terminal e Exceeding task execution delay requirement L th Requires unloading, probability function Pr (L i,th ＜L i,e ) The ratio of the number of tasks to be offloaded in the tasks carried by the ith vehicle terminal is 1 at the maximum.
The task priority of the ith vehicle terminal will be determined by a tuple p i Definition, represented by a feature vector, including its distance d to the MEC server i And calculates task offloading probability Pr i ，p i ＝＜d i ，Pr i >. The task priority characteristics p of the individual vehicle terminals are thus derived i 。
And S2, dividing the vehicle terminal into H clustering clusters by adopting a K-means user clustering method based on the task priority of the vehicle terminal by the MEC server at the edge side, and obtaining the priority of each clustering cluster.
Firstly, k vehicle terminals are selected as initial clustering centers and k clusters are formed, wherein each vehicle terminal is an initial clustering cluster, and the vehicle terminal is the center of the clustering cluster. And then calculating the distance between the rest vehicle terminals and the initial cluster center by adopting the Euclidean distance method, and distributing the rest vehicle terminals into the cluster closest to the rest vehicle terminals based on the Euclidean distance obtained by calculation. After assigning a vehicle terminal, the centroid c of its cluster will be recalculated h Wherein h is [1, k ]]The centroid is the center of the cluster.
Because the K value is selected to have a great influence on the effect of the K-means clustering method, in order to ensure the accuracy of distribution, the vehicle terminals are divided into K clusters according to the priority of the vehicle terminals by minimizing the error square sum.
The square error and SEE calculation formula is as follows, and the data size is used to indicate the difference between the current cluster number and the optimal cluster number.
Wherein p is i Representing the priority characteristics of the ith vehicle terminal, c h The centroid of the H-th cluster is represented, and H represents the current cluster number (at which time the optimal cluster number is not reached).
Simulation results prove that when the square sum of errors is 1, the current cluster number is the optimal cluster number, the mass center is not repeatedly calculated at the moment, and finally the vehicle terminal is determined to be divided into K cluster groups. The final centroid solution process is as follows:
first, calculate p i Cluster index h of (2) i 。
The calculation formula of the centroid is as follows:
wherein U represents the number of vehicle terminals to be clustered, p i Indicating the priority characteristics of the ith vehicle terminal, D h Representing the cluster h, h.epsilon.1, K]。
The centroid of the cluster reflects the priority, and the smaller the numerical value is, the higher the priority is.
And step S3, the MEC server respectively distributes the cluster clusters with the highest task priorities to the edge computing, and distributes the cluster clusters with the lowest task priorities to the local computing.
In this process, the MEC server may receive task offloading requests from a plurality of Vehicle terminals, and in step S2, the Vehicle terminals are classified into a plurality of clusters according to priority, and in order to be able to complete task processing at the fastest speed, the computing tasks of the vehicles in the clusters with the highest priority are first offloaded to the MEC server for computation, and for the computing tasks of the vehicles in the clusters with the lowest priority, considering the limited computing resources and the instability of V2V links (vehicles-vehicles), must be rejected.
And S4, for the rest users, the MEC server acquires an optimal distributed computing and unloading strategy through iterative training of a deep neural network learning algorithm, and the optimal distributed computing and unloading strategy has the lowest system cost.
The system cost of the system scene in the invention is different under each decision condition, and a single unloading mode is insufficient to minimize the system cost, so that the optimal unloading decision needs to be found through deep learning. It is assumed that each vehicle terminal in the same cluster has the same user priority. For vehicle clusters other than the highest and lowest priority, the MEC server learns the optimal distributed computing offloading policy through the deep neural network.
In the car networking scenario, the algorithm state, action and rewards routine setting of deep learning is shown in fig. 3, in the time slot t where the environment state is denoted as s, the agent selects one unloading action a to interact with the environment, in the next time slot, the environment state is changed to s 'according to the transition probability, and rewards r' are returned to the agent. The agent modifies the next action based on the rewards until an optimal policy is found. In the present invention, s t A is the current environmental state t For unloading motion vector (short "motion" or "unloading motion") in current state, r k Is s t In a state of executing the unloading action a t A prize value is generated. Within the iteration time t, cluster D is clustered h The unloading action of the vehicle terminal can be represented by a t And (3) representing. Currently, the method is thatState s t Then the current exploration information from the environment is represented. The present invention will use the channel gain g t Task queue f stored on vehicle terminal t And its remaining calculated capacity ratio r t Representing the current exploration information from the environment, i.e. s t ＝{g t ,f t ,r t }。
The invention uses a deep neural network learning algorithm and uses a Q value network as a strategy judgment standard, wherein the Q value represents < s t ,a t The value of > i.e. the system cost actually resulting from the remaining vehicle clusters performing the offloading decision in the current environmental state, and the Q value is stored by a Q table and the system cost in each state in the Q table can be looked up. The deep neural network learning algorithm adopts a three-layer DNN architecture, an activation function is a sigmoid function, an output function is a relu function, and an optimal unloading strategy is obtained by setting a reward value function.
The invention adopts a negative rewarding method to make a correct decision to distribute the system cost and accelerate the training progress. At iteration time t, cluster D h The rewards of each vehicle terminal are the system cost of the vehicle terminal. In the local computing mode, the system cost is the sum of the energy consumption and the actual task execution delay. In the edge computing mode, the system cost is the sum of the computing time delay (i.e. the actual task execution time delay) of the edge server, the transmission time delay, the computing energy of the edge server and the transmission energy.
The task unloading algorithm based on the deep neural network is used for obtaining an optimal distributed computing unloading scheme, and a priority classification result can be output after the position information and task attribute information of the current vehicle terminal are obtained, and meanwhile, an unloading strategy is further obtained through training the Q value network.
Wherein the offloading decision action implements the update according to the following formula:
wherein Q(s) t ,a t ) Represented byIs the system cost at t time in the Q table, C e,t Represents the energy consumption cost of the system by adopting edge calculation, alpha is the learning rate, gamma is the discount factor,for the minimum system cost at the next time.
The algorithm solves the decision problem by learning the optimal strategy, and the requirement for obtaining the correct convergence performance is that the Q value is continuously updated, so that the current Q value can be infinitely close to the target Q value Q m (s t ,a t The method comprises the steps of carrying out a first treatment on the surface of the θ), where θ is a weight vector, is continuously updated in each iteration by minimizing the Loss function, and finally the Loss function value Loss (θ) is minimized.
Loss(θ)＝E[(R e,t +Q(s t ,a t )-Q m (s t ,a t ；θ))] 2
Wherein R is e,t Representing the current cost function.
The invention uses an empirical pool to store state-action pairs, < s, a >, and uses empirical playback techniques to sample from the empirical pool and then use it to train with the goal of minimizing the loss function. And after training is finished, selecting an action with the minimum Q value as a calculation task unloading scheme, namely, an unloading scheme of the rest vehicle clusters except the clusters with the highest priority and the lowest priority.
Step S5: and the MEC server sends the strategy to the vehicle clusters in the coverage area of the RSU, so that each vehicle cluster performs data transmission of task unloading through the RSU according to the strategy.
According to the vehicle edge network task allocation and unloading method based on the vehicle clustering, if a new vehicle terminal enters the coverage area of the RSU, the vehicle terminal sends the position information and the carried calculation task attribute information to the MEC server through the RSU, the MEC server calculates the priority of each task, a more reasonable basis is provided for the calculation resource scheduling process of the MEC server, and the unloading execution success rate of the calculation task of the vehicle terminal is improved; after receiving the current calculation task of the vehicle terminal, the MEC server obtains an unloading strategy according to the current obtained vehicle state and generates a rewarding value; and finally, updating the unloading decision action by a task unloading method based on a deep neural network, and carrying out data transmission operation of task unloading by all vehicle terminals in the coverage area of the RSU according to an unloading strategy. Compared with the prior art, the method and the system for preprocessing the priority of the vehicle cluster through the K-means clustering algorithm provide more reasonable basis for the calculation resource scheduling process of the MEC server. The invention not only improves the unloading execution success rate of the calculation task of the vehicle terminal, but also is more suitable for the self-organizing network environment of the vehicle; and the execution success probability of the calculation task and the accuracy of the unloading decision are also improved, and the execution time delay of the calculation task is effectively reduced.
Compared with the prior art, the algorithm provided by the invention under the calculation tasks with different intensities has the lowest cost compared with the traditional unloading algorithm system, has stronger adaptability to occasions of the Internet of vehicles cluster scene, and simultaneously has better robustness and superiority.
It should be understood that, although the steps in the flowchart of fig. 2 are shown in sequence as indicated by the arrows, the steps are not necessarily performed in sequence as indicated by the arrows. The steps are not strictly limited to the order of execution unless explicitly recited herein, and the steps may be executed in other orders. Moreover, at least some of the steps in fig. 2 may include multiple sub-steps or stages that are not necessarily performed at the same time, but may be performed at different times, nor do the order in which the sub-steps or stages are performed necessarily performed in sequence, but may be performed alternately or alternately with at least a portion of the sub-steps or stages of other steps or other steps.
The technical features of the above embodiments may be arbitrarily combined, and all possible combinations of the technical features in the above embodiments are not described for brevity of description, however, as long as there is no contradiction between the combinations of the technical features, they should be considered as the scope of the description.
The above examples illustrate only a few embodiments of the invention, which are described in detail and are not to be construed as limiting the scope of the invention. It should be noted that it will be apparent to those skilled in the art that several variations and modifications can be made without departing from the spirit of the invention, which are all within the scope of the invention. Accordingly, the scope of protection of the present invention is to be determined by the appended claims.
Claims (7)
1. The vehicle edge network task allocation and unloading method based on the vehicle clustering is characterized by comprising the following steps of:
step S1: receiving the position information of a vehicle terminal in the coverage area of an RSU and task attribute information carried by the vehicle terminal in real time, calculating task unloading probability according to the task attribute information, and defining task priority of the vehicle terminal based on the position information and the task unloading probability;
step S2: dividing the vehicle terminal into H clustering clusters by adopting a K-means user clustering method based on the task priority of the vehicle terminal, and obtaining the task priority of each clustering cluster;
step S3: the cluster with the highest task priority is distributed to edge calculation, and the cluster with the lowest task priority is distributed to local calculation;
step S4: for other cluster groups between the highest and lowest task priorities, obtaining an optimal distributed computing and unloading strategy through iterative training of a deep neural network learning algorithm, wherein the optimal distributed computing and unloading strategy has the lowest system cost;
step S5: and sending the optimal distributed computing unloading strategy to each cluster in the coverage area of the RSU, so that the vehicle terminals in the cluster clusters perform data transmission of task unloading through the RSU according to the optimal distributed computing unloading strategy.
2. The vehicle edge network task allocation offloading method according to claim 1, wherein in the step S1, the location information of the vehicle terminal is a distance from the vehicle terminal to the MEC server; the task attribute information comprises a task execution time delay requirement, a task size and an actual task execution time delay;
the task priority feature of the vehicle terminal is composed of a tuple p i Definition, p i ＝＜d i ，Pr i >, wherein d i Representing the distance, pr, of the ith vehicle terminal to the MEC server i Representing a task offloading probability of an ith vehicle terminal;
task offloading probability Pr of the ith vehicle terminal i From probability function Pr (L i,th ＜L i,e ) Obtaining, wherein L th ＜L e Representing actual task execution time delay L e Exceeding task execution delay requirement L th Requires unloading, probability function Pr (L i,th ＜L i,e ) The ratio of the number of tasks to be offloaded in the tasks carried by the ith vehicle terminal is 1 at the maximum;
the actual task execution time delay calculation formula is as follows:
L i,e ＝m i /F i ；
wherein m is i Task size for ith vehicle terminal, F i The computing power of the i-th vehicle terminal.
3. The vehicle edge network task allocation and offloading method based on vehicle clustering as set forth in claim 1, wherein said step S2 specifically includes:
s21: k vehicle terminals are selected as initial clustering centers and k clustering clusters are formed;
s22: the Euclidean distance between the rest of the vehicle terminals and each cluster center is calculated, and the rest of the vehicle terminals are distributed to the cluster closest to the rest of the vehicle terminals based on the Euclidean distance;
s23: re-computing the centroid of the cluster after each allocation;
s24: returning to the step S22, and finally determining to divide the vehicle terminal into K clustering clusters until the error square sum SEE is 1; the centroid of the cluster reflects the task priority of the cluster, and the smaller the value of the centroid is, the higher the task priority is;
the calculation formula of the error square sum SEE is as follows:
wherein p is i Representing task priority characteristics of the ith vehicle terminal, c h The centroid of the H clustering cluster is represented, H represents the current clustering quantity, and the optimal clustering quantity is not reached at the moment;
the centroid calculation formula is as follows:
wherein,
wherein D is h Representing clusters, i.e. the h cluster, h.epsilon.1, K]U represents the number of vehicle terminals to be clustered, h i And (5) indicating the cluster index of the ith vehicle terminal.
4. The vehicle edge network task allocation and offloading method based on vehicle clustering as set forth in claim 1, wherein said step S4 specifically includes:
during the iteration time t, the current environmental state s t Representing current exploration information from the environment, a t Representing cluster clusters D h Unloading action of medium-terminal vehicle and clustering cluster D h The rewards of the terminal vehicles are the system components of the terminal vehiclesThe cost is high;
q value is expressed as < s using a network of Q values as a policy criterion t ,a t The value of > i.e. at the current environmental state s t The cluster of the lower remaining vehicles executes the unloading action a t The system cost actually generated;
by training the Q value network to iterate the unloading action, the update strategy is as follows:
wherein Q(s) t ,a t ) Representing the system cost at time t, C e,t Represents the energy consumption cost of the system by adopting edge calculation, alpha is the learning rate, gamma is the discount factor,minimum system cost for the next moment;
meeting Q value infinitely approximates to target Q value Q m (s t ,a t The method comprises the steps of carrying out a first treatment on the surface of the θ) and the corresponding unloading action strategy when the loss function value is the minimum is the optimal distributed calculation unloading strategy;
the Loss function value Loss (θ) is calculated as follows:
Loss(θ)＝E[(R e,t +Q(s t ,a t )-Q m (s t ,a t ；θ))] 2
wherein R is e,t Representing the current cost function, θ is the weight vector.
5. The vehicle edge network task allocation offload method based on vehicle clustering according to claim 4, wherein in the local calculation mode, the system cost is a sum of energy consumption and actual task execution delay; in the edge computing mode, the system cost is the sum of the actual task execution delay of the edge server, the transmission delay, the computing energy of the edge server and the transmission energy.
6. The vehicle edge network task allocation offloading method of claim 4, wherein a channel gain g is used t Task queue f stored on a terminal vehicle t And its remaining calculated capacity ratio r t Representing the current exploration information from the environment, i.e. s t ＝{g t ,f t ,r t }。
7. The vehicle edge network task allocation and offloading method based on vehicle clustering according to claim 4, wherein the deep neural network learning algorithm adopts a three-layer DNN architecture, the activation function is a sigmoid function, the output function is a relu function, and the optimum distributed computing and offloading policy is obtained by setting a reward value function.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210888963.4A CN115065683B (en) | 2022-07-27 | 2022-07-27 | Vehicle edge network task allocation and unloading method based on vehicle clustering |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210888963.4A CN115065683B (en) | 2022-07-27 | 2022-07-27 | Vehicle edge network task allocation and unloading method based on vehicle clustering |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN115065683A CN115065683A (en) | 2022-09-16 |
| CN115065683B true CN115065683B (en) | 2023-12-26 |
Family
ID=83206760
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210888963.4A Active CN115065683B (en) | 2022-07-27 | 2022-07-27 | Vehicle edge network task allocation and unloading method based on vehicle clustering |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115065683B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115964178B (en) * | 2023-01-09 | 2024-05-31 | 江南大学 | Internet of vehicles user computing task scheduling method, device and edge service network |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109872001A (en) * | 2019-02-28 | 2019-06-11 | 南京邮电大学 | Unmanned vehicle method for allocating tasks based on K-means and discrete particle cluster algorithm |
| CN113467851A (en) * | 2021-05-24 | 2021-10-01 | 南京邮电大学 | Dynamic vehicle calculation task unloading method and device based on vehicle clustering |
| CN113518326A (en) * | 2021-04-12 | 2021-10-19 | 南京邮电大学 | Joint distribution optimization method for vehicle-mounted edge network computing resources and communication resources |
| CN113645273A (en) * | 2021-07-06 | 2021-11-12 | 南京邮电大学 | Internet of vehicles task unloading method based on service priority |
| CN113661721A (en) * | 2019-05-07 | 2021-11-16 | 英特尔公司 | V2X service for providing trip specific QoS prediction |
| CN113918240A (en) * | 2021-10-15 | 2022-01-11 | 全球能源互联网研究院有限公司 | Task unloading method and device |
| CN114549886A (en) * | 2022-03-04 | 2022-05-27 | 重庆邮电大学 | V2X message clustering method and system based on k-means algorithm |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US11151682B2 (en) * | 2019-07-22 | 2021-10-19 | Verizon Patent And Licensing Inc. | System and methods for distributed GPU using multi-access edge compute services |
-
2022
- 2022-07-27 CN CN202210888963.4A patent/CN115065683B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109872001A (en) * | 2019-02-28 | 2019-06-11 | 南京邮电大学 | Unmanned vehicle method for allocating tasks based on K-means and discrete particle cluster algorithm |
| CN113661721A (en) * | 2019-05-07 | 2021-11-16 | 英特尔公司 | V2X service for providing trip specific QoS prediction |
| CN113518326A (en) * | 2021-04-12 | 2021-10-19 | 南京邮电大学 | Joint distribution optimization method for vehicle-mounted edge network computing resources and communication resources |
| CN113467851A (en) * | 2021-05-24 | 2021-10-01 | 南京邮电大学 | Dynamic vehicle calculation task unloading method and device based on vehicle clustering |
| CN113645273A (en) * | 2021-07-06 | 2021-11-12 | 南京邮电大学 | Internet of vehicles task unloading method based on service priority |
| CN113918240A (en) * | 2021-10-15 | 2022-01-11 | 全球能源互联网研究院有限公司 | Task unloading method and device |
| CN114549886A (en) * | 2022-03-04 | 2022-05-27 | 重庆邮电大学 | V2X message clustering method and system based on k-means algorithm |
Also Published As
| Publication number | Publication date |
|---|---|
| CN115065683A (en) | 2022-09-16 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Yang et al. | A parallel intelligence-driven resource scheduling scheme for digital twins-based intelligent vehicular systems | |
| CN111586696A (en) | Resource allocation and unloading decision method based on multi-agent architecture reinforcement learning | |
| CN113645273B (en) | Internet of vehicles task unloading method based on service priority | |
| CN112995289B (en) | Internet of vehicles multi-target computing task unloading scheduling method based on non-dominated sorting genetic strategy | |
| CN112911016B (en) | Edge-side collaborative computing unloading method and system, electronic equipment and storage medium | |
| CN110531996B (en) | Particle swarm optimization-based computing task unloading method in multi-micro cloud environment | |
| CN112153145A (en) | Method and device for unloading calculation tasks facing Internet of vehicles in 5G edge environment | |
| CN114189892A (en) | Cloud-edge collaborative Internet of things system resource allocation method based on block chain and collective reinforcement learning | |
| CN115297171B (en) | Edge computing and unloading method and system for hierarchical decision of cellular Internet of vehicles | |
| CN113316116B (en) | Vehicle calculation task unloading method | |
| CN112214301B (en) | Smart city-oriented dynamic calculation migration method and device based on user preference | |
| CN113641417B (en) | Vehicle security task unloading method based on branch-and-bound method | |
| CN115065683B (en) | Vehicle edge network task allocation and unloading method based on vehicle clustering | |
| CN116489708A (en) | Meta universe oriented cloud edge end collaborative mobile edge computing task unloading method | |
| Li et al. | DNN partition and offloading strategy with improved particle swarm genetic algorithm in VEC | |
| CN113709249B (en) | Safe balanced unloading method and system for driving assisting service | |
| CN116744367A (en) | Unloading method based on double-layer unloading mechanism and multi-agent algorithm under Internet of vehicles | |
| Maleki et al. | Reinforcement learning-based decision-making for vehicular edge computing | |
| CN117221951A (en) | Task unloading method based on deep reinforcement learning in vehicle-mounted edge environment | |
| CN113452625B (en) | Deep reinforcement learning-based unloading scheduling and resource allocation method | |
| CN115967430A (en) | Cost-optimal air-ground network task unloading method based on deep reinforcement learning | |
| CN114741191A (en) | Multi-resource allocation method for compute-intensive task relevance | |
| CN114138466A (en) | Task cooperative processing method and device for intelligent highway and storage medium | |
| Farimani et al. | Deadline-aware task offloading in vehicular networks using deep reinforcement learning | |
| CN116155750B (en) | Deep learning job resource placement method, system, equipment and storage medium |
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 | ||
| CB03 | Change of inventor or designer information |
Inventor after: Wu Song Inventor after: Ye Jian Inventor after: Zhao Haitao Inventor after: Feng Tianyi Inventor after: Tang Jiawen Inventor after: Yang Hongwei Inventor after: Chen Zhao Inventor after: Dong Xingxing Inventor before: Ye Jian Inventor before: Wu Song Inventor before: Zhao Haitao Inventor before: Feng Tianyi Inventor before: Tang Jiawen Inventor before: Yang Hongwei Inventor before: Chen Zhao Inventor before: Dong Xingxing |
|
| CB03 | Change of inventor or designer information | ||
| TA01 | Transfer of patent application right |
Effective date of registration: 20231010 Address after: 211112 Building 1, No. 1555, Tianyin Avenue, Jiangning District, Nanjing City, Jiangsu Province Applicant after: Duolun Internet Technology Co.,Ltd. Address before: 211112 No. 1555 Tianyin Avenue, Jiangning District, Nanjing City, Jiangsu Province Applicant before: DUOLUN TECHNOLOGY Corp.,Ltd. |
|
| TA01 | Transfer of patent application right | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |