CN113094857B - Controller layout method of energy-saving software-defined vehicle network - Google Patents

Abstract

The invention discloses a controller layout method of an energy-saving software-defined vehicle network, which comprises the following steps: s1, constructing a minimum controller selection algorithm based on an actual geographic scene; s2: adding a self-adaptive mechanism into the artificial bee colony algorithm; s3: and (2) constructing a multi-objective artificial bee colony algorithm by combining the pareto optimal principle on the basis of the step (S2). The controller layout scheme proposed by the patent is superior to other controller layout schemes. The IMABC provided by the invention has lower energy cost, lower delay and better delivery ratio, and fully shows the reliability and feasibility of the algorithm.

Description

Controller layout method of energy-saving software-defined vehicle network

Technical Field

The invention belongs to the technical field of wireless communication, and relates to an energy-saving controller layout method of a software-defined vehicle network.

Background

In recent years, energy costs in data center networks have attracted increased attention. As applications of large network architectures grow more and more, and in order to reduce access delays for applications, the data centers of these networks are geographically distributed, resulting in the consumption of significant amounts of power in the data center network. Also, with the rapid growth of vehicles and limited road infrastructure, traffic on roads is also faced with problems such as increased transmission delay and lack of management. Particularly with the advent of Software Defined Vehicle Networks (SDVNs), vehicles can also communicate with controllers. To guarantee the communication quality in an SDVN, controllers need to be reasonably geographically distributed and multiple controllers need to cooperate. The energy costs in SDVNs will inevitably increase to reduce the huge delay between the vehicle and the controller. Therefore, a reasonable number of controllers must be considered in an SDVN. Determining the number and location of controllers, also known as controller placement issues (Controller Placement Problem, CPP), has attracted considerable attention from researchers. As a core component of a software defined network (Software Defined Network, SDN), a controller assumes complex computational tasks in the SDN. However, due to the limited computational power of a single SDN controller, there is still a critical problem of insufficient scalability when handling large numbers of requests, which means that multiple controllers must be deployed to meet complex computations and dynamic network changes in large networks. CPPs were originally introduced and addressed as a facility site selection problem, a well known NP problem. Many performance metrics are used to evaluate CPP from a number of aspects, including energy consumption, latency, load balancing, and availability. As a basic goal of green communication system design, energy conservation may reduce the energy costs of the controller. Furthermore, this increases the runtime of controllers of the SDN. As one of the most important issues of SDN, latency can affect state synchronization between network devices. The load balancing index is another important index for balancing the load of the entire network and distributing it among a plurality of operable units.

Disclosure of Invention

In order to solve the controller layout problem in the SDVN of the large dynamic network and meet the energy-saving requirement, the invention provides an energy-saving controller layout method of a software-defined vehicle network. For an area, the optimization of minimum controller coverage and meeting a plurality of targets is realized, and the communication requirements of vehicles in the area are ensured while the energy consumption cost of the controller layout of the area is reduced.

The technical scheme provided by the invention is that the controller layout method of the energy-saving software-defined vehicle network comprises the following steps:

s1, constructing a minimum controller selection algorithm based on an actual geographic scene;

s2: adding a self-adaptive mechanism into the artificial bee colony algorithm;

s3: based on the step S2, constructing a multi-objective artificial bee colony algorithm by combining the pareto optimal principle;

a controller topology method of an energy efficient software defined vehicle network, further comprising: s4: a route calculation algorithm is employed to evaluate the performance of the route metric.

Further, in the step S1, the central controller is assigned to each controller; the controllers are divided into two types, namely a normally open controller and a dynamically opened controller;

the step S1 specifically comprises the following steps:

S11, collecting information of roads connected with each intersection by adopting a controller;

s12: judging the distance L between two intersection points i and j by providing the maximum communication radius com_ran (unit is meter) of the controller;

s13: for the controller dis (c) _i ,c _j ) Judging the distance between the two points;

and S14, adding traffic flow to manage the state of the controller.

Further, the step S12 is specifically that if L is greater than 4com_ran, the controller is placed tangentially; if L is greater than 2com_ran and less than 4com_ran, then placing a single controller in the middle of the two intersections; the positions of all controllers are stored in c_set, and optimization of c_set is performed; for each iteration, constructing a graph G (V, E) representing the current controller network, the node set V representing the controllers in c_set;

further, the step S13 specifically includes: if dis (c) _i ,c _j ) If the communication is smaller than com_ran, the communication is judged to be valid and stored in E, a connection subset con_set is generated according to G (V, E), the minimum and maximum x-axis coordinates and y-axis coordinates minx, maxx, miny and maxy of each connection set are obtained, and the rectangle corresponding to each connection set is calculated and stored in r_set;

traversing the rectangle in r_set if the rectangle r _i With another rectangle r _j Overlap, a new rectangle r is generated _k To cover r _i And r _j Then the overlapped rectangle r is deleted in r_set _i And r _j . For all r in r_set _i ，r _i All controllers within the coverage area will be rearranged by fish scale placement. If r _i With only two controllersAnd they are closely laid out, both controllers are deleted and a new controller is placed in between them. If there are multiple controllers in the rectangle, then all controllers will be purged and new controllers placed using the fish scale layout. All the addition or deletion operations to the controller are recorded and updated in c_set. This operation is cycled until a stable layout scheme is obtained. Traversing all segments in s_set; if the controller c_i in c_set does not overwrite any segment in s_set, c_i is marked as a useless controller and deleted from c_set.

Further, the step S14 specifically includes:

all segments in s_set are traversed. If the controller c in c_set _i Without covering any segment in s_set, then c _i Marked as useless controller and deleted from c_set. Next, the vehicle position v_set between the plurality of time slices is traversed and collected. If the vehicle v _i The closest controller c can be found _j And v _i At c _j Within communication range, then c _j Managed number of vehicles s [ j ]]Increase by 1. Controller c _j After all vehicles are recorded, the number of vehicles is counted by a number of vehicles counter s j]And two thresholds: gamma and delta are compared. Gamma and delta are determined by the transmission radius, the duration of the data packet and the sum of vehicles contained in the data packet. This step divides c_set into a fixed on controller set fix_c_set and a dynamically switchable controller set sw_c_set. If s [ i ]]>Delta, the corresponding controller c is turned on _i And c is carried out _i Fix_c_set is inserted. If s [ i ]]<Gamma, then the controller c is explained _i The managed vehicle nodes are island nodes. Since the number of islanding vehicle nodes is very small and the power consumption of the controller required to manage islanding nodes is too large, this will lead to poor routing performance, c is deleted from c_set _i . Other controllers are added to the dynamically switchable controller set sw_c_set. Further, for other controllers in fix_c_set, a rectangular set r_set is generated.

The controllers in the rectangle set r_set are sparsely placed in each rectangle;

if the traffic of a certain controller is low and the controller manages a limited quarantine node, fix_c_set and sw_c_set of the controller are deleted and the algorithm is ended.

Further, the adaptive mechanism in S2 is: all employed bees are first divided into two parts, based on a solution x in a randomly generated population p _i The upper half will arrive at the solution with the better fitness value and then perform a local search, the other half will be sent to the worse solution and based on the globally optimal solution x _best And carrying out global optimal search, updating the fitness value according to the solution of the last iteration after each iteration, and increasing the proportion k of the upper half bees along with the increase of the iteration times. The ratio of neighborhood searches is equal to the ratio of global optimum searches in the early stages, and more neighborhood searches are performed in the later stages. Because the solutions gradually gather around the current optimal solution as the number of iterations becomes larger, it is more necessary to avoid local optimality when performing neighbor searches;

finally, a new solution is created. Selecting the first n solutions from 2n newly generated optimal solutions as a new population p according to the fitness value _new ；

Further, the artificial bee colony algorithm comprises the following steps of:

each solution has a probability Pro _i Probability Pro _i Is determined based on the fitness value of the solution; solutions with higher fitness values have higher probability Pro _i Roulette mechanism is used to generate each solution x _i Roulette probability value rand (0, 1), probability Pro is determined for each solution _i Comparing with rand (0, 1), if Pro _i If the probability of (1) is greater than rand (0, 1), then the solution will generate a new solution based on the local search, and then add the new solution to the population; otherwise, probability Pro _i Worse than rand (0, 1), then the solution will not perform a local search and will be discarded; and comparing the next solution; once the population number reaches 2n, the optimization phase of the following bees is completed; from the newly generated population p by comparing fitness values of each solution _new Selecting n solutions with optimal fitness;

the improvement steps of the reconnaissance bees: based on the maximum number of iterations maxiter, if the optimal solution is stabilized beyond maxiter/4 times, the improved detection bees will reach the current optimal population, and then the population is mutated randomly by H times through neighbor search. Thus, H new populations are generated, the first n of all solutions of the H+1 populations are selected as new population p _new 。

Further, the multi-objective artificial bee colony algorithm in S3 includes:

s31, randomly generating an initial population p;

obtaining each solution x according to NSGA-II rapid non-dominant sorting method and diversity retention mechanism _i The non-dominant rank and the solution density value dens, selecting the solution with the minimum density value in the non-dominant ordered sequence of 1 as the global optimal solution;

s32: performing a multi-objective adaptive employment bee phase, filtering according to a non-dominant ranking of solutions;

the non-dominant rank of each solution will be determined and if the solution does not belong to a non-dominant rank of 1, the solution will be moved to the globally optimal solution in random steps. Otherwise, if the solution does belong to level 1, the solution will move from level 1 to the less dense solution in random steps.

If the non-dominant ranking value of the solution is not a level 1, the roulette probability is set to 0. Otherwise, calculating the roulette probability according to the density of the solution;

s33: selecting a solution according to the roulette probability;

the probability value for each solution is compared to the roulette probability and if the probability value for the solution is higher than its roulette probability, the solution will move to a solution that is less dense but also starts at level 1. Otherwise, if the probability value of the solution is low, discarding the solution and comparing the next solution until N new solutions are generated;

s34: a scout bee stage of multi-target self-adaption is improved;

if the population remains stable after each iteration, the stable value is incremented by 1, when the stable value reaches maxiter/4, a random mutation is performed, the best solution with the smallest density in class 1 is selected to be mutated in the random dimension, and this step generates N new solutions.

Further, the routing calculation algorithm in S4 specifically includes:

s41: pre-calculating a relationship between the controller and the vehicle as an input;

s42: acquiring a hop count list hop and a corresponding next hop node list next_hop;

s43: update v _i Is provided for the routing information of (a).

The invention provides an energy-saving controller layout method of a software-defined vehicle network, and firstly, a minimum controller selection algorithm is provided, which can reduce the number of controllers and ensure the coverage range of the area. In addition, an improved multi-objective artificial bee colony algorithm is provided on the basis of the proposed improved artificial bee colony algorithm. The multi-objective artificial bee colony algorithm can judge which controllers are to be opened for data transmission according to the real-time flow. Finally, a route calculation algorithm is proposed to evaluate the performance of the route metrics. Experiments show that the invention has better performance in terms of routing metric than other existing CPP schemes.

It is to be understood that both the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the disclosure of the invention as claimed.

Drawings

The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the invention and together with the description, serve to explain the principles of the invention.

In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings that are required to be used in the description of the embodiments or the prior art will be briefly described below, and it will be obvious to those skilled in the art that other drawings can be obtained from these drawings without inventive effort.

FIG. 1 is a schematic layout diagram of a controller layout method for an energy-efficient software-defined vehicle network according to an embodiment of the present disclosure;

fig. 2 is a schematic diagram of a fish scale layout according to an embodiment of the disclosure;

FIG. 3 is a schematic diagram of a sparse layout according to an embodiment of the present disclosure;

FIG. 4 is a topology of selected regions in accordance with an embodiment of the present disclosure;

FIG. 5 is a graph showing the coverage effect of the controller of the present invention when the coverage radius is 250 meters;

FIG. 6 is a graph of the coverage effect of a controller according to the present invention at a coverage radius of 500 meters;

FIG. 7 is a graph showing experimental data of the evaluation function F5 according to the present invention;

FIG. 8 is a graph showing experimental data of the evaluation function F7 according to the present invention;

FIG. 9 is a graph showing experimental data of the evaluation function F9 according to the present invention;

FIG. 10 is a graph showing experimental data of the evaluation function F10 according to the present invention;

FIG. 11 is a graph showing the effect of coverage experiments of NSGA-II according to the present invention;

FIG. 12 is a graph showing the effect of coverage experiments of MOPSO according to the present invention;

FIG. 13 is a graph of the coverage experimental effectiveness of IMABC according to the present invention;

FIG. 14 is a graph showing the comparison of experimental data of energy consumption of a controller layout scheme according to the present invention;

FIG. 15 is a graph showing the comparison of time delay experimental data of a controller layout scheme according to the present invention;

fig. 16 is a comparison chart of experimental data of delay jitter in a controller layout scheme according to the present invention;

FIG. 17 is a graph showing comparison of experimental data of the delivery rate of the controller layout scheme according to the present invention;

fig. 18 is a graph showing comparison of the average hop count experimental data of the controller layout scheme according to the present invention.

Detailed Description

Reference will now be made in detail to exemplary embodiments, examples of which are illustrated in the accompanying drawings. When the following description refers to the accompanying drawings, the same numbers in different drawings refer to the same or similar elements, unless otherwise indicated. The implementations described in the following exemplary examples do not represent all implementations consistent with the invention. Rather, they are merely examples of systems consistent with aspects of the invention as detailed in the accompanying claims.

With the advent of more and more large-scale network applications, the energy costs in data center networks have increased greatly. In Software-defined vehicle networks (SDVNs), as the number of vehicles increases, the communication delay between vehicles and between the vehicles and the controller also increases greatly. To reduce the large amount of latency, multiple controllers need to be laid out to assist in communication, which inevitably results in significant power consumption. Therefore, the number and location of controllers called controller placement problems (Controller Placement Problem, CPP) must be considered in reducing the energy costs in SDVNs. Although CPPs have been intensively studied in static networks, they have not been effectively addressed in highly dynamic and complex networks. The reasonable placement of the controller can reduce energy costs, thereby enabling green communications in the SDVN. This embodiment proposes a novel energy efficient CPP method in SDVN.

Since CPP is considered a facility location problem, a conventional approach to solving this problem is a heuristic algorithm. The invention uses a novel global optimization algorithm-artificial bee colony algorithm (Artificial Bee Colony algorithm, ABC) based on colony intelligence. ABC is designed according to the foraging behavior of bees, and the bees are classified into three types, namely employment bees, following bees and bees of a scouter. Each bee performs a different action. ABC shows better local and global search functionality and faces drawbacks such as premature convergence. Furthermore, ABC tends to fall into local optima and converges slowly in the later stages of the iteration. Thus, the present embodiment improves the ABC algorithm by modifying the search equation.

A controller placement method (Minimum cOntroller Selection mechanism, MOSA) for an energy efficient software defined vehicle network. The method may use as few controllers as possible to cover traffic segments in a given topology. A traffic judgment mechanism is added to further reduce the number of operation controllers. MOSA groups controllers into two types, normally open controllers and dynamically open controllers. This division will use a minimum of controllers to ensure energy efficient communication in the SDVN.

S1: to save energy consumption in guaranteeing SDVN communications, the present embodiment is directed to covering traffic lanes with a minimum of controllers. This can significantly reduce energy costs. In addition, the algorithm also strives to ensure the communication effect between all vehicles and the controller in this topology. The MOSA is assigned to each controller by the central controller.

The first step is to collect information of each intersection and the roads connecting the intersections. This operation records the latitude and longitude of each intersection. A controller is first placed at each intersection.

In a second step, the distance L between the two intersection points i and j is determined by providing the maximum communication radius com_ran of the controller. If L is greater than 4com_ran, the controllers are placed tangentially. If L is greater than 2com_ran and less than 4com_ran, then a single controller is placed in the middle of the two intersections. The positions of all controllers are stored in c_set. Then an optimization of c_set is performed. For each iteration, a graph G (V, E) is constructed representing the current controller network. Node set V represents the controller in c_set.

Third, for two controllers c _i ,c _j Distance dis (c) _i ,c _j ) And judging. If dis (c) _i ,c _j ) Less than com _ ran, then the communication is determined to be acceptable and stored in E. From G (V, E) the connected subsets con_set are generated, in which step the minimum and maximum x-axis coordinates and y-axis coordinates minx, maxx, miny, maxy of each connected subset are obtained. The corresponding rectangle of the connected subset can be calculated by the minimum and maximum coordinates and stored in r_set. Traversing the rectangle in r_set if the rectangle r _i With another rectangle r _j Overlap, a new rectangle r is generated _k To cover r _i And r _j Then the overlapped rectangle r is deleted in r_set _i And r _j . For all r in r_set _i ，r _i All controllers within the coverage area will be rearranged by fish scale placement. If r _i If there are only two controllers and they are closely laid out, both controllers are deleted and a new controller is placed in the middle of them. If there are multiple controllers in the rectangle, then all controllers will be purged and new controllers placed using the fish scale layout. All the addition or deletion operations to the controller are recorded and updated in c_set. This operation is cycled until a stable layout scheme is obtained.

In the next step, traffic flow will be added to further manage the state of the controller. All segments in s_set are traversed. If the controller c in c_set _i Without covering any segment in s_set, then c _i Marked as useless controller and deleted from c_set. Next, the vehicle position v_set between the plurality of time slices is traversed and collected. If the vehicle v _i The closest controller c can be found _j And v _i At c _j Within communication range, then c _j Managed number of vehicles s [ j ]]Increase by 1. Controller c _j After all vehicles are recorded, the number of vehicles is counted by a number of vehicles counter s j]And two thresholds: gamma and delta are compared. Gamma and delta are determined by the transmission radius, the duration of the data packet and the sum of vehicles contained in the data packet. This step divides c_set into a fixed on controller set fix_c_set and a dynamically switchable controller set sw_c_set. If s [ i ]]>Delta, the corresponding controller c is turned on _i And c is carried out _i Fix_c_set is inserted. If s [ i ]]<Gamma, then the controller c is explained _i The managed vehicle nodes are island nodes. Since the number of islanding vehicle nodes is very small and the power consumption of the controller required to manage islanding nodes is too large, this will lead to poor routing performance, c is deleted from c_set _i . Other controllers are added to the dynamically switchable controller set sw_c_set. Further, for other controllers in fix_c_set, a rectangular set r_set is generated. Unlike the operation described above, the controller in r_set is sparsely placed in each rectangle. Since the controller in fix_c_set is always in an on state, the routing performance of the area can be guaranteed, so that sparse layout can be used to reduce the amount of overheadThe number of controllers needed to save power consumption in the area. And outputting fix_c_set and sw_c_set, and ending the algorithm. The difference between the fish scale layout and the sparse layout can be seen in fig. 2 and 3. The fish scale layout ensures the full coverage effect of the controller coverage area, so that the number of controllers in the fish scale layout is more than that of the sparse layout. On the premise of ensuring the coverage effect of the controllers, the number of the controllers is reduced to the greatest extent by the sparse layout, so that the number of the controllers used by the sparse layout is less than that of the fish scale layout.

S2: this embodiment proposes an improved artificial bee colony algorithm, the improvement being based mainly on the behaviour of three bees to obtain. A unique adaptive fitness value mechanism is added. Compared with other improved ABC algorithms, the algorithm of the patent can find the approximate optimal solution cutting iteration time with the minimum error.

The invention improves the artificial bee colony algorithm, and mainly improves the artificial bee colony algorithm into a self-adaptive mechanism. Unlike other adaptation mechanisms, the adaptation mechanism proposed by the present invention first divides all employed bees into two parts. This process is based on a solution x in a randomly generated population p _i Is used for the adaptation value of the (c). The top half will reach a solution with a better fitness value and then perform a local search. The other half will be sent to the worse solution and based on the global optimal solution x _best And performing global optimal search. After each iteration, the fitness value is updated according to the solution of the previous iteration. As the number of iterations increases, the proportion k of the upper bees becomes greater. The ratio of neighborhood searches is equal to the ratio of global optimum searches in the early stages, and more neighborhood searches are performed in the later stages. Since the solutions gradually gather around the current optimal solution as the number of iterations becomes larger, it is more necessary to avoid local optimality when performing neighbor searches. Finally, a new solution is created. Selecting the first n solutions from 2n newly generated optimal solutions as a new population p according to the fitness value _new 。

The follow-up step of the bees still employs the original roulette mechanism, but with some modifications added. Each solution has a probability Pro _i . Probability Pro _i Is based on the fitness value determination of the solutionA kind of electronic device. Solutions with higher fitness values have higher probability Pro _i . Roulette mechanism is used primarily to generate each solution x _i The roulette probability value rand (0, 1). For each solution, the probability Pro _i Compared to rand (0, 1). If Pro is _i If the probability of (1) is greater than rand (0, 1), then the solution will generate a new solution based on the local search and then add the new solution to the population. Otherwise, probability Pro _i Worse than rand (0, 1), then the solution will not perform a local search and will be discarded. And the next solution is compared. Once the population number reaches 2n, the optimization phase following the bees is completed. From the newly generated population p by comparing fitness values of each solution _new N solutions with optimal fitness are selected.

The main improvement step of the scout bees is based on the maximum number of iterations maxiter. If the best solution is stabilized more than maxiter/4 times, the improved detection bees will reach the current best population, and then the population is mutated randomly H times by neighbor search. Thus, H new populations are generated. The first n solutions of all solutions of the H+1 populations are selected as the new population p _new 。

S3: in order to solve the CPP problem in the multi-objective dynamic scene, an improved multi-objective artificial bee colony algorithm (IMABC) based on improved ABC is provided. IMABC is used to select the switching state of the controller. In this way, the number of controllers that are turned on can be significantly reduced to achieve green communication.

Based on S2, this embodiment proposes a multi-objective artificial bee colony algorithm (IMABC). IMABC is based on the combination of our proposed improved artificial bee colony algorithm with the pareto optimal principle. Furthermore, IMABC uses the NSGA-II fast non-dominant ordering method and diversity retention mechanism. The reference point mechanism in NSGA-III was also used in the IMABC we proposed as an improvement. The next section will introduce the main ideas of the fast non-dominant ordering method, the diversity retention mechanism and the reference point mechanism.

In the fast non-dominant ordering method, each solution has two characteristics, one is a non-dominant solution ordering result, and the other is a solutionIs a density of (3). The non-dominant de-ordering result is obtained through multiple steps. For each solution x _m Two metrics are calculated. Is a dominant number n _m That is, the number of solutions that can dominate the solution m, denoted as s _m . Another metric is a set of solutions x _m Dominant solutions. Will dominate the count n _m All solutions of zero are set to the first layer non-dominant ordering. Then, for each solution x in the first layer non-dominant ordering _m At s _l Any other solution x with m in _l Dominant number n _l Will be decremented by 1. After this step, if the dominant number of any solution becomes zero, then the members are set to the second non-dominant ranking result. This process will continue until all of the ranked results are identified or the ranked results are 6.

In the diversity protection mechanism, a crowding distance method may be used to determine solution density. For each objective function, the congestion distance of the solution is calculated. The solutions are first sorted in ascending order according to the value of the objective function. The solution with the smallest and largest function values is called the boundary solution. The boundary de-allocates an infinite distance value. The crowding distance of the remaining solutions is equal to the absolute normalized difference of the function values of the two adjacent solutions. In other objective functions, the same operation of calculating the crowding distance is performed. The overall congestion distance value of the solution is calculated and defined as the sum of the individual distance values to each solution. For this purpose, solution densities can be obtained.

A reference point mechanism has been added to IMABC to provide and adaptively update extended reference points. First, non-dominant rank classification is performed on the entire population. All solutions ordered 1 to l from non-dominant are first included in the selected front row. If |sfont|=n, no further operation is required. Then, the next generation starts with s=sfront. For |sfront| >N, having selected a member from one to level (l-1) as a member of sfront, the remaining (k=n- |sfront|) solutions are selected from the solutions of the optional remaining ranking levels. First a reference point Zr on the hyperplane is defined. Zr is created on triangles whose vertices are (1, 0), (0, 1, 0) and (0, 1). Each solution x of sfront and ofront _i It is necessary to associate with its corresponding reference point. Thus, the pole of sfront is determined by determining the minimum of each objective function. Find the closest solution to the coordinate axis of each objective function and hyperplane based on Jie Shengcheng. Adaptive normalization is then performed on all solutions. The specific steps are to calculate the intercept of each coordinate axis on the hyperplane and set these intercepts to a normalized standard. From these intercepts, the normalized coordinates of all solutions are calculated. Also, each normalized solution ns _i And reference point rp _j The distance between them is calculated as sfront and ofront. Each solution nsi needs to be associated with the nearest reference point rp _j And (5) associating. If solution in sfront and rp _j Associated, rp _j The benefiting base value nc [ j ]]Increase by 1. To screen solutions, a "interest-based protection operation" is performed. Identifying the minimum Li Jizhi nc [ min ] ]Reference point rp of (2) _min 。

If nc [ min ]]=0 (meaning that there is no solution in sfront and reference point rp _min Associated), rp _min There are two results. If there are one or more solutions and reference points rp _min Associated with rp _min Adding the solution with the shortest distance to the population p, counting nc [ min ]]1 is added. However, if ofront is not aligned with reference point rp _min Any solution of the correlation, rp _min Is not within the current solution. If nc [ min ]]1(s) _f A member already associated with the reference point), then from ofront with reference point rp _min The associated random solution will be added to population p. Counting nc [ min ]]1 is added. After updating the benefit base value, the process is repeated K times to fill s _f Is not shown in the drawings).

The Multi-objective optimized swarm algorithm ((Improved Multi-objective Artificial Bee Colony algorithm, IMABC)) consists essentially of the steps of first randomly generating an initial population p, deriving from the two processes described above, the non-dominant rank of each solution xi and a solution density value of the solution, dens. Selecting the solution with the smallest density value in the non-dominant rank of 1 as the globally optimal solution, in the following steps, executing a Multi-objective adaptive employment stage, a Multi-objective adaptive follow-up stage and a Multi-objective adaptive reconnaissance stage, the improvement of the Multi-objective adaptive employment stage being based on the non-dominant rank and the solution density of the current N solutions, the solution being no longer ordered according to objective function values, but being screened according to the non-dominant rank of the solutions, at this stage, determining that the non-dominant rank of each solution is not 1, if the solution does not belong to the non-dominant rank, otherwise, moving in random steps from 1 to the lower density solution, at this step, then, the solution being set up with a probability value of the solution, and the solution of the probability of the new probability value being higher than the solution, the probability value being calculated at this stage, if the solution is not the solution is higher than the solution rank value, the probability value of the solution being higher than the solution rank value of the first, and the probability value being determined at this stage, and if the solution is not higher than the probability value is calculated, and the probability value is higher than the solution value is calculated at the next stage, and is determined according to the value of the solution value of the non-dominant rank, the solution will move to a lower density solution in level 1. Otherwise, if the probability value of a solution is low, the solution is discarded and the next solution is compared until N new solutions are generated. The last part is an improvement of the multi-target adaptive spy bee stage. If the population remains stable after each iteration, the stable value is increased by 1. When the stable value reaches maxiter/4, random mutagenesis will be performed. The best solution with the smallest density in class 1 is chosen to be mutated in the random dimension. This step generates N new solutions.

S4: in order to test the efficiency of the invention, a route calculation algorithm is designed, and under the layout scheme of the controller, the route indexes when the source vehicle node sends the data packet to the destination vehicle node are compared to judge the efficiency of the scheme.

The invention adopts the software-defined vehicle network of the LTE-V2X technology, and optimizes a plurality of targets by the energy-saving layout scheme of the controller of the area. The invention provides a minimum controller algorithm for the region, improves a traditional heuristic algorithm, provides a multi-objective optimization algorithm on the basis of the traditional heuristic algorithm, and improves the performance of the algorithm. A routing calculation algorithm is presented to evaluate the routing efficiency of the controller coverage scheme. The invention greatly improves the energy cost of the controller layout and improves the communication efficiency between vehicles.

The initial position of the controller c_set can be obtained by the MOSA algorithm. Furthermore, all dynamically placed controllers s [ n ] are handled using IMABC]Is a switching state of (a). The route calculation mechanism can obtain a fixed-on controller set fix_c_set and a dynamic switch set sw_c_set, and the fixed-on controller set fix_c_set and the dynamic switch set sw_c_set can be used in the subsequent route. S is a pareto solution set based on three indexes, wherein the three indexes are time delay, energy consumption and load balancing respectively. First, the relationship between the controller and the vehicle is used as an input. For each vehicle v _n Find the closest controller and store it in v_c [ n ]]Is a kind of medium. v _n And v_c [ n ]]The distance between them is calculated and stored. The next step is to acquire the current hop count hop list and the next hop next_hop list. This step may be obtained by recording the vehicle and the available controller closest to the vehicle. Finding all vehicles that can communicate with the controller in one hop, in particular if the vehicle vs]The closest controller v_c [ s ]]The distance between them is smaller than com_ran (next_hop)<com dis), the vehicle is recorded and the current hop count hop s is recorded]Record 1 and next hop next_hop [ s ]]Recorded as 0. In this way, all vehicles that are one hop away from the controller have been processed. Next, a multi-hop vehicle needs to be considered. Traversing all unprocessed vehicles v _i (hop[i]=0), this step can obtain v _i Is adjacent to the vehicle. If there is a processed adjacent vehicle (hop [ i ]]=1), which is stored in a set of available neighbor sets an_set. Then, calculate at v _i With the nearest controller v_c mini]The distance between them is smallest (min (dis [ i ]][n])+v_c_d[n]) Is provided. After this step, the slave v can be obtained _i The shortest distance to the controllers, mini, can act as a relay from the vehicle to the controllers. Finally, update v _i Is a hop list hop [ i ]]Updated to hop mini]+1, next hop list next_hop [ i ]]Updating to mini. In additionVehicle controller relation information v_c [ i ]]＝mini，v_c_d[n]＝dis[i]And mini+v_c_d [ mini ]]An update is needed for further iterations. After this iteration, this step process completes all two-hop vehicles. For more hops vehicles, the above process will be recursively traversed. The maximum number of hops for the algorithm is set to 5, which means that each vehicle can find its nearest controller within 5 hops and transmit the data packet. Through these steps, the method can be based on next_hop [ i ]]A routing path is obtained for all vehicles. Routing paths from the source and destination vehicles src and des to their controllers r_src and r_des may be through recursive processing of next_hop [ src ]]And next_hop [ des ]]Is obtained. At the same time, v_c [ src ] is calculated using Dijkstra]And v_c [ des ]]And a routing path r_con therebetween. Then, r_des is reserved and combined with r_src and r_con. In this way, the entire routing path from src to des is available. Through this path, the effect of the route index, such as delay, energy consumption, packet delivery rate, etc., when transmitting the packet from src to des can be evaluated.

The present invention selects the iron western region of Chinese Shenyang as topology as shown in FIG. 4. The lanes of the area were relatively clean and the subfields of the area were square placed so the area was suitable for simulation of the experiment. The network topology of the area is available from OpenStreetMap, and the geographical information of the topology can be edited by making some adjustments to the original topology using a Java OpenStreetMap editor (JOSM). Vehicle related information such as speed and number of vehicles is set from the SUMO platform. The experimental results of MOSA can be seen from both figures 5 and 6. As is clear from fig. 5, 177 controllers can be placed in this area when the transmission radius of the controllers is set to 250 meters. In addition, 81 controllers among the controllers may be set as fixed controllers. These 81 controllers will remain on while the other controllers are selected to be dynamically turned on or off. In fig. 6, when the transmission radius of the controller is set to 500 meters, the number of controllers is reduced to 40, and the number of fixed controllers is simultaneously reduced to 10. The transmission radius of the SDN controller may be adjusted according to parameters of different communication technologies. As the radius of the controller transition increases, the number of controllers decreases significantly.

The improved artificial bee colony algorithm provided by the patent is simulated on MATLAB and compared with the original Artificial Bee Colony (ABC), particle Swarm Optimization (PSO) and Gravitation Search Algorithm (GSA). To evaluate the performance of the proposed method, a standard basis function is used to evaluate the population intelligent algorithm. The ability to find the optimal solution under different functions is evaluated based on different benchmarks. The invention selects four representative functions in the standard reference functions, namely F5, F7, F9 and F10 functions respectively. These reference functions include unimodal and multimodal functions, and therefore the performance of the proposed algorithm of the present invention can be fully tested. Further, the parameters of the simulation were set as follows: the population is set to 30 and the number of iterations is set to 3000. The experimental results record the error curves for each algorithm at the time of testing, as shown in fig. 7-10. It is clear from the figure that the algorithm proposed by this patent always has minimal error in handling the different problems. When the solution falls into a local optimum for a particular time, the algorithm can effectively jump out of the local optimum. Meanwhile, compared with other algorithms, the time cost for finding the globally optimal solution is within a reasonable range.

Simulation of the multi-objective swarm algorithm is based on a PlatEMO experimental platform. In the simulation of this patent, the DTLZ2 target problem was used to evaluate the performance of the algorithm. In addition, the present experiment selects default values of parameters as parameters for the selected algorithm. Two common multi-objective evolutionary algorithms were selected for comparison with IMABC. These algorithms are non-dominant ordered genetic algorithm (NSGA-II) and multi-objective particle swarm optimization algorithm (MOPSO), respectively. In CPP problems, the uniformity of the algorithm is an important performance indicator for evaluating the controller distribution algorithm. Since these feasible solutions are Cheng Palei torr solution sets, each pareto solution set must be uniform, so that each solution is guaranteed to have a balanced effect, the uniformity is chosen as an evaluation index for the experiment. As can be seen from fig. 11 to 13, when evaluating the multi-objective DTLZ2 problem, the solution obtained by the IMABC algorithm proposed by this patent has the best distribution uniformity compared to MOPSO and NSGA-II.

To evaluate the CPP regimen in SDVN, the present patent selects three CPP regimens for comparison with the IMABC set forth in the present patent. These schemes are a greedy layout, a non-zero and gaming layout and an MHNSGA-II layout. These layout schemes are also evaluated using metrics such as delay, jitter, delivery rate, and energy consumption.

The results of the comparative simulation experiments are presented in fig. 14, showing the energy consumption experimental effects of the four algorithms. The index is used to evaluate the energy costs of all controllers. In these algorithms, the number of on controllers is different. The controllers in the greedy layout are all on. The number of open controllers in IMABC is less than the greedy layout. The number of MHNSGA-II and non-zero and gaming layouts is minimal. As the number of vehicles increases, the energy consumption of the algorithm theoretically increases. However, the energy consumption of the IMABC algorithm is less than that of the greedy algorithm, and is similar to the algorithm with less controller turn-on times. This shows that IMABC has advantages in terms of reduced energy costs. The increase in energy consumption of MHNSGA-II and non-zero and gaming is not significant as the number of vehicles increases. This is mainly because both algorithms are insensitive to variations in the number of vehicles. They propose solutions that do not capture the dynamic characteristics of the on-board network. This defect is also manifested in the experimental results of the transmission rate, delay and jitter.

Fig. 15 and 16 show delay and jitter performance in transmitting data packets. The IMABC is much less time-lapse than the MHNSGA-II and non-zero and gaming time-lapse. The delay of IMABC is close to the greedy delay, but slightly worse than the greedy algorithm. However, the greedy algorithm achieves a slight advantage in terms of latency, but makes a great sacrifice in terms of power consumption. In addition, the delay jitter of IMABC shows the stability of its algorithm.

Fig. 17 shows packet delivery rates under different layout algorithms. Under the index, the algorithm effect provided by the patent is almost the same as the performance of the greedy algorithm. In a greedy layout, good performance is justified because all controllers are already on. The number of open controllers in IMABC is less than the greedy algorithm, so this difference is acceptable. In addition, since the greedy packet delivery rate reaches 95%, this performance indicates the excellent performance of the MOSA algorithm of this patent. In the case of an average packet delivery rate of 92%, the performance of IMABC exceeds that of other algorithms, proving that the transmission of packets by IMABC in SDVN is also reliable.

The performance of the average hop count is shown in fig. 18. The present metric mainly describes the average number of hops of successfully transmitted packets from the source controller to the destination controller. When the data packet transmission of the controller fails, the vehicle uses the GPSR to forward the data packet by itself. In addition, if the transmission still fails, the vehicle may still attempt to transmit the data packet three times. Otherwise, the packet will be discarded. Similar to greedy, packet transmission in IMABC can be completed within 6 hops. However, MHNSGA-II and non-zero and gaming require nearly 8 hops to transmit the data packets. This indicates that both algorithms are unreliable when transmitting packets in an SDVN. This means that routing quality cannot be effectively guaranteed with MHNSGA-II or non-zero and gaming methods.

The invention provides an energy-saving scheme for solving the problem of controller layout in a dynamic SDVN scene with a plurality of targets. Based on the actual topology, the minimum controller selection algorithm (MOSA) proposed by this patent can ensure that a minimum number of controllers are used to cover an area. Further, in this case, considering the traffic flow, the controllers are divided into two types, an opened controller and a dynamically opened controller, to further reduce the number of controllers. An improved multi-objective artificial bee colony algorithm (IMABC) is provided, and based on the improved adaptive artificial bee colony algorithm provided by the invention, the IMABC can select the switching state of a controller according to traffic flow. In addition, the present invention also designs a routing computation algorithm to evaluate the performance of the CPP solution using routing metrics such as delay, jitter, cost, average hop count, and packet delivery rate. Experimental results show that the controller layout scheme provided by the invention is superior to other controller layout schemes. IMABC has lower energy costs, lower delays and better delivery rates, indicating the reliability and feasibility of the algorithm.

Other embodiments of the invention will be apparent to those skilled in the art from consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of the invention following, in general, the principles of the invention and including such departures from the present disclosure as come within known or customary practice within the art to which the invention pertains. It is intended that the specification and examples be considered as exemplary only, with a true scope and spirit of the invention being indicated by the following claims.

Claims (1)

1. A method for controller placement of an energy efficient software defined vehicle network, comprising:

s1, constructing a minimum controller selection algorithm based on an actual geographic scene:

assigning to each controller by the central controller; the controllers are divided into two types, namely a normally open controller and a dynamically opened controller;

the step S1 specifically comprises the following steps:

s11, collecting information of roads connected with each intersection by adopting a controller;

s12, judging the distance L between two intersection points i and j by providing the maximum communication radius com_ran of the controller, wherein the unit of the maximum communication radius com_ran is meter; if L is greater than 4com_ran, then the controller is placed tangentially; if L is greater than 2com_ran and less than 4com_ran, then placing a single controller in the middle of the two intersections; the positions of all controllers are stored in c_set, and optimization of c_set is performed; for each iteration, constructing a graph G (V, E) representing the current controller network, the node set V representing the controllers in c_set;

s13 for controller dis (c) _i ,c _j ) The distance between the two is judged: if dis (c) _i ,c _j ) If the communication is smaller than com_ran, the communication is judged to be valid and stored in E, a connection subset con_set is generated according to G (V, E), the minimum and maximum x-axis coordinates and y-axis coordinates minx, maxx, miny and maxy of each connection set are obtained, and the rectangle corresponding to each connection set is calculated and stored in r_set;

Traversing the rectangle in r_set if the rectangle r _i With another rectangle r _j Overlap, a new rectangle r is generated _k To cover r _i And r _j Then the overlapped rectangle r is deleted in r_set _i And r _j ；

For all r in r_set _i ，r _i All controllers within the coverage area are rearranged through fish scale placement; if r _i If there are only two controllers and they are closely laid out, both controllers are deleted and a new controller is placed in the middle of them;

if a plurality of controllers are in the rectangle, all controllers are cleared and new controllers are placed by using the fish scale layout; all the adding or deleting operations to the controller are recorded and updated in c_set; cycling this operation until a stable layout scheme is obtained; traversing all segments in s_set; if the controller c_i in c_set does not overwrite any segment in s_set, c_i is marked as a useless controller and deleted from c_set;

s14, adding traffic flow to manage controller status: c_set is divided into an open controller set fix_c_set and a switchable controller set sw_c_set; collecting and traversing vehicle position v_set between service time slices, if vehicle v _i The closest controller c can be found _j And v _i At c _j Within communication range, then c _j Managed number of vehicles s [ j ]]An increase of 1; controller c _j After all vehicles are recorded, the number of vehicles is counted by a number of vehicles counter s j]And two thresholds: comparing gamma with delta; gamma and delta are determined by the transmission radius, the duration of the data packet, and the sum of vehicles contained in the data packet; this step divides c_set into a fixed on controller set fix_c_set and a dynamically switchable controller set sw_c_set, if s [ i ]]>Delta, the corresponding controller c is turned on _i And c is carried out _i Inserting fix_c_set; if s [ i ]]<Gamma, then the controller c is explained _i The managed vehicle nodes are island nodes; since the number of islanding vehicle nodes is very small and the power consumption of the controller required to manage islanding nodes is too large, this will lead to poor routing performance, c is deleted from c_set _i The method comprises the steps of carrying out a first treatment on the surface of the Other controllers are added to the dynamically switchable controller set sw_c_set; for other controllers in fix_c_set, generating a rectangular set r_set;

the controllers in the rectangle set r_set are sparsely placed in each rectangle;

if the traffic of a certain controller is low and the controller manages a limited isolated node, then delete fix_c_set and sw_c_set of the controller and end the algorithm;

S2: adding a self-adaptive mechanism into the artificial bee colony algorithm; the self-adaptive mechanism is as follows: all employed bees are first divided into two parts, based on a solution x in a randomly generated population p _i The upper half will arrive at the solution with the better fitness value and then perform a local search, the other half will be sent to the worse solution and based on the globally optimal solution x _best After each iteration, the fitness value is updated according to the solution of the last iteration, and the proportion k of the upper half bees is larger and larger along with the increase of the iteration times; in the early stage, the ratio of the neighborhood searches is equal to the ratio of the global optimum searches, and in the later stage, more neighborhood searches are performed; because the solutions gradually gather around the current optimal solution as the number of iterations becomes larger, it is more necessary to avoid local optimality when performing neighbor searches;

finally, a new solution is created, the first n solutions are selected as new population p from 2n newly generated optimal solutions based on fitness values _new ；

The artificial bee colony algorithm comprises the following bee improvement steps:

each solution has a probability Pro _i Probability Pro _i Is determined based on the fitness value of the solution; solutions with higher fitness values have higher probability Pro _i Roulette mechanism is used to generate each solution x _i Roulette probability value rand (0, 1), probability Pro is determined for each solution _i Comparing with rand (0, 1), if Pro _i If the probability of (1) is greater than rand (0, 1), then the solution will generate a new solution based on the local search, and then add the new solution to the population; otherwise, probability Pro _i Worse than rand (0, 1), then the solution will not perform a local search and will be discarded; and comparing the next solution; once the population number has reached 2n,the optimization phase of following bees is completed; from the newly generated population p by comparing fitness values of each solution _new Selecting n solutions with optimal fitness;

the improvement steps of the reconnaissance bees: based on the maximum iteration number maxiter, if the optimal solution is stabilized more than maxiter/4 times, the improved investigation bees reach the current optimal population, and then the population is randomly mutated H times through neighbor searching; thus, H new populations are generated, the first n of all solutions of the H+1 populations are selected as new population p _new ；

S3: based on the step S2, constructing a multi-objective artificial bee colony algorithm by combining the pareto optimal principle; the multi-objective artificial bee colony algorithm comprises the following steps:

S31, randomly generating an initial population p;

obtaining each solution x according to NSGA-II rapid non-dominant sorting method and diversity retention mechanism _i The non-dominant rank and the solution density value dens, selecting the solution with the minimum density value in the non-dominant ordered sequence of 1 as the global optimal solution;

s32: performing a multi-objective adaptive employment bee phase, filtering according to a non-dominant ranking of solutions;

determining a non-dominant rank of each solution, and if the solution does not belong to the non-dominant rank of 1, moving the solution to the globally optimal solution in random step size; otherwise, if the solution does belong to level 1, the solution will move from level 1 to the lower density solution in random steps;

if the non-dominant ranking value of the solution is not level 1, the roulette probability is set to 0, otherwise, the roulette probability is calculated according to the density of the solution;

s33: selecting a solution according to the roulette probability;

comparing the probability value of each solution to the roulette probability, if the probability value of the solution is higher than its roulette probability, the solution will move towards a solution with a lower density but also starting from level 1; otherwise, if the probability value of the solution is low, discarding the solution and comparing the next solution until N new solutions are generated;

s34: a scout bee stage of multi-target self-adaption is improved;

If the population remains stable after each iteration, adding 1 to the stable value, executing random mutation when the stable value reaches maxiter/4, and selecting the optimal solution with the minimum density in the level 1 to mutate in a random dimension, wherein N new solutions are generated in the step;

s4: a routing computation algorithm is employed to evaluate the performance of a routing metric, the routing computation algorithm comprising:

s41: pre-calculating a relationship between the controller and the vehicle as an input;

s42: acquiring a hop count list hop and a corresponding next hop node list next_hop;

s43: update v _i Is provided for the routing information of (a).