Internet of vehicles network slice resource allocation method based on swarm intelligence optimization algorithm

Through the slicing resource allocation method of the Internet of Vehicles network based on group intelligence optimization algorithm, the problem of matching the complexity of Internet of Vehicles resource allocation and user needs is solved, efficient resource allocation and service quality satisfaction are achieved, and system costs are reduced.

CN119012156APending Publication Date: 2024-11-22JILIN UNIVERSITY
View PDF 0 Cites 3 Cited by

Patent Information

Application Number
CN202410961247.3
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-17
Publication Date
2024-11-22

AI Technical Summary

Technical Problem

The existing Internet of Vehicle resource allocation technology is difficult to effectively take into account users' differentiated service quality needs and resource utilization, especially in the case of limited resources, which leads to the complexity of the allocation of sliced ​​resources in the Internet of Vehicle network and is difficult to achieve efficient allocation.

Method used

Using the Internet of Vehicle Network slice resource allocation method based on the group intelligence optimization algorithm, the binary marine predator optimization algorithm with Logistic-Cubic cascade chaos and the multi-strategy improvement of African vulture optimization algorithm can be decomposed and the user-related decision-making and resource allocation problems are realized, and joint Optimize, optimize the total system cost and service quality constraints.

Benefits of technology

It effectively reduces system costs, realizes the reasonable allocation of limited resources in the Internet of Vehicles, meets users' differentiated service quality needs, and improves resource utilization efficiency.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119012156A_ABST
    Figure CN119012156A_ABST
Patent Text Reader

Abstract

The invention relates to an Internet of Vehicles network slice resource allocation method based on a swarm intelligence optimization algorithm, and belongs to the technical field of Internet of Vehicles. A multi-base-station and multi-vehicle-user Internet of Vehicles system model, a communication model, a calculation model and a slicing model are constructed, the Internet of Vehicles establishes a joint optimization problem model of user association decision, communication resource and calculation resource allocation based on network slicing under the constraint of limited communication resources and calculation resources, the total cost of the system is minimized, and the user association decision, communication resource and calculation resource allocation efficiency is improved. A joint optimization problem is decomposed into two sub-problems of user association decision and resource allocation, and iterative solution is performed by adopting a joint optimization method based on swarm intelligence. According to the method, the search capability of the algorithm is enhanced, the population is prevented from falling into local optimum, exploration and development are balanced, the diversity of the population is increased, differentiated service quality requirements and resource utilization of users can be considered, and reasonable allocation of limited resources of the Internet of Vehicles is realized.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The patent of this invention belongs to the field of Internet of Vehicles technology, and specifically relates to a method for allocating network slice resources of the Internet of Vehicles based on a swarm intelligence optimization algorithm. Background Art

[0002] With the advent of the 5G era, vehicles have been endowed with more computing, storage, perception and communication functions, and the Internet of Vehicles has become a popular paradigm that can provide reliable Internet services to drivers and passengers. Through intensive vehicle-to-vehicle and vehicle-to-infrastructure communications, the goals of improving road safety, traffic efficiency and infotainment experience can be achieved. Compared with traditional cellular networks, there are more diverse applications in the Internet of Vehicles scenario, which leads to heterogeneous service quality requirements for the Internet of Vehicles. For example, the service delay related to autonomous driving needs to be at most a few milliseconds to provide better safety. And other types of services, such as value-added services, whose main purpose is to reduce traffic congestion and provide passenger entertainment, allow longer delays and lower reliability. Although the diverse communication standards have led to heterogeneous user requirements for service quality, however, the existing Internet of Vehicles resource allocation technology solutions generally do not fully consider the differentiated service quality requirements of vehicle users.

[0003] 5G wireless networks use many emerging technologies. Network softwareization and intelligence are the key to 5G networks supporting different types of vehicle user services. Network slicing is a virtual network infrastructure whose core technologies are software-defined networking and network function virtualization. It can dynamically and flexibly allocate physical basic network resources to each logical network slice according to demand to support a wide range of use cases with various performance and service requirements. At present, network slicing research in the Internet of Vehicles scenario is mostly focused on network slicing architecture, and there is still room for further exploration in terms of resource allocation, service quality requirements, algorithm complexity, etc.

[0004] Network slicing is a promising solution for the scheduling and allocation of Internet of Vehicles resources. The design concept of resource sharing is adopted between network slices. However, the resources of the Internet of Vehicles system are limited, and the performance requirements of users are diverse, which greatly increases the complexity and difficulty of Internet of Vehicles resource scheduling. In addition, vehicles generate a large number of tasks that need to be processed during operation, which makes vehicle users have requirements for communication resources, computing resources and other resources. The joint allocation of multi-dimensional resources makes the resource allocation problem of vehicle network slicing more complicated. In addition, the existing related technical solutions generally take fairness as the research goal, lack of research on the joint allocation of multi-dimensional resources and the differentiated service quality requirements of vehicle users. Therefore, it is urgent to study effective strategies and algorithms to take into account the differentiated performance requirements and resource utilization of users under the constraints of limited resource quantity, so as to achieve efficient allocation of limited communication resources and computing in the Internet of Vehicles. Summary of the invention

[0005] The present invention proposes a method for allocating network slice resources of an Internet of Vehicles based on a swarm intelligence optimization algorithm to achieve efficient allocation of limited communication resources and computing in the Internet of Vehicles.

[0006] The technical solution adopted by the present invention comprises the following steps:

[0007] Step 1: Based on network slicing, build a multi-base station and multi-vehicle Internet of Vehicles system model:

[0008] Consider a heterogeneous vehicle network consisting of multiple base stations and multiple vehicle users in an access network enabled by mobile edge computing. All base stations are connected to an SDN controller, which is responsible for collecting network information and executing access network slicing decisions. Mobile edge computing servers are deployed at each base station. Vehicle users under different network slices need to associate with a suitable base station and offload their computing tasks to the base station for processing through wireless links. The base station set is represented as The slice collection is s=e represents enhanced Mobile Broadband (eMBB) slice, s=u represents Ultra Reliable Low Latency Communication (URLLC) slice, and the set of vehicle users served by each slice is The total number of system vehicle users is , vehicle user n s The association decision with base station m is expressed as represents vehicle user n s Associated with base station m, This means that the vehicle user is not associated with base station m;

[0009] Step 2: Establish the communication model, computing model and slicing model of the Internet of Vehicles system:

[0010] (1) Communication model:

[0011] make For vehicle users n s The channel gain between the base station m and the vehicle user is related to the distance between the vehicle user and the base station. There is independent and identically distributed additive Gaussian white noise in the Internet of Vehicles system. s The spectrum benefit between BS m is:

[0012]

[0013] Where p is the transmission power of the vehicle, σ 2 is the noise power,

[0014] make Assign base station m to vehicle user ns The proportion of communication resources to vehicle users n s The uplink transmission rate to base station m is expressed as:

[0015]

[0016] Among them B m is the total amount of communication resources of base station m, that is, the total amount of bandwidth resources;

[0017] (2) Calculation model:

[0018] The transmission power of the base station is usually much greater than the transmission power of the vehicle, and the data size after task processing is much smaller than the task size before processing. Therefore, this model ignores the downlink transmission delay after task processing. The service delay of vehicle users includes task offloading delay and task processing delay. Each vehicle user has a computing task to be processed, and its task can be expressed as in To calculate the data volume of the task, The number of CPU cycles required to complete a task, is the maximum tolerable delay of the task. s Associated with base station m, that is hour:

[0019] The task offloading delay of this user is expressed as:

[0020]

[0021] The task processing delay of this user is expressed as:

[0022]

[0023] The total delay of completing the user task is expressed as:

[0024]

[0025] in Assigned to user n by base station m s The proportion of computing resources, F m is the total amount of computing resources owned by base station m;

[0026] (3) Slice model:

[0027] The service provided by slice e has a high tolerance for latency and focuses on the transmission rate of vehicle users. Consider the vehicle user n accessing slice e. s The transmission rate obtained cannot be lower than the rate threshold R of the slice embb , its quality of service (QoS) requirement is expressed as:

[0028]

[0029] Slice u mainly provides security-related services, that is, latency-sensitive services, which require lower latency. Its QoS requirement is expressed as:

[0030]

[0031] Step 3: Under the constraints of limited communication resources and limited computing resources, the Internet of Vehicles establishes a joint optimization problem model of user association decision, communication resource allocation, and computing resource allocation based on network slicing to minimize the total cost of the system;

[0032] Step 4: Decompose the joint optimization problem into two sub-problems: user association decision and resource allocation. The joint optimization method based on group intelligence (JOMBGI) is used to iteratively solve the problem. For the user association decision sub-problem, the Logistic-Cubic Binary Marine Predators Algorithm (LCBMPA) based on Logistic-Cubic cascade chaos is used to solve it; for the resource allocation sub-problem, the Multi-strategy Improved African Vultures Optimization Algorithm (MIAVOA) is used to solve it.

[0033] The joint optimization problem model established in step 3 of the present invention is expressed as:

[0034] Under the constraints of limited communication resources and limited computing resources, with the goal of minimizing the total system cost, a joint optimization problem model of user association decision, communication resource allocation, and computing resource allocation is established:

[0035] The total system cost takes into account the operation cost of the slice and the violation cost of QoS constraints. The operation cost is related to the amount of communication resources and computing resources allocated to the slice by the system. The operation cost U o It is expressed as:

[0036]

[0037] Among them, w o,b is the communication resource allocation cost coefficient, w o,f Assign cost coefficients to computing resources;

[0038] For slice e, if the transmission rate of the vehicle user in the slice does not meet the requirements, it will cause the QoS constraint violation cost; for slice u, if the task completion time of the vehicle user in the slice exceeds the maximum tolerable delay of the task, it will also cause the QoS constraint violation cost. The QoS constraint violation cost is expressed as:

[0039]

[0040] Among them, w q,e is the cost coefficient of slice e violating the QoS constraint, w q,u is the cost coefficient of slice u violating the QoS constraint; is the indicator function. When x is true, When x is false,

[0041] The total cost of the system can be expressed as:

[0042] U=U o +U q (10)

[0043] Therefore, the joint optimization problem P0 is expressed as:

[0044] P0:min U(a,b,f)

[0045]

[0046] Among them, U(a,b,f) means that the total cost U of the system is a function of the association decision a, the communication resource allocation decision b, and the computing resource allocation decision f; constraint C1 means that the association decision is a 0-1 variable; constraint C2 means that the resources allocated by the base station to the vehicle user are part of the total resources; C3 and C4 respectively mean that the communication resources and computing resources allocated by the base station to the vehicle user do not exceed the total resources; C5 means that each vehicle user must be associated with one and only one base station for task offloading and task processing.

[0047] In step 4 of the present invention, a joint optimization method based on swarm intelligence is used to solve the joint optimization problem, and the problem is decomposed into two sub-problems: association decision-making and resource allocation. The binary marine predator optimization algorithm based on Logistic-Cubic cascade chaos is used to solve the association decision-making sub-problem, and the African vulture optimization algorithm based on multi-strategy improvement is used to solve the resource allocation sub-problem. The solution process is as follows:

[0048] Process (1) Initialize resource allocation, let communication resource allocation decision b = b (0) , computing resource allocation decision f = f (0) , set the current iteration number k of the joint optimization method to 0, and the maximum iteration number to K,

[0049] Process (2) Solve the associated decision sub-problems based on the binary marine predator optimization algorithm based on Logistic-Cubic cascade chaos:

[0050] The communication resource allocation decision b of the joint optimization algorithm at the kth iteration is b=b (k) , computing resource allocation decision f = f (k) Substituting into P0, the joint optimization problem P0 is transformed into P1:

[0051]

[0052] P1 is an optimization problem about the associated decision a, where a is a 0-1 variable. It is solved by the binary marine predator optimization algorithm based on Logistic-Cubic cascade chaos. The solution steps are as follows:

[0053] 1) Set algorithm-related parameters and initialize the population, randomly generate discrete available initial populations, use Logistic-Cubic cascade chaotic sequence to generate chaotic initial populations, and discretize them; calculate the fitness value of each individual in the randomly initialized population and chaotic population, and arrange them in order from small to large; select the first n most fit individuals as the initial population X0. Under the constraints, the fitness function of the optimization problem P1 is expressed as:

[0054] fitness=U(a,b (k) ,f (k) ) (14)

[0055] 2) The generated initial solution is used as the initial prey matrix Prey, and the optimal prey is used as the top predator Best i , and replicate it n times to construct the elite matrix Elite;

[0056] 3) Enter the iteration process:

[0057] (a) Update the parameters related to the number of iterations, calculate the fitness of individuals in Prey and find the top predator;

[0058] (b) Perform ocean memory storage and update the elite matrix Elite;

[0059] (c) Entering the main optimization process, according to the different speed ratios of predator and prey, the optimization process is divided into three main stages:

[0060] Phase 1: The high-speed ratio phase, also known as the exploration phase, occurs in the first third of the optimization process, when the prey moves faster than the predator. At this time, the algorithm focuses on exploration:

[0061]

[0062] in, is the prey's moving step length, is the position vector of the prey; represents Brownian motion and is a random number vector based on a standard normal distribution; the symbol It represents element-by-element multiplication, and the symbol Indicates that x is a vector; and The multiplication of simulates the Brownian motion of prey; P = 0.5 is a constant; is a vector of uniform random numbers in [0,1];

[0063] Stage 2: In the unit speed ratio stage, the speeds of predators and prey are almost equal. This happens in the middle third of the optimization process. Exploration tends to turn into exploitation. Prey uses Lévy motion to perform exploration, and predators use Brownian motion to perform exploitation. The population is divided into two, with half of the individuals used for exploration or exploitation.

[0064] For the first half of the population (i=1,...,n / 2):

[0065]

[0066] in, is a random number vector generated by Lévy distribution, representing Lévy motion. The random number z based on Lévy distribution is generated by formula (17). In formula (17), the random number Q obeys a normal distribution with a mean of 0 and a standard deviation of σ, that is, Q~N(0,σ 2 ); the random number Y follows the standard normal distribution, Y~N(0,1); σ is calculated by formula (18), α=1.5; Γ(·) represents the gamma function, Γ(1+α) equals α! , the Lévy motion of the prey is given by and The multiplication simulation, at the prey position Adding the step size above can simulate the movement of prey. The step size that follows the Lévy distribution is smaller, which can effectively search the nearby area in depth, thus facilitating exploitation.

[0067] For the second half of the population (i=1+n / 2,...,n):

[0068]

[0069] Among them, CF is an adaptive parameter that controls the predator's moving step size. and The multiplication of is a means to simulate the movement of the predator in the Brownian mode, and the prey updates its position according to the predator. t is the current iteration number, and Max_Iter is the maximum iteration number;

[0070] Phase 3: In the low speed ratio phase, the predator is faster than the prey, which is the development phase. This phase is achieved in the last third of the iterations of the optimization process. At low speed ratios, the best strategy for the predator is Lévy motion:

[0071]

[0072] (d) After each stage, the time-varying Sine transfer function TV is used Sin (·) Convert Prey to binary:

[0073] first, Restricted to In scope:

[0074]

[0075] Among them, max(·) and min(·) represent the maximum and minimum values ​​of the returned boundary values, respectively. and Respectively represent the upper and lower bounds of the prey position vector,

[0076] Through the time-varying Sine transfer function TV Sin (·) Convert to the range [0,1]:

[0077]

[0078] Mapping to binary vector

[0079]

[0080] in, is the transfer function TV Sin (·) control parameters, from Gradually decays to The algorithm shifts from exploration to development as iterations proceed. rand represents a random vector between 0 and 1.

[0081] (e) Calculate the fitness value of each individual according to the fitness function and update the top predator Best i ;

[0082] (f) Introducing a mutation factor with the number of iterations t as a parameter to the top predator Best i Expand the adaptive mutation and discretize it, use the greedy strategy to compare the fitness values ​​of individuals before and after the mutation, and select the best individual as the top predator:

[0083]

[0084] Among them, Best i ′ is the top predator after mutation;

[0085] (g) Performing ocean memory storage, which simulates the good memory of ocean predators. By comparing the fitness of the candidate solutions of the current iteration with the corresponding candidate solutions of the previous iteration, the poor candidate solutions are eliminated to simulate the memory of ocean predators;

[0086] (h) Environmental issues are a factor that can cause changes in the behavior of marine predators, such as fish aggregating devices (FADs). Apply the FADs effect to update the predator matrix Prey:

[0087]

[0088] In the formula, FADs = 0.2, which is the probability that FADs affects the optimization process. is a binary vector, constructed by generating a random vector in [0,1]. If a random number is less than 0.2, the random number at the corresponding position is changed to 0. If a random number is greater than 0.2, the random number at the corresponding position is changed to 1. r is a random number uniformly distributed in [0,1]. and are the upper and lower bound vectors of the solution, r1 and r2 represent random indices of the prey matrix;

[0089] (i) Based on the time-varying V-type transfer function TV V (·) Generate binary solution:

[0090]

[0091]

[0092] where arctan(·) is the inverse tangent function; σ and γ are TV V (·) control parameters; For the inversion operation, σ is taken from σ initial Gradually decays to σ final ,γfromγ initial Gradually decays to γ final :

[0093] σ=σ initial +t·(σ final -σ initial ) / Max_Iter (31)

[0094] γ=γ initial+t·(γ final -γ initial ) / Max_Iter (32)

[0095] (j) Calculate the fitness value of each individual, detect the position of the top predator, and update the elite matrix Elite and related parameters;

[0096] 4) Determine whether the current number of iterations t is equal to the maximum number of iterations Max_Iter. If so, output the current top predator position, i.e., the final association decision a=a (k+1) ; Otherwise, the number of iterations t is increased by 1 and returns to step 3) to enter the next iteration;

[0097] Process (3) solves the resource allocation sub-problem based on the multi-strategy improved African vulture optimization algorithm:

[0098] The user association decision a = a finally solved in process (2) (k+1) Substitute into P0, and the joint optimization problem P0 is transformed into P2:

[0099]

[0100] P2 is an optimization problem about the communication resource allocation decision b and the computing resource allocation decision f. Both b and f are continuous variables between 0 and 1. The African vulture optimization algorithm based on multi-strategy improvement is used to solve it. The solution steps are as follows:

[0101] 1) Set algorithm-related parameters and initialize the population, randomly generate an available initial population, and use the Logistic-Cubic cascade chaotic sequence to generate a chaotic initial population; calculate the fitness value of each individual in the randomly initialized population and the chaotic population, and arrange them in order from small to large; select the first n most fit individuals as the initial population X0. Under the constraints, the fitness function of the optimization problem P2 is expressed as:

[0102] fitness=U(b,f,a (k+1) ) (34)

[0103] 2) Enter the iteration process:

[0104] (a) The fitness value of each vulture individual is calculated according to the fitness function. The individual with the best fitness value is taken as the optimal vulture, and the second-best individual is taken as the second-best vulture. The optimal vulture and the second-best vulture guide the remaining individuals. The moving direction is expressed as:

[0105]

[0106] Among them, BestVulture1 represents the best vulture, BestVulture2 represents the second best vulture, L1 and L2 are preset constants, both between 0 and 1, and the sum is 1, p i From the roulette method:

[0107]

[0108] Among them, r i is a random number in the range [0,1],

[0109] (b) Calculate the hunger level of each vulture:

[0110]

[0111] Where rand1 is a random number between [0, 1]; z is a random number between [-1, 1]; t is the current iteration number; Max_Iter is the maximum iteration number, and g is calculated by formula (38):

[0112]

[0113] Where h is a random number between [-2,2]; w is a pre-set parameter, usually set to 2.5.

[0114] (c) When |F i |≥1, the vulture individual enters the exploration stage, and the individual position update formula is expressed as:

[0115]

[0116] Among them, δ and parameter p1 are in the range of [0,1]. p1 is given at the time of initialization and is used to determine which exploration behavior the individual performs. Both rand2 and rand3 are random numbers in the range [0,1]. i represents the distance between the current vulture and the optimal vulture or the suboptimal vulture, which is given by formula (41):

[0117] D i =|C·R i -X i | (41)

[0118] Where C is a random number uniformly distributed between [0,2].

[0119] (d) When |F i |<1, the vulture individual enters the development stage:

[0120] 0.5<|F i When |<1, it enters the middle stage of development, and the position update formula is expressed as:

[0121]

[0122] in, rand4 is a random number in the range of [0,1], and parameter p2 is in the range of [0,1], which is given at the time of initialization. When the vultures compete for food, The vulture performs circling flight behavior, and d is given by equation (43):

[0123] d=R i -X i (43)

[0124] S i1 and S i2 They are given by equations (44) and (45) respectively:

[0125]

[0126]

[0127] Among them, rand5 and rand6 are both random numbers in the range of [0,1];

[0128] |F i When |<0.5, it enters the late development stage, and the position update formula is expressed as:

[0129]

[0130] in, is a random number in the range [0,1]; parameter p3 is in the range [0,1] and is given at initialization. When , vultures perform gathering behavior, and vice versa, The vulture performs the attack behavior, r1 and r2 are both random numbers in the range of [0,1]; α is a fixed constant, usually set to 1.5; σ is given by formula (18);

[0131] A i1 and A i2 They are given by equations (49) and (50) respectively:

[0132]

[0133] (e) Based on the greedy differential evolution mutation strategy, the fitness values ​​of the vulture individuals before and after the mutation are compared, and the positions of the vulture individuals are updated:

[0134]

[0135] (f) Determine whether the current number of iterations t is equal to the maximum number of iterations Max_Iter. If so, output the position and best score of the current optimal vulture, i.e., the final resource allocation decision b (k+1) 、f (k+1) and system cost U (k+1) ; Otherwise, the number of iterations t is increased by 1 and returns to step 2) to enter the next iteration;

[0136] Process (4) obtains the current strategy π through the above steps (k+1) ={a (k+1) ,b (k+1) ,f (k+1)}, update π according to formula (53) (k+1) , if through π (k+1) The total cost U (k+1) Than the strategy π obtained in the previous iteration (k) The total cost U (k) Small, then keep the strategy π (k+1) As the current policy, otherwise the current policy π (k+1) Update to π (k) , determine whether k has reached the maximum number of iterations K of the set joint optimization method, and if so, output the final result π (k+1) , that is, the best association decision and resource allocation decision. If the maximum number of iterations is not reached, k = k + 1, and jump to process (2) for the next iteration;

[0137]

[0138] The solution step 1) in process (2) and the solution step 1) in process (3) of the present invention both adopt the strategy of Logistic-Cubic cascade chaos to generate the cascade chaos initial population. The Logistic-Cubic cascade chaos mapping expression is as follows (13):

[0139] x i+1 =ρ·[μ·x i ·(1-x i )]·(1-[μ·x i ·(1-x i )] 2 ) (13)

[0140] Among them, x i ∈[0,1]; ρ and μ are control parameters. When x0=0.3,ρ=2.595,μ=4,Logistic-Cubic cascade chaos has good chaotic ergodicity. i+1,L =μx i (1-x i) is the Logistic chaotic mapping expression, x i+1,C =ρx i (1-x i 2 ) is the Cubic chaos map expression.

[0141] The specific process of solving step (f) in process (2) of the present invention, introducing the adaptive t distribution operator to perform mutation disturbance on the optimal individual, is:

[0142] Best i ′=Best i +Best i trnd(t) (26)

[0143] Here, trnd(t) represents the t distribution.

[0144] In the solution step (c) of the process (3) described in the present invention, an adaptive trust δ is introduced to improve the exploration phase. The adaptive trust δ is defined as:

[0145]

[0146] In the solution step (d) of the process (3) of the present invention, an adaptive weight coefficient is introduced in the later stage of the development stage to improve the development stage. The adaptive weight coefficients θ1 and θ2 are defined as:

[0147]

[0148] In the solution step (e) of the process (3) of the present invention, based on the greedy differential evolution mutation strategy, the position of the vulture individual is updated, and the differential mutation individual X i,DE It is expressed as:

[0149] X i,DE =X i +F i ·(X r1 -X r2 ) (51)

[0150] Among them, X r1 , X r2 A vulture individual was randomly selected.

[0151] Beneficial effects of the present invention:

[0152] 1. The present invention takes into account the limited communication resources and computing resources in the Internet of Vehicles, and takes into account the differentiated service quality requirements and resource utilization of users, and jointly optimizes the association decision, communication resources and computing resource allocation to minimize the system cost. The proposed Internet of Vehicles network slice resource allocation method based on swarm intelligence optimization algorithm decomposes the problem into two sub-problems: user association decision and resource allocation, and adopts a joint optimization method based on swarm intelligence for iterative solution, taking into account the differentiated service quality requirements and resource utilization of users, effectively reducing the system cost, and realizing the reasonable allocation of limited resources in the Internet of Vehicles.

[0153] 2. A binary marine predator optimization algorithm based on Logistic-Cubic cascade chaos is used to solve the associated decision subproblems: the original MPA randomly generates the initial population, resulting in an uneven distribution of the initial population. The chaotic sequence can generate an initial population with better diversity according to certain rules. However, a single chaotic mapping and its improved forms often have the disadvantages of small full mapping parameter space and low initial value sensitivity. Cascade chaos can increase the complexity of the chaotic system, with a larger parameter space, higher initial value sensitivity, and stronger randomness. Therefore, the present invention adopts the Logistic-Cubic cascade chaos mapping strategy to make the position distribution of the initial population more uniform; in order to increase population diversity and avoid the original MPA from falling into the local optimum, the present invention introduces a t-distribution adaptive perturbation strategy in the optimal individual determination stage, and selects the optimal individual based on the greedy strategy, thereby reducing the probability of the algorithm falling into the local optimum.

[0154] 3. The resource allocation sub-problem is solved by using an improved African vulture optimization algorithm based on multiple strategies: the Logistic-Cubic cascade chaos mapping strategy is used in the initialization stage to make the position distribution of the initial population more uniform; the original AVOA algorithm over-trusts the current optimal vulture in the global exploration stage, so the present invention introduces an adaptive trust degree δ, which increases nonlinearly with the number of iterations t, and reaches 1 in the middle of the iteration, which fully trusts the current optimal vulture, enhances the exploration ability of the vulture, and balances exploration and development. When the original AVOA algorithm performs the aggregation behavior in the late stage of the development stage, it is believed that the optimal vulture and the suboptimal vulture have the same degree of influence on the current vulture, so a certain exploration ability still needs to be added in the middle and late stages of the algorithm optimization stage. To this end, the present invention introduces adaptive weight coefficients θ1 and θ2 for control. θ1 and θ2 increase and decrease respectively as the number of iterations increases. In the initial iteration, the suboptimal vulture has a greater impact on the current vulture. As the iteration proceeds, the impact of the suboptimal vulture decreases nonlinearly, and the impact of the optimal vulture increases nonlinearly and exceeds the suboptimal vulture. The exploration capability is quickly converted into the development capability, the balance between exploration and development is achieved, and the local optimum is avoided. The idea of ​​the differential evolution algorithm is applied, the vulture individuals are mutated and disturbed by the differential information between the vulture individuals, and it is determined whether to update the vulture individuals to the mutated individuals based on the greedy strategy, thereby enhancing the diversity of the population and avoiding falling into the local optimum. BRIEF DESCRIPTION OF THE DRAWINGS

[0155] Figure 1 It is a flow chart of the implementation of the present invention;

[0156] Figure 2 is the system model of the present invention;

[0157] Figure 3 It is a flowchart of the binary marine predator optimization algorithm based on Logistic-Cubic cascade chaos;

[0158] Figure 4 This is a flowchart of the African vulture optimization algorithm based on multi-strategy improvements;

[0159] Figure 5 It is a flow chart of JOMBGI, a joint optimization method based on swarm intelligence;

[0160] Figure 6 It is a schematic diagram of the convergence of the algorithm JOMBGI proposed in the present invention compared with different algorithms;

[0161] Figure 7 It is a curve chart showing the change of QoS satisfaction rate of the algorithm of the present invention and different algorithms in simulation experiments as the number of iterations changes;

[0162] Figure 8 It is a curve chart showing the change of total system cost as the total amount of system communication resources increases;

[0163] Fig. 9 It is a curve graph showing the change of QoS satisfaction rate as the total amount of system communication resources increases;

[0164] Fig.10 It is a curve chart showing the change of total system cost as the total amount of system computing resources increases;

[0165] Fig.11 It is a curve graph showing the change of QoS satisfaction rate as the total amount of system computing resources increases;

[0166] Fig.12 It is a curve diagram showing how the QoS satisfaction rate varies with the number of iterations when the total number of users is different under the default simulation parameters of the algorithm proposed in the present invention. DETAILED DESCRIPTION

[0167] The present invention will be further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0168] The implementation flow chart of the present invention is as follows Figure 1 As shown, the following steps are included:

[0169] Step 1: Based on network slicing, build a multi-base station and multi-vehicle Internet of Vehicles system model:

[0170] Consider a heterogeneous vehicle network consisting of multiple base stations and multiple vehicle users in a mobile edge computing-enabled access network. All base stations are connected to an SDN controller, which is responsible for collecting network information and executing access network slice decisions. The mobile edge computing server is deployed at each base station. Vehicle users in different network slices need to associate with a suitable base station and offload their computing tasks to the base station for processing through wireless links. The base station set is represented as The slice collection is s=e represents enhanced Mobile Broadband (eMBB) slice, s=u represents Ultra Reliable Low Latency Communication (URLLC) slice, and the set of vehicle users served by each slice is The total number of system vehicle users is Vehicle User s The association decision with base station m is expressed as represents vehicle user n s Associated with base station m, It means that the vehicle user is not associated with base station m. The vehicle networking system model constructed by the present invention is shown in the figure below: Figure 2 shown.

[0171] Step 2: Establish the communication model, calculation model and slicing model of the Internet of Vehicles system:

[0172] (1) Communication model:

[0173] make For vehicle users n s The channel gain between the vehicle user and the base station m is related to the distance between the vehicle user and the base station. There is independent and identically distributed additive Gaussian white noise in the Internet of Vehicles system. s The spectrum benefit between BS m is:

[0174]

[0175] Where p is the transmission power of the vehicle, σ 2 is the noise power;

[0176] make Assign base station m to vehicle user n s The proportion of communication resources to vehicle users n s The uplink transmission rate to base station m is expressed as:

[0177]

[0178] Among them B m is the total amount of communication resources of base station m, that is, the total amount of bandwidth resources;

[0179] (2) Calculation model:

[0180] The transmission power of the base station is usually much greater than that of the vehicle, and the data size after task processing is much smaller than the task size before processing. Therefore, this model ignores the downlink transmission delay after task processing. The service delay of vehicle users includes task offloading delay and task processing delay. Each vehicle user has a computing task to be processed, and its task can be expressed as in To calculate the data volume of the task, The number of CPU cycles required to complete a task, is the maximum tolerable delay of the task. s When associated with base station m, that is, hour:

[0181] The task offloading delay of this user is expressed as:

[0182]

[0183] The task processing delay of this user is expressed as:

[0184]

[0185] The total delay of completing the user task is expressed as:

[0186]

[0187] in Assigned to user n by base station m s The proportion of computing resources, F m is the total amount of computing resources owned by base station m;

[0188] (3) Slice model:

[0189] The service provided by slice e has a high tolerance for latency and focuses on the transmission rate of vehicle users. Consider a vehicle user n accessing slice e. s The transmission rate obtained cannot be lower than the rate threshold R of the slice embb Its quality of service (QoS) requirement is expressed as:

[0190]

[0191] Slice u mainly provides security-related services, that is, latency-sensitive services, which require lower latency. Its QoS requirements are expressed as:

[0192]

[0193] Step 3: Under the constraints of limited communication resources and computing resources, the Internet of Vehicles establishes a joint optimization problem model of user association decision, communication resource allocation, and computing resource allocation based on network slicing to minimize the total cost of the system:

[0194] The total system cost takes into account the operation cost of the slice and the violation cost of QoS constraints. The operation cost is related to the amount of communication resources and computing resources allocated to the slice by the system. The operation cost U o It is expressed as:

[0195]

[0196] Among them, w o,b is the communication resource allocation cost coefficient, w o,f Assign cost coefficients to computing resources;

[0197] For slice e, if the transmission rate of the vehicle user in the slice does not meet the requirements, it will cause the QoS constraint violation cost; for slice u, if the task completion time of the vehicle user in the slice exceeds the maximum tolerable delay of the task, it will also cause the QoS constraint violation cost. The QoS constraint violation cost is expressed as:

[0198]

[0199] Among them, w q,e is the cost coefficient of slice e violating the QoS constraint, w q,u is the cost coefficient of slice u violating the QoS constraint, is the indicator function. When x is true, When x is false,

[0200] The total cost of the system can be expressed as:

[0201] U=U o +U q (10)

[0202] Therefore, the joint optimization problem P0 is expressed as:

[0203] P0:min U(a,b,f)

[0204]

[0205] Among them, U(a,b,f) means that the total cost U of the system is a function of the association decision a, the communication resource allocation decision b, and the computing resource allocation decision f; constraint C1 means that the association decision is a 0-1 variable; constraint C2 means that the resources allocated by the base station to the vehicle user are part of the total resources; C3 and C4 respectively mean that the communication resources and computing resources allocated by the base station to the vehicle user do not exceed the total resources; C5 means that each vehicle user must be associated with one and only one base station for task offloading and task processing.

[0206] Step 4: Decompose the joint optimization problem into two sub-problems: user association decision and resource allocation, and use the joint optimization method based on group intelligence (JOMBGI) to iteratively solve it. For the user association decision sub-problem, the Logistic-Cubic Binary Marine Predators Algorithm (LCBMPA) based on cascade chaos is used to solve it; for the resource allocation sub-problem, the Multi-strategy Improved African Vultures Optimization Algorithm (MIAVOA) based on multi-strategy improvement is used to solve it;

[0207] The Marine Predators Algorithm (MPA) is a meta-inspiration algorithm proposed by S. Mirjalili et al. in 2020. The algorithm simulates the different foraging behaviors of predators in the ocean when they are at different speed ratios with their prey. Its advantages are high efficiency and flexibility, simplicity and ease of implementation, excellent robustness and high convergence accuracy, etc., and it performs well in solving practical problems. Although the original MPA is efficient and flexible, the algorithm still has the problem of being easy to fall into local optimality and the convergence accuracy needs to be improved. In order to overcome the shortcomings of the original MPA and further improve the solution performance of MPA, the present invention proposes a binary marine predator optimization algorithm based on Logistic-Cubic cascade chaos.

[0208] For iterative swarm intelligence optimization algorithms, a high-quality initial population can often bring better optimization performance. Compared with the random generation of the initial population by MPA, the chaotic sequence can generate an initial population with better diversity according to certain rules. However, a single chaotic mapping and its improved forms often have the disadvantages of small full mapping parameter space and low initial value sensitivity. Cascade chaos can increase the complexity of the chaotic system, with a larger parameter space, higher initial value sensitivity, and stronger randomness. Therefore, the present invention adopts the Logistic-Cubic cascade chaos mapping strategy to make the position distribution of the initial population more uniform; in order to increase the population diversity and avoid the original MPA from falling into the local optimum, the present invention introduces the t-distribution adaptive perturbation strategy in the optimal individual determination stage, and selects the optimal individual based on the greedy strategy, thereby reducing the probability of the algorithm falling into the local optimum.

[0209] The African Vultures Optimization Algorithm (AVOA) is a metaheuristic algorithm proposed by Abdollahzadeh et al. in 2021, which simulates the living habits and foraging behavior of African vultures. Its advantages are fast convergence speed and high convergence accuracy. However, when solving practical problems, the algorithm still has problems such as unbalanced development mechanism and exploration mechanism, and easy to fall into local optimal solution. In order to overcome the shortcomings of the original AVOA, the present invention proposes an African vulture optimization algorithm based on multi-strategy improvement.

[0210] In the AVOA initialization stage, the Logistic-Cubic cascade chaos mapping strategy is also used to make the position distribution of the initial population more uniform; in order to enhance the exploration ability of individual vultures in the exploration stage, the present invention introduces adaptive trust to prevent other vultures from approaching the current optimal vulture too early in the early stage of algorithm exploration to avoid falling into the local optimum; in order to enhance the exploration ability of the algorithm in the middle and late stages of the optimization stage to avoid falling into the local optimum, the present invention introduces two adaptive weight coefficients to control the importance of the optimal vulture and the suboptimal vulture, achieve a balance between exploration and development, and avoid falling into the local optimum; the idea of ​​the differential evolution algorithm is applied, the vulture individuals are mutated and disturbed by the differential information between the vulture individuals, and whether the vulture individuals are updated to the mutated individuals is determined based on the greedy strategy, thereby enhancing the diversity of the population and avoiding falling into the local optimum.

[0211] The joint optimization problem P0 is a mixed integer programming problem, and the communication resource allocation is coupled with the computing resource allocation. Therefore, the joint optimization problem is decomposed into two sub-problems: user association decision and resource allocation. LCBMPA is used to solve the association decision sub-problem, and MIAVOA is used to solve the resource allocation sub-problem. Figure 3 This is the flow chart of LCBMPA. Figure 4 This is the flow chart of MIAVOA. Figure 5 This is the flowchart of JOMBGI. The specific process of solving the problem using JOMBGI is as follows:

[0212] Process (1) Initialize resource allocation, let communication resource allocation decision b = b (0) , computing resource allocation decision f = f (0) , set the current number of iterations of the joint optimization method k=0, and the maximum number of iterations is K.

[0213] Process (2) Solve the associated decision sub-problems based on the binary marine predator optimization algorithm based on Logistic-Cubic cascade chaos:

[0214] The communication resource allocation decision b of the joint optimization algorithm at the kth iteration is b=b (k) , computing resource allocation decision f = f (k) Substituting into P0, the joint optimization problem P0 is transformed into P1:

[0215]

[0216] P1 is an optimization problem about the associated decision a, where a is a 0-1 variable. It is solved by the binary marine predator optimization algorithm based on Logistic-Cubic cascade chaos. The solution steps are as follows:

[0217] 1) Set algorithm-related parameters and initialize the population, randomly generate discrete available initial populations, use Logistic-Cubic cascade chaotic sequence to generate chaotic initial populations, and discretize them; calculate the fitness value of each individual in the randomly initialized population and chaotic population, and arrange them in order from small to large; select the first n most fit individuals as the initial population X0, and the Logistic-Cubic chaotic mapping expression is as shown in formula (13):

[0218] x i+1 =ρ·[μ·x i ·(1-x i )]·(1-[μ·x i ·(1-x i )] 2 ) (13)

[0219] Among them, x i ∈[0,1]; ρ and μ are control parameters. When x0=0.3,ρ=2.595,μ=4,Logistic-Cubic cascade chaos has good chaotic ergodicity. i+1,L =μx i (1-x i ) is the Logistic chaotic mapping expression, x i+1,C =ρx i (1-x i 2 ) is the Cubic chaos mapping expression;

[0220] Compared with the traditional random initialization method, the chaotic sequence traverses all states in the search space according to the specified rules, and can generate an initial population with wide distribution and good diversity, which helps to improve the algorithm's traversal and convergence accuracy. However, a single chaotic map and its improved form often have shortcomings such as small full mapping parameter space and low initial value sensitivity. Cascade chaos can improve the complexity of the chaotic system, with a larger parameter space, higher initial value sensitivity, and stronger randomness. Compared with Logistic single chaos and Cubic single chaos, Logistic-Cubic cascade chaos mapping has stronger chaotic characteristics and wider traversal ability.

[0221] Under the constraints, the fitness function of the optimization problem P1 is expressed as:

[0222] fitness=U(a,b (k) ,f (k) ) (14)

[0223] 2) The generated initial solution is used as the initial prey matrix Prey, and the optimal prey is used as the top predator Best i , and replicate it n times to construct the elite matrix Elite;

[0224] 3) Enter the iteration process:

[0225] (a) Update the parameters related to the number of iterations, calculate the fitness of individuals in Prey, and find the top predator;

[0226] (b) Perform ocean memory storage and update the elite matrix Elite;

[0227] (c) Entering the main optimization process. According to the different speed ratios of predators and prey, the optimization process is divided into three main stages:

[0228] Phase 1: The high-speed ratio phase, also known as the exploration phase, occurs in the first third of the optimization process, when the prey moves faster than the predator. At this time, the algorithm focuses on exploration:

[0229]

[0230] in, is the prey's moving step length, is the position vector of the prey; represents Brownian motion and is a random number vector based on a standard normal distribution; the symbol It represents element-by-element multiplication, and the symbol Indicates that x is a vector; and The multiplication of simulates the Brownian motion of prey; P = 0.5 is a constant; is a vector of uniform random numbers in [0,1];

[0231] Stage 2: In the unit speed ratio stage, the speeds of predators and prey are almost equal. This occurs in the middle third of the optimization process. Exploration tends to turn into exploitation. Prey uses Lévy motion to perform exploration, and predators use Brownian motion to perform exploitation. The population is divided into two, with half of the individuals used for exploration or exploitation.

[0232] For the first half of the population (i=1,...,n / 2):

[0233]

[0234] in, is a random number vector generated by Lévy distribution, representing Lévy motion. The random number z based on Lévy distribution is generated by formula (17). In formula (17), the random number Q obeys a normal distribution with a mean of 0 and a standard deviation of σ, that is, Q~N(0,σ 2); the random number Y follows the standard normal distribution, Y~N(0,1); σ is calculated by formula (18), α=1.5; Γ(·) represents the gamma function, Γ(1+α) equals α! , the Lévy motion of the prey is given by and The multiplication simulation, at the prey position Adding the step size to the above can simulate the movement of prey. The step size that follows the Lévy distribution is smaller, which can effectively search the nearby area in depth, thus helping to exploit;

[0235] For the second half of the population (i=1+n / 2,...,n):

[0236]

[0237] Among them, CF is an adaptive parameter that controls the predator's moving step size. and The multiplication of is a means to simulate the movement of the predator in the Brownian mode, and the prey updates its position according to the predator. t is the current iteration number, and Max_Iter is the maximum iteration number;

[0238] Phase 3: In the low speed ratio phase, the predator is faster than the prey, which is the development phase. This phase is achieved in the last third of the iterations of the optimization process. At low speed ratios, the best strategy for the predator is Lévy motion:

[0239]

[0240] (d) After each stage, the time-varying Sine transfer function TV is used Sin (·) Convert Prey to binary:

[0241] first, Restricted to In scope:

[0242]

[0243] Among them, max(·) and min(·) represent the maximum and minimum values ​​of the returned boundary values, respectively. and They represent the upper and lower bounds of the prey position vector respectively;

[0244] Through the time-varying Sine transfer function TV Sin (.) Convert to the range [0,1]:

[0245]

[0246] Mapping to binary vector

[0247]

[0248] in, is the transfer function TV Sin (.) control parameters, from Gradually decays to The algorithm shifts from exploration to development as iterations proceed. rand represents a random vector between 0 and 1.

[0249] (e) Calculate the fitness value of each individual according to the fitness function and update the top predator Best i ;

[0250] (f) Introducing a mutation factor with the number of iterations t as a parameter to the top predator Best i Expand the adaptive mutation, use the greedy strategy to compare the fitness values ​​of individuals before and after the mutation, select the best individual as the top predator, and introduce the adaptive t distribution operator to perform mutation disturbance on the best individual. The specific process is as follows:

[0251] Best i ′=Best i +Best i trnd(t) (26)

[0252] Taking the number of iterations as a parameter, in the early stage of iteration, the number of iterations is small, the t distribution is smoother than the Cauchy distribution, the t distribution operator obtains a larger value with a larger probability, and the step length of the position mutation is larger; in the middle stage of iteration, the t distribution is between the Cauchy distribution and the normal distribution, and the t distribution operator takes a relatively compromise value with a large probability; in the late stage of iteration, the t distribution is similar to the standard normal distribution, and the t distribution operator takes a smaller value with a larger probability, resulting in a smaller step length for mutation. Therefore, when the t distribution is introduced to perturb the optimal individual, the algorithm's search ability is improved and the probability of the algorithm falling into the local optimum is reduced;

[0253]

[0254] Among them, trnd(t) represents t distribution, Best i ′ is the top predator after mutation;

[0255] (g) Perform ocean memory storage, which simulates the good memory of ocean predators. By comparing the fitness of the candidate solutions of the current iteration with the corresponding candidate solutions in the previous iteration, the poor candidate solutions are eliminated to simulate the memory of ocean predators;

[0256] (h) Environmental issues are a factor that can cause changes in the behavior of marine predators, such as fish aggregating devices (FADs). Apply the FADs effect and update the predator matrix Prey:

[0257]

[0258] Where, FADs = 0.2, which is the probability that FADs affects the optimization process. is a binary vector constructed by generating a random vector in [0,1]. If a random number is less than 0.2, the random number at the corresponding position is changed to 0. If a random number is greater than 0.2, the random number at the corresponding position is changed to 1. r is a random number uniformly distributed in [0,1]. and are the upper and lower bound vectors of the solution, r1 and r2 represent random indices of the prey matrix;

[0259] (i) Based on the time-varying V-type transfer function TV V (.) Generate binary solution:

[0260]

[0261] Among them, arctan(.) is the inverse tangent function; σ and γ are TV V (.) control parameters; For the inversion operation, σ is taken from σ initial Gradually decays to σ final ,γfromγ initial Gradually decays to γ final :

[0262] σ=σ initial +t·(σ final -σ initial ) / Max_Iter (31)

[0263] γ=γ initial +t·(γ final -γ initial ) / Max_Iter (32)

[0264] (j) Calculate the fitness value of each individual, detect the position of the top predator, and update the elite matrix Elite and related parameters;

[0265] 4) Determine whether the current number of iterations t is equal to the maximum number of iterations Max_Iter. If so, output the current top predator position, i.e., the final association decision a=a (k+1) ; Otherwise, the number of iterations t is increased by 1 and returns to step 3) to enter the next iteration;

[0266] Process (3) solves the resource allocation sub-problem based on the multi-strategy improved African vulture optimization algorithm:

[0267] The user association decision a = a finally solved in process (2) (k+1) Substitute into P0, and the joint optimization problem P0 is transformed into P2:

[0268]

[0269] P2 is an optimization problem about the communication resource allocation decision b and the computing resource allocation decision f. Both b and f are continuous variables between 0 and 1. The African vulture optimization algorithm based on multi-strategy improvement is used to solve it. The solution steps are as follows:

[0270] 1) Set algorithm-related parameters and initialize the population, randomly generate an available initial population, and use the Logistic-Cubic cascade chaotic sequence to generate a chaotic initial population; calculate the fitness value of each individual in the randomly initialized population and the chaotic population, and arrange them in order from small to large; select the first n most fit individuals as the initial population X0. Under the constraints, the fitness function of the optimization problem P2 is expressed as:

[0271] fitness=U(b,f,a (k+1) ) (34)

[0272] 2) Enter the iteration process:

[0273] (a) The fitness value of each vulture individual is calculated according to the fitness function. The individual with the best fitness value is taken as the optimal vulture, and the second-best individual is taken as the second-best vulture. The optimal vulture and the second-best vulture guide the remaining individuals. The moving direction is expressed as:

[0274]

[0275] Among them, BestVulture1 represents the best vulture, BestVulture2 represents the second best vulture, L1 and L2 are preset constants, both between 0 and 1, and the sum is 1, p i From the roulette method:

[0276]

[0277] Among them, r i is a random number in the range [0,1];

[0278] (b) Calculate the hunger level of each vulture:

[0279]

[0280] Where rand1 is a random number between [0,1]; z is a random number between [-1,1]; t is the current iteration number, and Max_Iter is the maximum iteration number. g is calculated by formula (38):

[0281]

[0282] Where h is a random number between [-2,2]; w is a pre-set parameter, usually set to 2.5;

[0283] (c) When |F i |≥1, the vulture individual enters the exploration phase, and the adaptive trust δ is introduced to improve the exploration phase. The adaptive trust is defined as:

[0284]

[0285] The original African Vultures Optimization Algorithm (AVOA) algorithm over-trusts the current optimal vulture in the global exploration phase. The introduction of adaptive trust can prevent other vultures from approaching the current optimal vulture too early in the early exploration phase of the algorithm, avoid falling into the local optimum, and increase the diversity of the early algorithm. The adaptive trust δ increases nonlinearly with the number of iterations t, reaching 1 in the middle of the iteration, fully trusting the current optimal vulture, enhancing the exploration ability of the vulture, and balancing exploration and development;

[0286] The individual position update formula is expressed as:

[0287]

[0288] Among them, δ and parameter p1 are in the range of [0,1]. p1 is given at the time of initialization and is used to determine which exploration behavior the individual performs. Both rand2 and rand3 are random numbers in the range [0,1]. i represents the distance between the current vulture and the optimal vulture or the suboptimal vulture, which is given by formula (41):

[0289] D i =|C·R i -X i | (41)

[0290] Where C is a random number uniformly distributed between [0,2];

[0291] (d) When |F i |<1, the vulture individual enters the development stage:

[0292] 0.5<|F iWhen |<1, it enters the middle stage of development, and the position update formula is expressed as:

[0293]

[0294] in, rand4 is a random number in the range of [0,1], and parameter p2 is in the range of [0,1], which is given at the time of initialization. When the vultures compete for food, The vulture performs circling flight behavior. d is given by equation (43):

[0295] d=R i -X i (43)

[0296] S i1 and S i2 They are given by equations (44) and (45) respectively:

[0297]

[0298] Among them, rand5 and rand6 are both random numbers in the range of [0,1];

[0299] |F i When |<0.5, it enters the late stage of development, and the adaptive weight coefficient is introduced to improve the development stage. The adaptive weight coefficients θ1 and θ2 are defined as:

[0300]

[0301] When the original AVOA algorithm performs the aggregation behavior in the late development stage, it is considered that the optimal vulture and the suboptimal vulture have the same influence on the current vulture. However, in the middle and late stages of the algorithm optimization stage, it is still necessary to increase a certain exploration capability to avoid falling into the local optimum. If it is assumed that the optimal vulture and the suboptimal vulture always have the same influence on the current vulture, it may cause the algorithm to be unable to jump out of the local optimum. The introduced adaptive weight coefficients θ1 and θ2 increase and decrease respectively with the increase in the number of iterations. At the beginning of the iteration, the suboptimal vulture has a greater influence on the current vulture. As the iteration proceeds, the influence of the suboptimal vulture decreases nonlinearly, while the influence of the optimal vulture increases nonlinearly and exceeds the suboptimal vulture, quickly converting the exploration capability into the development capability, achieving a balance between exploration and development, and avoiding the local optimum;

[0302] The position update formula is expressed as:

[0303]

[0304] in, is a random number in the range [0,1]; parameter p3 is in the range [0,1] and is given at initialization. When , vultures perform gathering behavior, and vice versa, The vulture performs the attack behavior, r1 and r2 are both random numbers in the range of [0,1]; α is a fixed constant, usually set to 1.5; σ is given by formula (18);

[0305] A i1 and A i2 They are given by equations (49) and (50) respectively:

[0306]

[0307] (e) Based on the greedy differential evolution mutation strategy, the fitness values ​​of the vulture individuals before and after the mutation are compared, and the positions of the vulture individuals are updated:

[0308] Differential mutation individual X i,DE It is expressed as:

[0309] X i,DE =X i +F i ·(X r1 -X r2 ) (51)

[0310] Among them, X r1 , X r2 The vultures are randomly selected individuals. The differential evolution algorithm is applied to perturb the individuals by using the differential information between the vultures, which enhances the diversity of the population and avoids falling into the local optimum.

[0311] The position update formula is expressed as:

[0312]

[0313] (f) Determine whether the current number of iterations t is equal to the maximum number of iterations Max_Iter. If so, output the position and best score of the current optimal vulture, i.e., the final resource allocation decision b (k+1) 、f (k+1) and system cost U (k+1) ; Otherwise, the number of iterations t is increased by 1 and returns to step 2) to enter the next iteration;

[0314] Process (4) obtains the current strategy π through the above steps (k+1) ={a (k+1) ,b (k+1) ,f (k+1)}, update π according to formula (53) (k+1) , if through π (k+1) The total cost U (k+1)Than the strategy π obtained in the previous iteration (k) The total cost U (k) Small, then keep the strategy π (k+1) As the current policy, otherwise the current policy π (k+1) Update to π (k) , determine whether k has reached the maximum number of iterations K of the set joint optimization method, and if so, output the final result π (k+1) , that is, the best association decision and resource allocation decision. If the maximum number of iterations is not reached, k = k + 1, and jump to step (2) for the next iteration;

[0315]

[0316] The technical effects of the present invention are further described in detail below in conjunction with simulation experiments.

[0317] The simulation takes 15 vehicle users, two slices of eMBB and uRLLC, and two base stations with different resource amounts as an example, and compares the joint optimization method based on swarm intelligence proposed in the present invention, namely JOMBGI, with the following schemes:

[0318] 1. MIBMPA-AVOA: MIBMPA is used to solve association decisions and AVOA is used to solve resource allocation;

[0319] 2. MIBMPA-WOA: MIBMPA is used to solve association decisions, and Whale Optimization Algorithm (WOA) is used to solve resource allocation;

[0320] 3. MIBMPA-Average: Using MIBMPA to solve association decisions, each base station allocates resources evenly to associated users;

[0321] 4. Random: A vehicle user is randomly associated with a base station, and each base station randomly allocates resources to the associated users.

[0322] The main difference between MPA and other swarm intelligence optimization algorithms is that the optimization process is divided into three stages according to the number of iterations, simulating the entire life cycle of predators and prey, and transitioning from a global search of the solution space to a local exploration of the current optimal solution in the solution space. Compared with the random generation of the initial population by MPA, the chaotic sequence can generate an initial population with better diversity according to certain rules. However, a single chaotic mapping and its improved forms often have the disadvantages of small full mapping parameter space and low initial value sensitivity. Cascade chaos can increase the complexity of the chaotic system, with a larger parameter space, higher initial value sensitivity, and stronger randomness. Therefore, the present invention adopts the Logistic-Cubic cascade chaos mapping strategy to make the position distribution of the initial population more uniform; in order to increase the population diversity and avoid the original MPA from falling into the local optimum, the present invention introduces the t-distribution adaptive perturbation strategy in the optimal individual determination stage, and selects the optimal individual based on the greedy strategy, thereby reducing the probability of the algorithm falling into the local optimum.

[0323] The main difference between AVOA and other swarm intelligence optimization algorithms is that the algorithm simulates the living habits and foraging behavior of African vultures, and the optimal vulture and the suboptimal vulture lead the rest of the vultures to find food. By defining the hunger degree of the vultures, the optimization process is divided according to three situations of the hunger degree. The original AVOA randomly generates the initial population, so the present invention also adopts the Logistic-Cubic cascade chaos mapping strategy in the AVOA initialization stage to make the position distribution of the initial population more uniform; in order to enhance the exploration ability of the vulture individuals in the exploration stage, the present invention introduces adaptive trust to prevent other vultures from approaching the current optimal vulture too early in the early stage of algorithm exploration to avoid falling into the local optimum; in order to enhance the exploration ability of the algorithm in the middle and late stages of the optimization stage and avoid falling into the local optimum, the present invention introduces two adaptive weight coefficients to control the importance of the optimal vulture and the suboptimal vulture, achieve a balance between exploration and development, and avoid falling into the local optimum; the idea of ​​the differential evolution algorithm is applied, and the vulture individuals are mutated and disturbed by using the differential information between the vulture individuals, and whether the vulture individuals are updated to the mutated individuals is determined based on the greedy strategy, thereby enhancing the diversity of the population and avoiding falling into the local optimum.

[0324] 1. Simulation parameter settings

[0325] In the simulation experiment, two base stations with different resource amounts are considered, with a coverage radius of 600m. The ratio of various resource amounts of the two base stations is 3:2. By default, the bandwidth resources are set to 6MHz and 4MHz, respectively, and the computing resources are set to 9GHz and 6GHz, respectively. By default, the total number of system vehicle users within the coverage of the two base stations is 15, and the ratio of the number of vehicle users under the eMBB slice to the number of vehicle users under the uRLLC slice is 1:2. The transmission power of the vehicle user p = 23dBm, the noise power σ 2=-114dBm, the path loss is modeled as 128+37.6lg 10 (d), d represents the distance between the base station and the vehicle user, in km, and shadow fading follows a log-normal distribution with a mean of 0 and a standard deviation of 8 dB. Data size Both are set to 10Kbit, the number of CPU cycles required Both are 50MHz. For eMBB users, task tolerance delay It is set to 200ms, and the tolerance delay of uRLLC users is set to 100ms. The communication resource allocation cost coefficients of the two base stations are 20 and 25 respectively, and the computing resource allocation cost coefficients are 20 and 25 respectively. The base station with less resources allocates the same amount of resources at a higher cost. The cost coefficients of slice e and slice u violating QoS constraints are both 400. When solving the sub-problems, the population size in the MIBMPA, MIAVOA, AVOA, and WOA algorithms are all set to 30, the maximum number of iterations is 500, and the fitness function is the total system cost. The maximum number of iterations for JOMBGI and the four comparison algorithms is 1000.

[0326] 2. Simulation content and result analysis

[0327] Figure 6 The figure shows the convergence of the proposed algorithm JOMBGI compared with other algorithms. It can be observed from the figure that although the proposed algorithm and the comparison algorithm can converge within a limited number of iterations, proving the effectiveness of the swarm intelligence optimization algorithm in solving this problem, the proposed algorithm JOMBGI has a better convergence. Figure 6 It can be seen that JOMBGI reaches the optimal value at the 43rd iteration, with a minimum system total cost of 52.98, the fastest convergence speed and the highest convergence accuracy. LCBPMA-WOA converges the slowest and has the lowest convergence accuracy, proving that AVOA and its improved algorithm perform better than the classic WOA in solving this type of resource allocation problem. However, LCBMPA-AVOA reaches the optimal value of 452.90 at the 100th iteration. Compared with this, the improved algorithm JOMBGI has a convergence speed of 57 times higher and a convergence accuracy of 88%, proving that MIAVOA performs better than the original AVOA in solving this type of resource allocation problem. This is because MIAVOA combines Logistic-Cubic cascade chaos mapping, adaptive trust, adaptive weight coefficient, and differential mutation to improve AVOA, avoiding the algorithm from falling into local optimality and enhancing the global optimization ability of the algorithm.

[0328] Figure 7The graph is a comparison of the QoS satisfaction rate of different algorithms with the number of iterations of the algorithm proposed in the present invention. The QoS satisfaction rate is used to measure the satisfaction of the selected solution for the QoS requirements of all vehicle users. The higher the QoS satisfaction rate, the more vehicle users' QoS requirements are met. The QoS satisfaction rate of the algorithm proposed in the present invention is better than that of other algorithms. Figure 7 It can be observed that with the increase in the number of iterations, the QoS satisfaction rate under each algorithm generally increases and gradually converges to a maximum value. Specifically, the QoS satisfaction rate of JOMBGI reaches the optimal value 1, i.e. 100%, after 43 iterations, indicating that the QoS requirements of all users can be met, while the comparison algorithms LCBMPA-AVOA and LCBMPA-WOA converge more slowly, and the QoS satisfaction rates can only reach 93.3% and 66.7%, respectively. Compared with LCBMPA-AVOA, the QoS satisfaction rate of the proposed algorithm is improved by 6.7%, indicating that the improved MIAVOA has better performance. Combined with Figure 6 ,It can be seen that JOMBGI can meet the QoS requirements of users with the minimum system cost, ,which reflects that the proposed algorithm JOMBGI can achieve more effective utilization of limited resources.

[0329] Figure 8 is the curve of the total system cost changing with the increase of the total amount of system communication resources. The richer the communication resources, the easier it is for users to obtain a higher information transmission rate through a reasonable resource allocation scheme, the easier it is for users to meet their QoS requirements, and the lower the total cost. Compared with other algorithms, the total cost of the algorithm JOMBGI proposed in this invention is smaller and can better meet the QoS requirements of all users. Figure 8 It can be observed that as the total amount of system communication resources increases, the total system cost obtained by each algorithm shows a downward trend and finally stabilizes. LCBMPA-WOA and random algorithms perform the worst. Even when the total amount of resources is relatively abundant, the total system cost is still high, which means that LCBMPA-WOA and random algorithms cannot flexibly allocate the resources of the two base stations to meet the QoS requirements of all users. The proposed algorithm JOMBGI has a similar curve to LCBMPA-Average. When the total amount of resources is small, LCBMPA-Average performs better, but as the amount of resources increases to 5.5MHz, the total system cost achieved by JOMBGI is 26.30 lower than that of LCBMPA-Average, and as the amount of resources increases further, the total system cost achieved by JOMBGI is always slightly lower than that of LCBMPA-Average, and the utilization of resources is better.

[0330] Fig. 9 is the curve of QoS satisfaction rate changing with the increase of total system communication resources. Fig. 9It can be observed that as the total amount of system communication resources increases, the QoS satisfaction rate obtained by each algorithm shows an upward trend and finally stabilizes. The curves of the proposed algorithm JOMBGI and LCBMPA-Average are relatively similar. When the amount of resources is not less than 5.5MHz, the QoS satisfaction rate can reach 100%. LCBMPA-AVOA can only reach 93.3% at most. LCBMPA-WOA and random algorithms perform the worst, and their QoS satisfaction rates are always much lower than those of the other three algorithms. Although the QoS satisfaction rate obtained by LCBMPA-Average in the later stage also reaches the optimal level, JOMBGI is Figure 8 The corresponding total system cost is lower, indicating that JOMBGI can more effectively allocate communication resources to appropriate users to meet the QoS requirements of all users.

[0331] Fig.10 is the relationship curve between the total system cost and the total amount of system computing resources. The richer the computing resources, the easier it is for users to obtain a smaller task processing delay through a reasonable resource allocation scheme. The JOMBGI algorithm proposed in this invention has the best performance. Fig.10 It can be seen that the relationship curves between the total system cost of all algorithms and the total amount of system computing resources show a downward trend. As the amount of computing resources increases, the three algorithms JOMBGI, LCBMPA-AVOA, and LCBMPA-Average gradually converge to the minimum value, while the total system cost of LCBMPA-WOA and random algorithms is much higher than these three algorithms even when the resources are abundant. When the amount of resources is small, LCBMPA-Average performs best. After the amount of resources increases to 15.5GHz, the total cost of the proposed JOMBGI algorithm is the smallest. When the amount of resources increases to 18GHz, the total cost of the proposed JOMBGI algorithm is 16.54 lower than that of LCBMPA-AVOA and 50.79 lower than that of LCBMPA-Average, and the utilization of computing resources is better.

[0332] Fig.11 is the relationship curve between QoS satisfaction rate and total system computing resources. Fig.11 It can be seen that the curves of all algorithms show an upward trend. After the total amount of computing resources reaches 15.5GHz, the QoS satisfaction rates of the three algorithms JOMBGI, LCBMPA-AVOA, and LCBMPA-Average can all be stabilized at 100%, while the QoS satisfaction rates of LCBMPA-WOA and random algorithms are generally much lower than these three algorithms, and can only reach more than 90% when the amount of resources reaches 23GHz. Fig.10 It can be seen that JOMBGI can meet the QoS requirements of more users at a lower total cost and achieve rational use of computing resources.

[0333] Fig.12 This is the relationship curve between QoS satisfaction rate and iteration number when the total number of users is different under the default simulation parameters of the JOMBGI algorithm proposed in the present invention. Fig.12 It can be observed that all curves converge within a finite number of iterations, but the fewer the total number of users, the faster the algorithm converges. When the total number of users is 27, the algorithm needs 298 iterations to converge to the optimal value. This is because the larger the total number of users, the higher the dimension of the optimization problem, the increased complexity of the algorithm, and the slower convergence. When the total number of users is less than 15, the QoS satisfaction rate can eventually reach 100%, while when the total number of users is 21 and 27, the QoS satisfaction rate can only reach 76.2% and 55.6% at most. It can be seen that when the total number of users is not less than 27, the QoS requirements of a large number of users cannot be met under the set default resource conditions.

[0334] In summary, Figure 6 and Figure 7 The effectiveness of the algorithm improvement of the present invention can be effectively demonstrated, and the proposed JOMBGI algorithm can converge to a better value in fewer iterations. Figures 8 to 11 , which effectively proves the effectiveness of the user association and resource allocation strategy of the proposed JOMBGI algorithm, and can effectively allocate communication resources and computing resources to appropriate users under different total resource conditions. Fig.12 The proposed algorithm is used to explore the QoS satisfaction of users under different total number of users. It can be concluded that the algorithm proposed in this invention takes into account the differentiated service quality requirements and resource utilization of users under different slices, satisfies the QoS requirements of users to a greater extent with fewer resources, and realizes the reasonable allocation of limited resources in the Internet of Vehicles.

Claims

1. A method for allocating network slice resources of an Internet of Vehicles based on a swarm intelligence optimization algorithm, characterized in that: The following steps are involved: Step 1: Based on network slicing, build a multi-base station and multi-vehicle Internet of Vehicles system model: Consider a heterogeneous vehicle network consisting of multiple base stations and multiple vehicle users in an access network enabled by mobile edge computing. All base stations are connected to an SDN controller, which is responsible for collecting network information and executing access network slicing decisions. Mobile edge computing servers are deployed at each base station. Vehicle users under different network slices need to associate with a suitable base station and offload their computing tasks to the base station for processing through wireless links. The base station set is represented as The slice collection is s=e represents enhanced Mobile Broadband (eMBB) slice, s=u represents Ultra Reliable Low Latency Communication (URLLC) slice, and the set of vehicle users served by each slice is The total number of system vehicle users is Vehicle User s The association decision with base station m is expressed as represents vehicle user n s Associated with base station m, This means that the vehicle user is not associated with base station m; Step 2: Establish the communication model, computing model and slicing model of the Internet of Vehicles system: (1) Communication model: make For vehicle users n s The channel gain between the base station m and the vehicle user is related to the distance between the vehicle user and the base station. There is independent and identically distributed additive Gaussian white noise in the Internet of Vehicles system. s The spectrum benefit between BS m is: Where p is the transmission power of the vehicle, s 2 is the noise power, make Assign base station m to vehicle user n s The proportion of communication resources to vehicle users n s The uplink transmission rate to base station m is expressed as: Among them B m is the total amount of communication resources of base station m, that is, the total amount of bandwidth resources; (2) Calculation model: The transmission power of the base station is usually much greater than the transmission power of the vehicle, and the data size after task processing is much smaller than the task size before processing. Therefore, this model ignores the downlink transmission delay after task processing. The service delay of vehicle users includes task offloading delay and task processing delay. Each vehicle user has a computing task to be processed, and its task can be expressed as in To calculate the data volume of the task, The number of CPU cycles required to complete a task, is the maximum tolerable delay of the task. s Associated with base station m, that is hour: The task offloading delay of this user is expressed as: The task processing delay of this user is expressed as: The total delay of completing the user task is expressed as: in Assigned to user n by base station m s The proportion of computing resources, F m is the total amount of computing resources owned by base station m; (3) Slice model: The service provided by slice e has a high tolerance for latency and focuses on the transmission rate of vehicle users. Consider the vehicle user n accessing slice e. s The transmission rate obtained cannot be lower than the rate threshold R of the slice embb , its Quality of Service (QoS) requirement is expressed as: Slice u mainly provides security-related services, that is, latency-sensitive services, which require lower latency. Its QoS requirement is expressed as: Step 3: Under the constraints of limited communication resources and limited computing resources, the Internet of Vehicles establishes a joint optimization problem model of user association decision, communication resource allocation, and computing resource allocation based on network slicing to minimize the total cost of the system; Step 4: Decompose the joint optimization problem into two sub-problems: user association decision and resource allocation. The joint optimization method based on group intelligence (JOMBGI) is used to iteratively solve the problem. For the user association decision sub-problem, the Logistic-Cubic Binary Marine Predators Algorithm (LCBMPA) based on Logistic-Cubic cascade chaos is used to solve it; for the resource allocation sub-problem, the Multi-strategy Improved African Vultures Optimization Algorithm (MIAVOA) is used to solve it.

2. The method for allocating network slice resources of the Internet of Vehicles based on the swarm intelligence optimization algorithm according to claim 1 is characterized in that: The joint optimization problem model established in step 3 is expressed as: Under the constraints of limited communication resources and limited computing resources, with the goal of minimizing the total system cost, a joint optimization problem model of user association decision, communication resource allocation, and computing resource allocation is established: The total system cost takes into account the operation cost of the slice and the violation cost of QoS constraints. The operation cost is related to the amount of communication resources and computing resources allocated to the slice by the system. The operation cost U o It is expressed as: Among them, w o,b is the communication resource allocation cost coefficient, w o,f Assign cost coefficients to computing resources; For slice e, if the transmission rate of the vehicle user in the slice does not meet the requirements, it will cause the QoS constraint violation cost; for slice u, if the task completion time of the vehicle user in the slice exceeds the maximum tolerable delay of the task, it will also cause the QoS constraint violation cost. The QoS constraint violation cost is expressed as: Among them, w q,e is the cost coefficient of slice e violating the QoS constraint, w q,u is the cost coefficient of slice u violating the QoS constraint; is the indicator function. When x is true, When x is false, The total cost of the system can be expressed as: U=U o +U q (10) Therefore, the joint optimization problem P0 is expressed as: Among them, U(a,b,f) means that the total cost U of the system is a function of the association decision a, the communication resource allocation decision b, and the computing resource allocation decision f; constraint C1 means that the association decision is a 0-1 variable; constraint C2 means that the resources allocated by the base station to the vehicle user are part of the total resources; C3 and C4 respectively mean that the communication resources and computing resources allocated by the base station to the vehicle user do not exceed the total resources; C5 means that each vehicle user must be associated with one and only one base station for task offloading and task processing.

3. The method for allocating network slice resources of an Internet of Vehicles based on a swarm intelligence optimization algorithm according to claim 1 is characterized in that: The step 4 adopts a joint optimization method based on swarm intelligence to solve the joint optimization problem, decomposes the problem into two sub-problems: association decision-making and resource allocation, adopts a binary marine predator optimization algorithm based on Logistic-Cubic cascade chaos to solve the association decision-making sub-problem, and adopts an African vulture optimization algorithm based on multi-strategy improvement to solve the resource allocation sub-problem. The solution process is as follows: Process (1) Initialize resource allocation, let communication resource allocation decision b = b (0) , computing resource allocation decision f = f (0) , set the current iteration number k of the joint optimization method to 0, and the maximum iteration number to K, Process (2) Solve the associated decision sub-problems based on the binary marine predator optimization algorithm based on Logistic-Cubic cascade chaos: The communication resource allocation decision b of the joint optimization algorithm at the kth iteration is b=b (k) , computing resource allocation decision f = f (k) Substituting into P0, the joint optimization problem P0 is transformed into P1: P1 is an optimization problem about the associated decision a, where a is a 0-1 variable. It is solved by the binary marine predator optimization algorithm based on Logistic-Cubic cascade chaos. The solution steps are as follows: 1) Set algorithm-related parameters and initialize the population, randomly generate discrete available initial populations, use Logistic-Cubic cascade chaotic sequence to generate chaotic initial populations, and discretize them; Calculate the fitness value of each individual in the randomly initialized population and the chaotic population, and arrange them in order from small to large; select the first n most fit individuals as the initial population X0. Under the constraints, the fitness function of the optimization problem P1 is expressed as: fitness=U(a,b (k) ,f (k) ) (14) 2) The generated initial solution is used as the initial prey matrix Prey, and the optimal prey is used as the top predator Best i , and replicate it n times to construct the elite matrix Elite; 3) Enter the iteration process: (a) Update the parameters related to the number of iterations, calculate the fitness of individuals in Prey and find the top predator; (b) Perform ocean memory storage and update the elite matrix Elite; (c) Entering the main optimization process, according to the different speed ratios of predator and prey, the optimization process is divided into three main stages: Phase 1: The high-speed ratio phase, also known as the exploration phase, occurs in the first third of the optimization process, when the prey moves faster than the predator. At this time, the algorithm focuses on exploration: in, is the prey's moving step length, is the position vector of the prey; represents Brownian motion and is a random number vector based on a standard normal distribution; the symbol It represents element-by-element multiplication, and the symbol Indicates that x is a vector; and The multiplication of simulates the Brownian motion of prey; P = 0.5 is a constant; is a vector of uniform random numbers in [0,1]; Stage 2: In the unit speed ratio stage, the speeds of predators and prey are almost equal. This happens in the middle third of the optimization process. Exploration tends to turn into exploitation. Prey uses Lévy motion to perform exploration, and predators use Brownian motion to perform exploitation. The population is divided into two, with half of the individuals used for exploration or exploitation. For the first half of the population (i=1,...,n / 2): in, is a random number vector generated by Lévy distribution, representing Lévy motion. The random number z based on Lévy distribution is generated by formula (17). In formula (17), the random number Q obeys a normal distribution with a mean of 0 and a standard deviation of s, that is, Q~N(0,s 2 ); the random number Y follows the standard normal distribution, Y~N(0,1); s is calculated by formula (18), α=1.5; Γ(·) represents the gamma function, Γ(1+α) equals α!, the Lévy motion of the prey is given by and The multiplication simulation, at the prey position Adding the step size above can simulate the movement of prey. The step size that follows the Lévy distribution is smaller, which can effectively search the nearby area in depth, thus facilitating exploitation. For the second half of the population (i=1+n / 2,...,n): Among them, CF is an adaptive parameter that controls the predator's moving step size. and The multiplication of is a means to simulate the movement of the predator in the Brownian mode, and the prey updates its position according to the predator. t is the current iteration number, and Max_Iter is the maximum iteration number; Phase 3: In the low speed ratio phase, the predator is faster than the prey, which is the development phase. This phase is achieved in the last third of the iterations of the optimization process. At low speed ratios, the best strategy for the predator is Lévy motion: (d) After each stage, the time-varying Sine transfer function TV is used Sin (·) Convert Prey to binary: first, Restricted to In scope: Among them, max(·) and min(·) represent the maximum and minimum values ​​of the returned boundary values, respectively. and Respectively represent the upper and lower bounds of the prey position vector, Through the time-varying Sine transfer function TV Sin (·) Convert to the range [0,1]: Mapping to binary vector in, is the transfer function TV Sin The control parameters of (·), from Gradually decays to The algorithm shifts from exploration to development as iterations proceed. rand represents a random vector between 0 and 1. (e) Calculate the fitness value of each individual according to the fitness function and update the top predator Best i ; (f) Introducing a mutation factor with the number of iterations t as a parameter to the top predator Best i Expand the adaptive mutation and discretize it, use the greedy strategy to compare the fitness values ​​of individuals before and after the mutation, and select the best individual as the top predator: Among them, Best i ′ is the top predator after mutation; (g) Performing ocean memory storage, which simulates the good memory of ocean predators. By comparing the fitness of the candidate solutions of the current iteration with the corresponding candidate solutions of the previous iteration, the poor candidate solutions are eliminated to simulate the memory of ocean predators; (h) Environmental issues are a factor that can cause changes in the behavior of marine predators, such as fish aggregating devices (FADs). Apply the FADs effect to update the predator matrix Prey: In the formula, FADs = 0.2, which is the probability that FADs affects the optimization process. is a binary vector, constructed by generating a random vector in [0,1]. If a random number is less than 0.2, the random number at the corresponding position is changed to 0. If a random number is greater than 0.2, the random number at the corresponding position is changed to 1. r is a random number uniformly distributed in [0,1]. and are the upper and lower bound vectors of the solution, r1 and r2 represent random indices of the prey matrix; (i) Based on the time-varying V-type transfer function TV V (·) Generate binary solution: where arctan(·) is the inverse tangent function; s and γ are TV V (·) control parameters; For the inversion operation, s is taken from s initial Gradually decays to s final ,γfromγ initial Gradually decays to γ final : s=s initial +t·(s final -s initial ) / Max_Iter (31) c = c initial +t·(γ final -c initial ) / Max_Iter (32) (j) Calculate the fitness value of each individual, detect the position of the top predator, and update the elite matrix Elite and related parameters; 4) Determine whether the current number of iterations t is equal to the maximum number of iterations Max_Iter. If so, output the current top predator position, i.e., the final association decision a=a (k+1) ; Otherwise, the number of iterations t is increased by 1 and returns to step 3) to enter the next iteration; Process (3) solves the resource allocation sub-problem based on the multi-strategy improved African vulture optimization algorithm: The user association decision a = a finally solved in process (2) (k+1) Substitute into P0, and the joint optimization problem P0 is transformed into P2: P2 is an optimization problem about the communication resource allocation decision b and the computing resource allocation decision f. Both b and f are continuous variables between 0 and 1. The African vulture optimization algorithm based on multi-strategy improvement is used to solve it. The solution steps are as follows: 1) Set algorithm-related parameters and initialize the population, randomly generate an available initial population, and use the Logistic-Cubic cascade chaotic sequence to generate a chaotic initial population; calculate the fitness value of each individual in the randomly initialized population and the chaotic population, and arrange them in order from small to large; select the first n most fit individuals as the initial population X0. Under the constraints, the fitness function of the optimization problem P2 is expressed as: fitness=U(b,f,a (k+1) ) (34) 2) Enter the iteration process: (a) The fitness value of each vulture individual is calculated according to the fitness function. The individual with the best fitness value is taken as the optimal vulture, and the second-best individual is taken as the second-best vulture. The optimal vulture and the second-best vulture guide the remaining individuals. The moving direction is expressed as: Among them, BestVulture1 represents the best vulture, BestVulture2 represents the second best vulture, L1 and L2 are preset constants, both between 0 and 1, and the sum is 1, p i From the roulette method: Among them, r i is a random number in the range [0,1], (b) Calculate the hunger level of each vulture: Where rand1 is a random number between [0, 1]; z is a random number between [-1, 1]; t is the current iteration number; Max_Iter is the maximum iteration number, and g is calculated by formula (38): Where h is a random number between [-2,2]; w is a pre-set parameter, usually set to 2.

5. (c) When |F i |≥1, the vulture individual enters the exploration stage, and the individual position update formula is expressed as: Among them, δ and parameter p1 are in the range of [0,1]. p1 is given at the time of initialization and is used to determine which exploration behavior the individual performs. rand p1 , rand2, and rand3 are all random numbers in the range [0,1]. i represents the distance between the current vulture and the optimal vulture or the suboptimal vulture, which is given by formula (41): D i =|C·R i -X i | (41) Where C is a random number uniformly distributed between [0,2]. (d) When |F i |<1, the vulture individual enters the development stage: 0.5<|F i When |<1, it enters the middle stage of development, and the position update formula is expressed as: Among them, rand p2 , rand4 are all random numbers in the range [0,1], and parameter p2 is in the range [0,1] and is given at initialization. When the vultures compete for food, The vulture performs circling flight behavior, and d is given by equation (43): d=R i -X i (43) S i1 and S i2 They are given by equations (44) and (45) respectively: Among them, rand5 and rand6 are both random numbers in the range of [0,1]; F i When |<0.5, it enters the late development stage, and the position update formula is expressed as: Among them, rand p3 is a random number in the range [0,1]; parameter p3 is in the range [0,1] and is given at initialization. When , vultures perform gathering behavior, and vice versa, The vulture performs the attack behavior, r1 and r2 are both random numbers in the range of [0,1]; α is a fixed constant, usually set to 1.5; s is given by formula (18); A i1 and A i2 They are given by equations (49) and (50) respectively: (e) Based on the greedy differential evolution mutation strategy, the fitness values ​​of the vulture individuals before and after the mutation are compared, and the positions of the vulture individuals are updated: (f) Determine whether the current number of iterations t is equal to the maximum number of iterations Max_Iter. If so, output the position and best score of the current optimal vulture, i.e., the final resource allocation decision b (k+1) 、f (k+1) and system cost U (k+1) ; Otherwise, the number of iterations t is increased by 1 and returns to step 2) to enter the next iteration; Process (4) obtains the current strategy π through the above steps (k+1) ={a (k+1) ,b (k+1) ,f (k+1) }, update π according to formula (53) (k+1) , if through π (k+1) The total cost U (k+1) Than the strategy π obtained in the previous iteration (k) The total cost U (k) Small, then keep the strategy π (k+1) As the current policy, otherwise the current policy π (k+1) Update to π (k) , determine whether k has reached the maximum number of iterations K of the set joint optimization method, and if so, output the final result π (k+1) , that is, the best association decision and resource allocation decision. If the maximum number of iterations is not reached, k = k + 1, and jump to process (2) for the next iteration; 4. The method for allocating network slice resources of the Internet of Vehicles based on the swarm intelligence optimization algorithm according to claim 3 is characterized in that: The solution step 1) in the process (2) and the solution step 1) in the process (3) both adopt the strategy of Logistic-Cubic cascade chaos to generate the cascade chaos initial population. The Logistic-Cubic cascade chaos mapping expression is as follows (13): x i+1 =ρ·[μ·x i ·(1-x i )]·(1-[μ·x i ·(1-x i )] 2 ) (13) Among them, x i ∈[0,1]; ρ and μ are control parameters. When x0=0.3,ρ=2.595,μ=4,Logistic-Cubic cascade chaos has good chaotic ergodicity. i+1,L =μx i (1-x i ) is the Logistic chaotic mapping expression, x i+1,C =ρx i (1-x i 2 ) is the Cubic chaos map expression.

5. The method for allocating network slice resources of Internet of Vehicles based on swarm intelligence optimization algorithm according to claim 3 is characterized in that: The specific process of solving step (f) in the process (2) and introducing the adaptive t distribution operator to perform mutation disturbance on the optimal individual is: Best i ′=Best i +Best i ·trnd(t) (26) Here, trnd(t) represents the t distribution.

6. The method for allocating network slice resources of Internet of Vehicles based on swarm intelligence optimization algorithm according to claim 3 is characterized in that: In the solution step (c) of the process (3), an adaptive trust δ is introduced to improve the exploration phase. The adaptive trust is defined as:

7. The method for allocating network slice resources of the Internet of Vehicles based on swarm intelligence optimization algorithm according to claim 3 is characterized in that: In the solution step (d) of the process (3), adaptive weight coefficients are introduced in the late development stage to improve the development stage. The adaptive weight coefficients θ1 and θ2 are defined as:

8. The method for allocating network slice resources of Internet of Vehicles based on swarm intelligence optimization algorithm according to claim 3 is characterized in that: In the solution step (e) of the process (3), based on the greedy differential evolution mutation strategy, the position of the vulture individual is updated, and the differential mutation individual X i,DE It is expressed as: X i,DE =X i +F i ·(X r1 -X r2 ) (51) Among them, X r1 , X r2 A vulture individual was randomly selected.

Citation Information

Cited By

  • Network dynamic optimization method and system for computer architecture

    CN119854849A

  • Distributed load flow calculation method and system for active power distribution network

    CN120341885A

  • Scenarized forwarding method and system for multi-source seismic service data

    CN121884548A