LU500343B1 - Self-Adaptive Energy-Optimal Vehicle Clustering Method based on fuzzy C-means - Google Patents

Abstract

Disclosed is a self-adaptive energy-optimal vehicle clustering method based on fuzzy C-means, which includes the following steps: connecting a vehicle to a Vehicular Ad Hoc Network (VANET); determining a neighbor list of the vehicle through angular neighbor detection for the vehicle; determining the number of clusters after vehicle clustering according to a self-adaptive energy-optimal model; grouping the vehicles into clusters by means of a fuzzy C-means algorithm; and selecting a cluster-head vehicle by using a vehicle movement direction, a weighted mobility value, and entropy as measures of a weighted clustering algorithm, where optimization of an election process of the cluster head can obtain a stable network. The method of the present invention improves the stability of the VANET, equalizes the power consumption, and reduces network overheads.

Description

Self-Adaptive Energy-Optimal Vehicle Clustering Method based on fuzzy C-means 999943

BACKGROUND OF THE INVENTION Field of the Invention The present invention relates to a self-adaptive energy-optimal vehicle clustering method based on fuzzy C-means, which relates to the field of Internet of Vehicles (IoV) communications technologies.

Description of Related Art In recent years, research on a Vehicular Ad Hoc Network (VANET) has attracted more and more attention from the academia and industry, and has been studied together with some popular technologies such as data security and integrity of a vehicular cloud computing network.

The VANET is formed by connecting vehicles via an ad hoc network, and consists of a road side unit (RSU) and an on board unit (OBU), where the OBU is mounted in the vehicle and may be regarded as a mobile node. The basic form of vehicle communication is classified into Vehicle to Vehicle (V2V) and Vehicle to Infrastructure (V2I). The VANET is an extension of a Mobile Ad-hoc Network (MANET) in the IoV scenario. However, because of high-speed mobility of vehicle nodes, the conventional network topology of the MANET is incompatible with the VANET and thus direct use of the topology is infeasible. In addition, data forwarding in the VANET requires low-latency transmission and high reliability. However, due to a large number of vehicle nodes and a few router deployments, the use of a flat network architecture of the conventional ad hoc network may cause poor scalability of the network. Therefore, academic scholars have proposed to build a clustering structure in the VANET to manage the network topology, so as to reduce changes in the VANET topology. Moreover, as the mobile vehicle nodes store more and more user information (for example, vehicle positions and movement trajectories), related security and privacy issues also need urgent attention. It is also a major challenge to ensure secure and reliable transmission of the information in the VANET.

Currently, a massive amount of data in the IoV environment is processed through conventional cloud computing. Although the cloud processing has a very strong storage and computing capability, various services in the IoV environment have high requirements on latency, thus failing to ensure high service quality. Therefore, it is required to introduce Mobite00543 Edge Computing (MEC) technology into the IoV, to "sink" latency-sensitive services to be close to users for processing. An MEC-based IoV architecture can meet the processing requirements of ultra-high reliability and low latency for various latency-sensitive services, and further can optimize network resources and improve the security of user privacy. In addition, compared to the conventional cloud computing architecture, message transmission or task processing in the edge cloud close to user equipment (UE) facilitates protection of private information of the users. Ideally, edge servers should be deployed as many as possible to reduce the latency of service requests, which, however, inevitably results in significant energy consumption. This case not only causes high production costs, but also is contrary to the concept of green communication. Therefore, energy consumption in the MEC technology is a huge challenge.

Therefore, in combination with the characteristics of the VANET, designing an efficient clustering topology control algorithm in the MEC-based IoV environment to optimize network connectivity and stability 1s of great significance to VANET protocol development and network management. Moreover, in the IoV edge network environment, the computing capability of the vehicles can be fully utilized, which may be regarded as auxiliary edge servers to provide services to surrounding neighboring vehicles or pedestrians. Thus, optimization of energy consumption of the vehicular edge servers needs to be further carried out in vehicle clustering.

SUMMARY OF THE INVENTION Technical Problem To solve the deficiencies in the prior art, the present invention provides a self-adaptive energy-optimal vehicle clustering method based on fuzzy C-means, so as to solve the high energy consumption problem in the MEC technology in the prior art.

Technical Solution To achieve the foregoing objective, the present invention adopts the following technical solution: A self-adaptive energy-optimal vehicle clustering method based on fuzzy C-means is provided, including the following steps:

S1: connecting a vehicle to a VANET; LU500343 S2: determining a neighbor list of the vehicle through angular neighbor detection for the vehicle; S3: determining the number of clusters after vehicle clustering according to a self-adaptive energy-optimal model, S4: grouping the vehicles into clusters by means of a fuzzy C-means algorithm; and SS: selecting a cluster-head vehicle by using a vehicle movement direction, a weighted mobility value, and entropy as measures of a weighted clustering algorithm, where optimization of an election process of the cluster head obtains a stable network.

Further, step S2 specifically includes: by using an angle between vehicle speed vectors as a measure, and when and only when the included angle between the vehicle speed vectors is an acute angle, clustering the two vehicles into one cluster; after the vehicle receives hello data broadcast from its surrounding vehicles, acquiring position information by using a vehicular GPS, considering a vehicle whose angle with the vehicle is within a threshold range as a potential neighboring vehicle, and ignoring hello data broadcast from other surrounding vehicles.

Further, step S3 specifically includes: dynamically adjusting the allocation of CPU resources occupied by virtual machines (VMs) based on popularities of virtual network function (VNF) units, to obtain an energy-optimal model; and obtaining an optimal value Æ according to the energy-optimal model.

Further, in step S4, one cluster includes one cluster head and a plurality of cluster members (namely, vehicle nodes); and each vehicle is only allocated to one cluster, ensuring that the clusters obtained after grouping do not overlap each other.

Further, step S4 specifically includes: grouping the vehicles into clusters by means of a fuzzy C-means algorithm according to vehicle modeling parameters, where the parameters include a vehicle speed, distance, and angle.

Advantageous Effect The present invention has the following beneficial effects: The method of the present invention has good performance with the evaluation indicators of both the power consumption and a cluster life cycle, which can effectively reduce the whole power consumption of vehicular edge servers and network overheads, and further improve robustness and stability of the VANET.

BRIEF DESCRIPTION OF THE DRAWINGS 500345 FIG. 1 is a schematic flowchart of a method of the present invention; and FIG. 2 is a schematic diagram of a two-dimensional image for calculating a vehicle movement angle in the method of the present invention.

DETAILED DESCRIPTION OF THE INVENTION For ease of understanding by persons skilled in the art, the present invention is further described below with reference to the embodiments and accompanying drawings, but the content mentioned in the implementations is not intended to limit the present invention.

Referring to FIG. 1, a self-adaptive energy-optimal vehicle clustering method based on fuzzy C-means of the present invention is applied in a VANET, which determines a neighbor list of a vehicle by calculating an angle between vehicles, to prevent incorrect clustering of vehicles on a crossroad. The vehicles are effectively grouped into clusters by means of a fuzzy C-means clustering algorithm, where a cluster value after clustering based on fuzzy C-means is pre-determined according to a self-adaptive energy-optimal model. Further, the present invention proposes that a cluster head is selected by using a vehicle movement direction, a weighted mobility value, and an entropy value as indicators of a weighted clustering algorithm, which prolongs a life cycle of the cluster head and reduces a re-clustering frequency, thus improving the stability of the VANET and reducing network overheads. Specific steps of the method are as follows: S1: A collaborative relationship is established between heterogeneous communication networks by means of existing IoV technology, so as to provide reliable Internet services to a vehicle and connect the vehicle to the VANET.

S2: In order to ensure stability of clusters, a neighbor list of the vehicle is selected necessarily through angular neighbor detection, rather than discovering neighboring vehicles by exchanging Hello data packets in the whole communication range.

Specifically, an angle between vehicle speed vectors is used as a measure, and when and only when the included angle between the vehicle speed vectors is an acute angle, two vehicles are clustered into one cluster. Because the vehicles are all fitted with a GPS, position information of vehicles vi and w at time points 7 and 7-1 can be obtained. As shown in FIG. 2, at the time point f, a position vector of the vehicle v1 is (x1, y1) and a position vector of the vehicle v2 is (x2,

V2); and at the time point 7-1, a position vector of the vehicle vi is (x1', yı') and a position vedigp00343 of the vehicle ” is (x2', V2").

Then, an included angle 0 between the vehicles v1 and v2 may be obtained based on the law of cosines: 6 = arccos a * Ac + An * Ay, | After the vehicle vi receives hello data from its surrounding vehicles, only a vehicle whose angle with the vehicle vi is within the threshold range is considered as a potential neighboring vehicle, and the hello data broadcast from other surrounding vehicles is ignored. That is, the threshold range is O<p. When the included angle 0 exceeds 18°, it is regarded that the vehicles travel on different roads, and thus the threshold ¢ is set to 18°. When the included angle 0 is within the threshold range, the two vehicles are identified as travelling in the same direction. When the included angle exceeds the threshold, it is regarded that the vehicles travel in different directions.

S3: The number of clusters after vehicle clustering is determined according to a self- adaptive energy-optimal model.

Specifically, the allocation of CPU resources occupied by VMs is dynamically adjusted based on popularities of virtual network function (VNF) units. According to a difference in popularity of request flows reaching MEC servers, by means of replacing equal CPU allocation, the computing resources of each VM in the vehicle are dynamically adjusted by adapting to the popularities of different types of requests, thus minimizing the power of vehicular server resources.

The present invention groups edge servers into clusters based on Service Function Chaining (SFC), and there are in total m edge servers which are group into Æ clusters that do not overlap each other. The servers in each cluster are classified into two types: local cluster head servers (LC) and member servers (Member); and each flock has only one cluster head. There are À LC edge servers and (m/k-1) member edge servers. Based on an edge server clustering model set forth in the present invention, it is assumed that at least one LC server and at least one member server exist in each cluster, namely, 1<k<m/2. When the edge servers are grouped into only one cluster, there may be mc-n idle VMs; and when the cluster number is greater than 1, there may be mec-(k-1)c-n idle VMs. When k>(m+1-n/c), (k-1)c VMs are occupied by an LVNF (a cluster head virtual network function unit). Therefore, in this case, Nu MVNFs (member virtual network function units) are not installed, where Nu =max[n-(m-k+1)c, 0].

For each flock, a member edge server node can store different types of VNF units, and 1200343 LC server has VNF distribution information of the flock.

With regard to a connection request from a UE source node to a destination node, a data flow controller is connected to an LC node by using a signaling packet including a source ID, a destination ID, and VNF required by a head portion.

During creation of a VNF service chain, each data flow is started by the LC node.

Each data flow needs to provide information about the {source ID, destination ID, and required VNF } in the header of a data packet of the data flow.

Based on a field of the required VNF, the LC node determines a route needed in the cluster to direct flows to the VNF in the link.

If the required VNF does not exist in the flock, it is migrated from another nearest flock which has the required VNF, and then an MC (member server) node creates a VNF chain.

Within a time interval of 7(Æ), the number of flows received by edge server virtual machines VESs is ny, and each flow contains n,er requests.

Therefore, there are in total ne requests from the edge servers within the time interval of 7(k). An average number of requests from a single RH, LC edge server virtual machine LVE within the time interval of 7(k) is : i = FE? J Then, the probability of the jth LVE being busy per second is Pre; ? T{ k) k * T(Kk} ÆT{k) Flr k where f denotes a processing time by the jth LVE when a request reaches this LVE, and rez; and m; denotes a proportion of time dedicated to processing task request flows by the jth LVE.

C= Pug 2 = 2M, Therefore, a CPU load of a single LVE per second is 7 k When a request reaches the LVE, it is assumed that the CPU operates at full power.

That is, the VM fully processes the request, and 1; #1 If CPU resources are evenly distributed to > £2; = # each single VM in the edge server, 7; is a constant value, / , and a CPU load of the ; à AP; RAR LC edge server is hi 2! 2 k k Based on Network Function Virtualized (NFV) technology, one edge server has a plurality of VMs and one VM is corresponding to one VNF unit.

The types and number of the VNF units in the LC edge server are fixed, while the VNF units running in the member server are replaced according to popularities.

That is, VM migration occurs.

When all the VMs in the edge serydp00343 are occupied, a VNF unit with the lowest popularity may be replaced.

Therefore, an average CPU load /irember of the member servers is classified into /-ep (an average CPU load required for replacing VNF in the member servers) and /w (an average CPU load required for running VNF in the edge servers). Therefore, the total power consumption is Be (k) = kP (Le }+ (m k) Pl + Lop ) (1) CPU load per second during replacement of MVEs (virtual machines of member edge servers) When a request flow reaches a flock, a VNF unit required by a task request has not yet been installed.

In this case, a VNF unit with the lowest popularity in the flock may be replaced with a VNF unit required by the task flow.

Such dynamic replacement occupies a CPU resource in the server, producing extra power consumption.

That is, in the case of [mc-(k-1)c] < n, it is required to replace a low-popularity VE that has not yet been installed.

Because MVEs that are not installed are dynamically replaced, an interval range of a sum pu of the popularities of the m, MVEs that are not installed needs to be restricted, to obtain a value of pv.

Because the MVE with the lowest popularity is uninstalled at the initial time, a 2 Pi 7 f N ’ WN, =i ; EU Fly 2e, t minimum critical value of puis /““ where Na == ty + Ze . When the AVE ; ; AT request flow requires MVE 7 to process the task, where ja «4, MVEn-nu having the lowest popularity is replaced with MVE" If another request flow requires MVE 7 to „N - AT ot 2‘ nrg : py OLA process the task, where / © No and J #7 ‚ MVEZ is further replaced with MVE, 2 P; Therefore, a maximum critical value of py is FEN , where Na ={n—ngn-n, +1: 1} > P, Sp SY P, Therefore, an interval range of puis =" Fete Because a difference value between /“Na and 4“ Na is Pan, "Pr which is x negligible, the value of pu may be estimated as /“N« where Ny is a collection of MVESs that are not installed, and Nu = {n-nu+1, n-nu+2, ..., n}. Then, the CPU load during replacement of MVEs per second is

Cs ; A LU500343 / — al tk) drop a ort, — on Peg BE; Lan TP (m-k)T{k} m-k m—k (2) CPU load per second when running requests in the member edge servers When [me ~(k~1) c] = ” few servers process centralized requests, while most remaining servers are idle.

Therefore, it is not required to migrate the VMs.

To calculate a CPU load per second when running requests in the member edge servers, it is required to determine an average popularity of the member servers, which is equal to a sum of the popularities of the MVEs installed in the member servers divided by the number of the member servers.

The sum of the popularities of all the MVEs installed in the member servers is 1-P;-Pu, and then a total pis of p,= 1-P,—R, average popularities of the MVESs installed in each member server is m=k When a request reaches the MVE, a processing time of the MVE is #, where po n, T{k) Ji >, Ve . . . Pres.

Then, the probability Pine, of the MVEs of the vth kind in the ith cluster being busy per second is: ey By Teg Pj 1,7 (k) —F Pre, Fe fr a a SD à Hiv T'(k) oH T'(k) Frey come Pell 2 mkt, veil2-et; where tr } ca C3 $ and when a request reaches the MVE and the VM fully processes the request (namely, when the CPU operates at full power), 1, =1, I / =P ne M, rn Therefore, a CPU load of a single MVE per second is ”* Dame © Pili . Because the MVEs in the member server do not interfere with each other during operation, a CPU load per second when all the MVEs in the member servers operate is: ly = 2 Ay 1.5, veil 2, - €, 2 2 Pin RI Pu Pu je! ee kt vel1.2…-e! . Because ‘Sid mkiveliod , the CPU load per second when running requests in the member server is: ly = > Hells, Py, veld 2 3

In a system model, calculation of power consumption of the MEC servers mainly inclulig900343 two parts: calculation of power consumption of the LC edge servers and calculation of power consumption of the member servers. A total power consumption is: Pince (k ) = kP(I © ) + (m -k ) PL tomper =kP (lc )+(m=k)P(1,; + Le) Therefore, the self-adaptive energy-optimal model of a server may be converted to a convex optimization problem to be resolved: min Be. (4) = kP(1; 0) + m—=K)P(1y +1.) st. T (k) ST Isksm/2 Because a total energy consumption of the vehicular edge servers in the whole IoV edge network environment is reduced, a task processing time is increased. However, the present invention ensures that an average latency T(k) of processing task request flows in the vehicular MEC servers maintains at an acceptable level, that is, 7(X)<7;eg. Further, in order to ensure that at least one LC server and at least one member server exist in each cluster, a constraint condition of 1<k <m/2 needs to be further met. — — JE FE fe = PT A CPU load formula of the LC edge servers is j A" To minimize the c CC Mn . N min/,e = ZU; = >, (PE CPU load of the LC edge servers, = aR is required, which is min 4 Len, converted to = to be calculated. Therefore, a minimum value of ‚#1 is calculated according to the Cauchy inequality in the present invention, namely: min > pn ‚= min (Di + Pal +" PA. )

A ; 1 1 1 2 (pa + Pan Hp werd 2(Jp +p. +p +p.) Because non fe , | 1,1. 1 | c — HA Fee = mo M Tis obtained according to a CPU resource model of the server. Based on

$ pme (Jp Jp, + Jp; +.) Dil; <= the Cauchy inequality theorem, +“ can be obtained, ni=p.n’ =pn° 1. and equality holds if and only if PIL "Pl PAL That is, a minimized CPU load of the vehicular LC edge servers is: 2 | wp, NP +p +p) min/, =min > a ja k ke Likewise, the CPU load when the member servers operate is: ; na Je, + Jp, + Jp; +p) TR 5 ree (m-k}e Therefore, a model of minimizing power consumption of the vehicular MEC servers adapting to different request flow popularities 1s as follows: min Pince (k) = kP{ Le ) + (m 7 &)P {romper } { 2 inlip +d. +p + Jp. =kP EP Bar, ep) | c : ° ‘ f (2° / ; 2 iP (ye + fra + fps +p) (m IP PCR Dach Be) HC ME a | ke 1 (m=k}e \ } \ st. T{k)ST,,1<k<m/2 0d +1, <1 An optimal value k, namely, an optimal cluster number Æ for minimizing the overall energy consumption of the vehicle servers, is obtained according to this energy-optimal model. In order to determine the optimal clustering value k, it is required to ensure that the Pmec(Æ) function is a convex function, so as to effectively calculate an optimal number of clusters to minimize the power consumption of the MEC servers. Therefore, such an optimization issue needs to be tackled in two cases.

Under the condition that the convexity of the power model is met, the optimal cluster number is determined by means of convex optimization. When the model does not conform to the convexity condition, the optimal cluster number k is found out by means of a traversal search algorithm.

S4: The vehicles are grouped into clusters by means of a fuzzy C-means algorithm according to parameters such as a vehicle speed, distance, and angle.

Specifically, a data set X= {x1, x2, ..., xn} having attribute values of the vehicles vi, V», LU500343 Vn 1s grouped into c € {2, ..., n} clusters, where the data set X has three attributes of the vehicle speed, distance, and angle. The cluster grouping status may be represented by a membership matrix U of kxn, where the value of u indicates a degree of membership of the vehicle v; to a cluster j, and u; € (0, 1). When #;=0, it indicates that the vehicle v; does not belong to the cluster J, and when w;=1, it indicates that the vehicle v; is allocated to the cluster j. The greater the value of uj; indicates a higher degree of membership of the vehicle v; to the cluster j, and u; meets the following formula:

C C Du =1 1; >0 i=l JA i=12- GG J=1,2, A mean vector matrix of the clusters is defined as B= {u1, 2, 3, ..., Hc}, and a distance between the vehicle data set X and the mean vector matrix B is shown by a matrix D. The distance matrix D is calculated as follows by using the Euclidean distance ; ; T dy = yl X 746 Xe, 7 55) dy, tr d, D=| : +. : dy + da Then, a target function is: H CC 2 rn CC 2 — 77 TH — m _ Jude = Da” TELA i=l j=l TE = where m is a fuzzy coefficient which is usually equal to 2. By means of Lagrangian optimization, the Lagrangian multiplier À is introduced, and the target function is written into: Fl C € T= urd 2 | Yu, |=1 H i i i=l j=l i=l à oJ —=0,—=0 2; © ou, .

Partial derivatives are let to be / by using the method of Lagrange multipliers, to deduce updating formulas of the degree of membership and the clustering center mean vector as follows:

Uy / > | #2 #1 w=, [Sul jt / J=l | SS: It is proposed that a cluster-head vehicle is selected by using a vehicle movement direction, a weighted mobility value, and entropy as measures of a weighted clustering algorithm, where optimization of an election process of the cluster head can obtain a more stable network.

Specifically, after the vehicles are clustered by means of the fuzzy C-means algorithm, weighted sums of the vehicles v; in the cluster are calculated based on a WCA algorithm, and a vehicle node with the minimum weighted sum 1s selected as the cluster head.

Characteristic parameters considered in the WCA are replaced with the vehicle position information Le | the weighted mobility value So (1) , and the entropy value H, (£ Af):

1. The vehicle position information is Mo = [MM - {| which indicates whether the vehicles are far away from each other or close to each other.

2. The weighted mobility value Sy (7) of the vehicle measures the relative stability of the vehicle to its neighboring vehicles.

3. The entropy value H, (f, Ar) of the vehicle reflects whether a network with a vehicle v as the center is stable. Then, the weighted sum of the vehicle v is calculated as follows: W.(t)=w (7) + SE) (4, (t.A1 )) ; 3 | > ow, =], w >0 where wi, wa, W3 are corresponding weighting factors, and #4 ; and values of the weighting factors may be set according to a degree of importance of corresponding parameters.

By selecting the cluster head with the vehicle movement direction, the weighted mobility value, and the entropy value as indicators of a weighted clustering algorithm, a life cycle of the cluster head can be prolonged and a re-clustering frequency can be reduced, thus improving the stability of the VANET and reducing network overheads.

Although the embodiments of the present invention are described above, the present invention is not limited to the specific embodiments and application fields described above.

FH@00543 above specific embodiments are merely schematic and instructive, rather than restrictive.

Enlightened by this specification, those of ordinary skill in the art can also make various forms without departing from the protection scope of the claims of the present invention, which all belong to the protection scope of the present invention.

Claims (10)

CLAIMS LU500343

1. A self-adaptive energy-optimal vehicle clustering method based on fuzzy C-means, comprising the following steps: S1: establishing a collaborative relationship between heterogeneous communication networks by means of an Internet of Vehicles (IoV) technology, so as to connect a vehicle to a Vehicular Ad Hoc Network (VANET); S2: determining a neighbor list of the vehicle according to an angle between speed vectors of the vehicle and another vehicle in the VANET, wherein the neighbor list of the vehicle comprises neighboring vehicles of the vehicle; and clustering the vehicle and its neighboring vehicles into one cluster; S3: determining a self-adaptive energy-optimal model according to a CPU load per second during replacement of virtual machines (VMs) of member edge servers and a CPU load per second when running requests in the member edge servers, and determining the number of clusters after vehicle clustering by using the self-adaptive energy-optimal model, S4: grouping the vehicles into clusters by means of a fuzzy C-means algorithm according to vehicle modeling parameters, wherein the vehicle modeling parameters comprise a vehicle speed, distance, and angle; and SS: selecting a cluster-head vehicle by using a vehicle movement direction, a weighted mobility value, and entropy as measures of a weighted clustering algorithm.

2. The self-adaptive energy-optimal vehicle clustering method based on fuzzy C-means according to claim 1, wherein step S2 specifically comprises: when the angle between speed vectors of the vehicle and another vehicle is an acute angle, clustering the two vehicles into one cluster; after the vehicle receives hello data broadcast from its surrounding vehicles, acquiring position information by using a vehicular GPS, determining a vehicle whose speed vector angle with the vehicle is within a threshold range as a potential neighboring vehicle and ignoring hello data broadcast from those which are not the potential neighboring vehicles.

3. The self-adaptive energy-optimal vehicle clustering method based on fuzzy C-means according to claim 2, wherein the determining a vehicle whose speed vector angle with the vehicle is within a threshold range as the potential neighboring vehicle comprises: atthe time point #, defining a position vector of a vehicle vı as (x1, y1) and a position vector of a vehicle v as (x2, y); and at the time point 7-1, defining a position vector of the vehicle vı as 599349 yı') and a position vector of the vehicle v; as (x2', 32"); determining an included angle 0 between speed vectors of the vehicles vi and ” by using the following formula: \ 8 = arccos | A EA ANA | after the vehicle vi receives hello data from its surrounding vehicles, determining the vehicle v, as the potential neighboring vehicle of the vehicle v1 when 0 < ¢, wherein ¢ is a threshold.

4. The self-adaptive energy-optimal vehicle clustering method based on fuzzy C-means according to claim 1, wherein step S3 specifically comprises: dynamically adjusting the allocation of CPU resources occupied by VMs based on popularities of virtual network function (VNF) units; according to a difference in popularity of request flows reaching Mobile Edge Computing (MEC) servers, by means of replacing equal CPU allocation, dynamically adjusting computing resources of each VM in the vehicle by adapting to the popularities of different types of requests, thus minimizing the power of vehicular server resources and then determining the self-adaptive energy-optimal model.

5. The self-adaptive energy-optimal vehicle clustering method based on fuzzy C-means according to claim 4, wherein when the self-adaptive energy-optimal model meets the convexity of the power model, an optimal cluster number k is determined by means of convex optimization; and when the self-adaptive energy-optimal model does not meet the convexity of the power model, the optimal cluster number Æ is found out by means of a traversal search algorithm.

6. The self-adaptive energy-optimal vehicle clustering method based on fuzzy C-means according to claim 1, wherein one cluster in step S4 comprises one cluster head and a plurality of cluster members, namely, vehicle nodes; and each vehicle is only allocated to one cluster, ensuring that the clusters obtained after grouping do not overlap each other.

7. The self-adaptive energy-optimal vehicle clustering method based on fuzzy C-means according to claim 1, wherein the grouping the vehicles into clusters by means of a fuzzy C- means algorithm according to vehicle modeling parameters comprises: grouping a data set X= {x1, x2, ..., Xn} having attribute values of the vehicles vi, V2, ..., Vn into €

; LU500343 € {2, ..., n} clusters, wherein the data set X has three attributes of the vehicle speed, distance, and angle; the cluster grouping status is represented by a membership matrix U of kxn, the value of uj; indicates a degree of membership of the vehicle v; to a cluster j, and 2; € (0, 1); when u;=0, it indicates that the vehicle v; does not belong to the cluster j; and when u;=1, it indicates that the vehicle v; is allocated to the cluster j; and the greater the value of #; indicates a higher degree of membership of the vehicle v; to the cluster J, and u, meets the following formula: © © Zuy= Lu >0 i=l J=1 i=12- 60 J=1,2, nn

8. The self-adaptive energy-optimal vehicle clustering method based on fuzzy C-means according to claim 7, wherein a mean vector matrix of the clusters is defined as B= {wı, 42, us, ..., He}, and a distance between the vehicle data set X and the mean vector matrix B is shown by a matrix D; and the distance matrix D is calculated as follows by using the Euclidean distance dy = yx — 4; Wx, —#, ) dy, a ) dy, D=, : +. : 5 d, ... d., then, a target function is: Fi ES 3 N © 2 3 ofl — HI ~ = 2 d’ 2 lu = i= j= i=l j= wherein m is a fuzzy coefficient which is usually equal to 2; by means of Lagrangian optimization, the Lagrangian multiplier 4 is introduced, and the target function is written into: JS urd (Zu )- mE = ; and à cf — =0, —=0 LL a au, ; CL partial derivatives are let to be 7 by using the method of Lagrange multipliers,

to deduce updating formulas of the degree of membership and the clustering center mean vector as follows: 2 € d. Niel uw, = 1 + ’ / Se i p= ul, [> ul Fl JA

9. The self-adaptive energy-optimal vehicle clustering method based on fuzzy C-means according to claim 1, wherein step S5 specifically comprises: after the vehicles are clustered by means of the fuzzy C-means algorithm, calculating weighted sums of the vehicles v; in the cluster based on a WCA algorithm, and selecting a vehicle node with the minimum weighted sum as the cluster head; and replacing characteristic parameters considered in the WCA with vehicle position information a = S:, (1) [ [ , a weighted mobility value ‘”“ /, and an entropy value H, (7, Af), wherein the vehicle RE] position information is * “1, which indicates whether the vehicles are far away . | SY, (1) . from each other or close to each other; the weighted mobility value “**/ of the vehicle measures the relative stability of the vehicle to its neighboring vehicles; and the entropy value Hy (t, Ar) of the vehicle reflects whether a network with a vehicle v as the center is stable.

10. The self-adaptive energy-optimal vehicle clustering method based on fuzzy C-means according to claim 9, wherein the weighted sum of the vehicle v is calculated as follows: W,(t)=w, {=p + wy Sy, (1) + wy (=H, (1.1) 3 > ow =1,w >0 wherein wi, wa, W3 are corresponding weighting factors, and ‘=! ; and values of the weighting factors are set according to a degree of importance of corresponding parameters.