A method for large-scale multi-objective edge computing resource allocation that considers user satisfaction

By constructing a multi-objective edge computing resource allocation model in the Internet of Vehicles (IoV), and combining adaptive covariance matrix evolution and dual-weight analysis, resource allocation is optimized, solving the problem of insufficient vehicle computing resources. This achieves low-energy consumption, low-latency, and high-load-balanced resource allocation, improving user satisfaction and system stability.

CN117742965BActive Publication Date: 2026-11-17HEBEI UNIV OF TECH
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202311812930.2
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2023-12-27
Publication Date
2026-11-17
Estimated Expiration
2043-12-27

AI Technical Summary

Technical Problem

Vehicle computing resources are limited and cannot meet the time latency requirements of in-vehicle applications. The cloud computing center is far from the vehicle, resulting in large service transmission latency and reduced user satisfaction.

Method used

This paper proposes a large-scale multi-objective edge computing resource allocation method that takes user satisfaction into account. Through mathematical modeling, user satisfaction, time latency, energy consumption and load balancing are taken as objective functions. The resource allocation is optimized by using an adaptive covariance matrix evolution strategy and a dual-index selection strategy. Finally, the optimal allocation scheme is selected by combining the dual-weight analysis method.

Benefits of technology

While ensuring user satisfaction, it achieves minimal energy consumption and time latency, improves load balancing, balances the workload of cloud computing and edge computing resources, and enhances the system's privacy, scalability, and reliability.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN117742965B_ABST
    Figure CN117742965B_ABST
Patent Text Reader

Abstract

A large-scale multi-objective edge computing resource allocation method considering user satisfaction is proposed. First, user satisfaction, time latency, load balancing, and energy consumption are considered as objective functions to mathematically model a multi-objective resource allocation optimization model for the Internet of Vehicles (IoV), and a node association matrix is ​​constructed based on this model. Then, a dual-index selection strategy is used to improve the adaptive covariance matrix evolution strategy, obtaining a set of optimal resource allocation schemes for the IoV. Finally, a dual-weight analysis method is used to select suitable edge computing resource allocation schemes from this set of optimal resource allocation schemes. This invention seeks to achieve resource allocation with minimal energy consumption, minimal time latency, and high load balancing, thereby balancing the workload of cloud computing and edge computing resources, while ensuring user satisfaction. Combined with a uniform computing strategy based on small subpopulations, a resource allocation scheme that varies with computing resource costs is derived.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention belongs to the field of edge computing resource allocation, specifically a large-scale multi-objective edge computing resource allocation method that takes into account user satisfaction. Background Technology

[0002] With the rapid development of vehicle-to-everything (V2X) technology, in-vehicle software with high latency requirements is being widely used in V2X. However, vehicles have limited computing resources, which cannot meet the latency requirements of in-vehicle applications. Developing edge computing is an effective way to solve the problem of limited vehicle computing resources in V2X.

[0003] In the vehicle-to-everything (V2X) architecture, the cloud-based software-defined network (SDN) control node maintains the global network state and forwards different types of service rules and policies to the application-layer roadside unit (LSU) control nodes. This two-layer control model simplifies management and configuration and improves the system's real-time performance, stability, and scalability. When a vehicle cannot access the LSU control node due to coverage or bandwidth limitations, it can send a service request to a base station with wider coverage. The base station will then forward the received service request to a nearby edge server.

[0004] To meet the computing resource requirements of all vehicles in the entire connected vehicle architecture, it is essential to allocate resources rationally between cloud computing and edge computing. While cloud computing models can reduce energy consumption, they cannot meet the real-time demands of connected vehicle applications. To reduce data transmission and communication latency, connected vehicle devices can transmit computing tasks to edge nodes closer to them.

[0005] The resource allocation objective function is used to describe the following four issues: (1) reduce energy consumption while meeting the needs of vehicle networking services; (2) achieve load balancing in the allocation of resources in cloud computing and edge computing; (3) reduce the time delay in processing computing tasks; and (4) increase user satisfaction. Summary of the Invention

[0006] The purpose of this invention is to address the resource allocation problem in the Internet of Vehicles (IoV), where vehicles with limited computing resources often cannot handle large-scale computational tasks. Cloud computing is needed to alleviate this problem, but cloud computing centers are often located far from vehicles, resulting in significant service transmission latency and reduced user satisfaction. To solve this issue, this invention proposes a large-scale, multi-objective edge computing resource allocation method for IoV that considers user satisfaction as the objective function.

[0007] The technical solution of this invention:

[0008] A large-scale multi-objective edge computing resource allocation method considering user satisfaction is proposed. First, based on the mathematical model mentioned in the paper "Quantization and Deployment of Edge Servers Offloading Social Media Services in Industrial Internet of Things" by Xu X, Shen B, Yin X et al., load balancing is added as a fourth optimization objective, in addition to the three objectives of user satisfaction, time latency, and energy consumption. User satisfaction, time latency, load balancing, and energy consumption are considered as objective functions to mathematically model the multi-objective resource allocation optimization model for the Internet of Vehicles (IoV), and a node association matrix is ​​constructed based on the mathematical model. An adaptive covariance matrix evolution strategy, which generates the next generation by sampling in each generation, is improved using a dual-index selection strategy to obtain the optimal resource allocation scheme set for the IoV. Finally, a dual-weight analysis method is used to select a suitable edge computing resource allocation scheme from the optimal resource allocation scheme set for the IoV. The specific steps are as follows:

[0009] (1) Construct the objective function with the goal of minimizing energy consumption to reduce costs, task M i On server s j The energy consumption is calculated using the following formula:

[0010] P(s j M i )=λ×b(M i )×(V(s j )) 2

[0011] Among them, b(M) i ) represents task M i Size, V(s) j ) indicates server s j The computational speed, where λ represents the energy factor;

[0012] Using the energy factor λ, the total energy consumption is determined by the sum of the energy used to perform all computational tasks, which can be expressed mathematically as follows:

[0013]

[0014] Wherein, P(s) j ,*) indicates that the computation task is in s j Energy consumption on the server;

[0015] (2) Construct the objective function with the minimum time delay in executing the computational task, specifically:

[0016] In the Internet of Things (IoT), most data is processed and stored by edge computing clusters, thanks to their proximity to end users and negligible transmission latency. When tasks are sent to the cloud for processing, which is far from the user, the transmission time needs to be considered, using C... i The total service completion time is calculated as follows:

[0017]

[0018] in, The value is either 0 or 1, used to identify task C. i Is it on the edge server? j Up operation;

[0019]

[0020] When task M i On edge servers j During the above operation, Not on edge servers j During the above operation, S j =0; when task M i On cloud server j During the above operation, Not on cloud server j During the above operation, S j =0;

[0021] The total task delay is calculated as follows:

[0022] S t =max(S) j j = 1, 2, ..., m

[0023] (3) Construct the objective function with the goal of maximizing load balancing of computing resources, specifically:

[0024]

[0025] Average workload is calculated using the workload per server obtained from the following formula, expressed as follows:

[0026]

[0027] Where d represents the energy coefficient, l j Indicates server s j The workload;

[0028] The variance of server workload in the system is a metric used to evaluate the effectiveness of load balancing strategies, and it is calculated using the following formula:

[0029]

[0030] Load balancing index, denoted by f l This indicates that the average workload of servers performing computing tasks is used, expressed in l average express;

[0031] (4) Construct an objective function that maximizes user satisfaction with the expected time delay not exceeding the maximum tolerable time delay, specifically:

[0032]

[0033] Among them, the maximum service time delay that the user application can tolerate is denoted by k. i ×a i It means that k i ≥1 is a constant;

[0034] When the expected time delay t(M) i ,s j The time delay is not greater than the maximum tolerable time delay k. i ×a i hour, When the expected time delay t(M) i ,s j () greater than the maximum tolerable time delay k i ×a i At that time, user satisfaction dropped to 0;

[0035] Task M i Send to server s j The service time required for processing is denoted by t(M). i ,s j ) indicates; if from t(M i ,s j Subtract the expected processing time a from the input. i And taking the maximum value of this result and 0, the quantity obtained is [t(M i ,s j )-a i ] + =max{t(M i ,s j )-a i ,0};When the actual service time is less than or equal to the expected service time, this quantity is [t(M i ,s j )-a i ] + =0; in this case, user satisfaction S a (M i ,s j ) = 1, indicating complete satisfaction;

[0036] If the actual service time falls within [a] i ,k i ×a i Within a certain range, user satisfaction decreases as service time increases, with the degree of impact varying depending on factor k. i Decision; k i The value represents the tolerance range for the actual service time of a given task; when the actual service time exceeds k... i ×a i At the maximum tolerance level, user satisfaction drops to 0;

[0037] (5) Using the above objective function, a mathematical model containing i tasks and j server nodes was constructed. Next, the node association matrix of the system was constructed as follows:

[0038]

[0039] Let q j Let q represent the j-th individual in the population. ji This indicates that the i-th task is executed on the j-th node;

[0040] (6) For the node association matrix of the above-mentioned multi-objective resource allocation optimization model of the Internet of Vehicles, an improved algorithm based on the dual-index selection strategy is proposed, and the improved large-scale multi-objective optimization algorithm is used to solve the above four-objective optimization problem; the dual-index selection strategy combines the Pareto dominance relationship and the uniformity calculation strategy, specifically as follows:

[0041] The index E1(q) of the first part is calculated for each individual based on the Pareto dominance relationship. j ):

[0042]

[0043] Among them, the index E1(q) j ) represents an individual q j The number of times it is dominated by all other individuals;

[0044] Based on the uniformity calculation strategy, the second part of the index is calculated for further individual selection:

[0045]

[0046] Where L is a constant greater than 1; finally, the individual fitness is calculated, and the individuals with the highest fitness are selected and retained; the assessment of individual fitness is carried out using the following methods:

[0047] Fitness (q) j )=E1(q j )+E2(q j )

[0048] The value range of E1 is positive integer or 0, and E2 ∈ (0,1). Using this method, E1 is used to select individuals first. When E1 is equal, Pareto has no effect. Then E2 is used to select individuals.

[0049] (7) After obtaining the optimal resource allocation scheme set for the Internet of Vehicles (IoV) through a dual-index selection strategy, a dual-weight analysis method is used to select a final edge computing resource allocation scheme from the optimal resource allocation scheme set for the IoV. The dual-weight analysis method combines the first weight coefficient and the second weight coefficient to obtain the preference ranking calculation method, specifically:

[0050] First, determine the first weighting coefficient, Z. xy Let Z represent the importance of the x-th objective function to the y-th objective function, and Z... xy +Z yx =1, then calculate the first weight vector according to the following formula, and sort the weight vectors corresponding to each target:

[0051]

[0052] Then, determine the second weighting coefficient and establish a G. n×i The solution set of G, where n is the number of optimal solutions, i is the objective number, and g is the solution set of G. n×i A solution in the solution set, where g a×o This represents the o-th objective value in the a-th solution;

[0053] Before calculating the weight values, the solution is normalized according to the following formula:

[0054]

[0055] The importance of the normalized 0th objective value is calculated using the following formula:

[0056]

[0057] The second weighting coefficient for the 0th objective value is calculated using the following formula:

[0058]

[0059] Finally, to calculate the individual preference value, the first and second weighting coefficients are combined according to the following formula:

[0060] t o =θt 1x +(1-θ)t 2o

[0061] Wherein, θ is used to represent the difference between the first weighting coefficient and the second weighting coefficient;

[0062] Next, using the merged weight coefficients, construct the normalized matrix R = (r ao ) n×i ;

[0063] Where r ao =t o h ao

[0064] In evolutionary algorithms, when the objective value is as small as possible, the ideal solution is represented as:

[0065]

[0066] When the objective value is as large as possible, the ideal solution is expressed as:

[0067]

[0068] Therefore, R = (r ao ) n×i The distance to the ideal solution is:

[0069]

[0070] or

[0071] Calculate the preference values ​​for all individuals:

[0072]

[0073] The higher an individual's preference value, the higher their ranking. Based on the above formula, a suitable edge computing resource allocation scheme is selected from the set of optimal resource allocation schemes for the Internet of Vehicles.

[0074] The beneficial effects of this invention are:

[0075] This invention, while ensuring user satisfaction, seeks a resource allocation method that minimizes energy consumption, time latency, and load balancing to balance the workloads of cloud computing and edge computing resources. Combined with our proposed uniform computing strategy based on small subpopulations, a flexible resource allocation scheme that adapts to varying computing resource costs is derived. Furthermore, this framework employs a two-layer SDN control model, which offers several advantages, including:

[0076] 1) Enhanced privacy: Each SDN controller has its own management area, which makes it easy to implement different privacy policies in different areas.

[0077] 2) Increased scalability: When the number of devices managed by SDN exceeds a certain limit, systems using a single controller become very challenging to scale. Therefore, a two-layer controller model is able to deploy a fairly large network.

[0078] 3) Improved reliability: The two-layer control model utilizes multiple controllers, reducing the risk of single-point failures caused by the high concentration of controllers. Attached Figure Description

[0079] Figure 1 This is a flowchart illustrating the method of the present invention.

[0080] Figure 2 The experimental results for the four algorithms in terms of time delay are shown.

[0081] Figure 3 The experimental results of four algorithms in load balancing are presented.

[0082] Figure 4 The results show the energy consumption of the four algorithms.

[0083] Figure 5 The results of the experiments on user satisfaction for the four algorithms are shown. Detailed Implementation

[0084] The specific embodiments of the present invention will be further described below with reference to the accompanying drawings and technical solutions.

[0085] In the Internet of Vehicles (IoV), edge computing plays a crucial role in computing resources. IoV architectures frequently encounter conflicts such as time latency, energy consumption, and load balancing, making the rational allocation of computing resources essential. While cloud computing models can reduce energy consumption, they cannot meet the real-time requirements of IoV applications. To reduce data transmission and communication latency, IoV devices can transfer computing tasks to edge nodes closer to them. Therefore, combining cloud computing and edge computing to solve resource allocation problems is worthy of investigation. This invention aims to address how to allocate resources between edge computing and cloud computing.

[0086] The idea behind this invention is as follows: Since automotive computing and storage resources are limited, computing tasks need to be uploaded to cloud computing centers. However, cloud computing centers are typically far from the service requesters, resulting in long transmission delays. Therefore, edge servers are suitable for handling a portion of the task demands. The goal is to find a resource allocation method that minimizes energy consumption and time latency, and maximizes load balancing, while ensuring user satisfaction, thereby balancing the workload of cloud computing resources and edge computing resources.

[0087] See Figure 1This invention provides a method for allocating large-scale multi-objective edge computing resources in the Internet of Vehicles (IoV) that considers user satisfaction, comprising the following steps:

[0088] Construction of the S1 Vehicle-to-Everything (V2X) Model

[0089] The topology of the computing resource network is acquired to construct a software-defined vehicle-to-everything (V2X) model, including a cloud computing center, software-defined network control nodes, roadside units (Roadside Units), Roadside Unit control nodes, base stations, edge servers, vehicles, and users. Roadside Units collect and transmit road and vehicle service information to Roadside Unit control nodes for processing. Roadside Unit control nodes execute decision-making algorithms, making local decisions and allocating resources based on service types, reducing the burden on the software-defined network control nodes, and making real-time decisions near the end of the process. Furthermore, edge servers are deployed at the edge of the V2X network to support timely vehicle responses. The cloud-based software-defined network control nodes maintain the global network state and forward different types of service rules and policies to the application-layer Roadside Unit control nodes. This two-layer control model of Roadside Unit control nodes and software-defined network control nodes simplifies management and configuration and improves the system's real-time performance, stability, and scalability. When a vehicle cannot access a Roadside Unit control node normally due to coverage and bandwidth limitations, it can send service requests to base stations with wider coverage.

[0090] Construction of the S2 objective function

[0091] This invention mathematically models a multi-objective resource allocation optimization model for the Internet of Vehicles (IoV) by considering user satisfaction, time latency, load balancing, and energy consumption as objective functions. Under the premise of ensuring user satisfaction, it seeks a resource allocation method that minimizes energy consumption, achieves minimal time latency, and maximizes load balancing to balance the workload of cloud computing and edge computing resources.

[0092] (1) The optimization objective is to minimize energy consumption:

[0093] To achieve sustainable computing and optimize service quality while minimizing costs, prioritizing energy consumption as a metric for resource allocation is crucial. Therefore, it is necessary to calculate the energy consumption of collaborative resource allocation in a cloud context. Energy consumption is represented as follows:

[0094]

[0095] Where b(M) i ) represents task M i Size, V(s) j This indicates the server's computing speed.

[0096] The energy factor λ was used in the calculations. The total energy consumption is determined by the sum of the energy used to perform all computational tasks, and can be expressed mathematically as follows:

[0097]

[0098] (2) Optimization objective is to maximize user satisfaction:

[0099] In most IoT applications, service requests have an expected time delay and a maximum tolerable time delay requirement. If the actual service time is less than the expected time, user satisfaction is 100%. However, if the actual service time exceeds the expected time, users may still find it acceptable, but their satisfaction decreases as the delay increases. The maximum tolerable service time delay for a user application is denoted by k. i ×a i It means that k i ≥1 is a constant. If service time exceeds this maximum tolerance, user satisfaction will drop to 0. To evaluate resource allocation schemes, this chapter introduces a user satisfaction metric, calculated using the following formula:

[0100]

[0101] Task M i Send to server s j The service time required for processing is denoted by t(M). i ,s j This is represented by ). If we subtract the expected processing time ai from this value and take the maximum value of this result and 0, we get the quantity [t(M i ,s j )-a i ] + =max{t(M i ,s j )-a i According to formula (3.10), when the actual service time is less than or equal to the expected service time, this quantity is [t(M ,0}. i ,s j )-a i ] + =0. In this case, user satisfaction S a (M i ,s j =1, indicating complete satisfaction.

[0102] However, if the actual service time falls within [a i ,k i ×a i Within a certain range, user satisfaction decreases as service time increases, with the degree of impact varying depending on factor k.i Decision. k i The value represents the tolerance range for the actual service time of a given task. When the actual service time exceeds k... i ×a i At the maximum tolerance level, user satisfaction drops to 0.

[0103] (3) Optimize for maximum load balancing:

[0104] Load balancing resource allocation strategies can effectively balance the workload of computing resources, reduce task response time, and improve user satisfaction. In cloud edge collaboration systems, workloads should be distributed as evenly as possible to improve task response speed, rather than concentrating tasks on one or a few servers. Server workload can be calculated as follows:

[0105]

[0106] Average workload can be calculated using the workload per server obtained from the formula, which is expressed as follows:

[0107]

[0108] Where d represents the energy coefficient, l j Indicates server s j The workload;

[0109] The variance of server workload in a system is a metric used to evaluate the effectiveness of a load balancing strategy. It is calculated using the following formula:

[0110]

[0111] Load balancing index, denoted by f l This indicates that the average workload of servers performing computing tasks is used, expressed in l average express.

[0112] (4) Minimize time delay as the optimization objective:

[0113] Due to the limited computing and storage capabilities of in-vehicle equipment, computational tasks are often processed on servers. j Execute the task i Required time (T) i j )for:

[0114]

[0115] In the Internet of Things (IoT), most data is processed and stored by edge computing clusters, thanks to their proximity to end users and negligible transmission latency. However, when tasks are sent to the cloud for processing, which is far from the user, the transmission time of the tasks needs to be taken into account, using C... i The total computation time is expressed as follows:

[0116]

[0117] in, The value is either 0 or 1, used to identify task M. i Is it in s j Up operation.

[0118]

[0119] The total task delay is calculated as follows:

[0120] S t =max(S) j ), j = 1, 2, ..., k

[0121] Construction of S3 vehicle location and computing resource nodes

[0122] Based on mathematical modeling combined with vehicle location and computing resource nodes, a software-defined vehicle network system model is constructed.

[0123]

[0124] Let n represent the vehicle size and i represent the number of tasks; the first generation population is formed through random selection, with a total of n individuals; let q x Let q represent the x-th individual in the population. xy Let y represent the decision variable of the x-th individual.

[0125] Solving the S4 multi-objective optimization problem

[0126] Large-scale multi-objective optimization algorithms, such as Non-Dominant Sorting Genetic Algorithm (NSGA2), Covariance Matrix Adaptive Evolutionary Strategy (CMA-ES), and Bi-objective Evolutionary Algorithm (BiGE), can be used to solve the above-mentioned large-scale multi-objective edge computing resource allocation optimization model in the Internet of Vehicles that takes into account user satisfaction.

[0127] S5 combines the Pareto dominance relation and the uniformity calculation strategy to obtain a dual-index selection strategy. Employing this dual-index selection strategy improves the efficiency and convergence of large-scale multi-objective optimization algorithms in obtaining the optimal allocation scheme set.

[0128] S5.1 Part 1: Determination of Indicators

[0129] The first part is the index E1(p) i The calculation is based on Pareto dominance for each individual:

[0130]

[0131] Among them, the index E1(p) i ) represents an individual p i The number of times a group of diversity variables is dominated by all other individuals.

[0132] S5.2 Part Two: Determination of Indicators

[0133] The second part of the index, namely the estimation strategy based on uniformity calculation, can be used to further select individuals, and the formula is expressed as follows.

[0134]

[0135] In the formula, L is a constant greater than 1. Finally, the individual fitness is calculated, and the individuals with the highest fitness are selected and retained. The following methods are used to assess individual fitness:

[0136] Fitness (p) i )=E1(p i )+E2(p i )

[0137] The value of E1 is a positive integer or 0, and E2 ∈ (0,1). Using this method, we can first use E1 to select individuals. When E1 is equal, Pareto has no effect, and then use E2 to select individuals.

[0138] S6 employs a dual-weighted analysis method to select a suitable edge computing resource allocation scheme from the set of optimal resource allocation schemes. By combining the first and second weighting coefficients, the resulting decision scheme satisfies both the rational allocation of computing resources and improves the user experience.

[0139] S6.1 Determination of the first weighting coefficient

[0140] First, Z ij Let z represent the importance of the i-th objective function to the j-th objective function, and z ij +z ji =1, then calculate the first weight vector according to the following formula, and sort the weight vectors corresponding to each target:

[0141]

[0142] S6.2 Determination of the Second Weighting Coefficient

[0143] First, establish a Gn×m The solution set, where n is the number of optimal solutions and m is the objective number, where g i×j This represents the j-th objective value in the i-th solution.

[0144] Before calculating the weight values, the solution is normalized according to the following formula:

[0145]

[0146] The importance of the j-th objective value after normalization is calculated according to the following formula:

[0147]

[0148] Finally, the second weighting coefficient for the j-th objective value is calculated using the following formula:

[0149]

[0150] S6.3 Calculating Individual Preference Values

[0151] First, combine the first and second weighting coefficients according to the following formula:

[0152] t j =λw 1i +(1-λ)w 2j

[0153] λ can be used to change the different proportions of the first weighting coefficient and the second weighting coefficient.

[0154] Next, using the merged weight coefficients, construct the normalized matrix R = (r ij ) n×m

[0155] r ij =t j v ij

[0156] Finally, calculate the preference values ​​for all individuals:

[0157]

[0158] Example:

[0159] The simulation framework used in this embodiment includes a collaborative cloud-edge computing system consisting of five cloud servers and fifty edge servers. The computational tasks in the vehicle network are randomly generated, with a size ranging from [1, 10] × 10. 6 The instructions follow a uniform distribution within the interval. Considering a population size of 200, each algorithm is executed independently 30 times. Other experimental parameters are detailed in Table 1.

[0160] Table 1 Experimental parameters of resource allocation algorithm in the vehicle-to-everything (V2X) environment

[0161]

[0162]

[0163] First, the topology of the computing resource network is obtained, and a software-defined vehicle network model is constructed, including a cloud computing center, software-defined network control nodes, roadside units, roadside unit control nodes, base stations, edge servers, vehicles, and users.

[0164] A mathematical model for the multi-objective resource allocation optimization model of the vehicle-to-everything (V2) network is constructed using the four objective functions in S2. Then, based on vehicle locations and computing resource nodes, a software-defined V2X system model is built using S3. Next, n deployment schemes are randomly generated, and these n schemes will serve as the initial population for a large-scale multi-objective optimization algorithm.

[0165] These n deployment schemes are imported into the four objective functions of S2. A large-scale multi-objective optimization algorithm is used to solve for the objective function values ​​of these deployment schemes. Based on the index selection strategy, individual indexes are selected for crossover or mutation to generate offspring for the next iteration. After reaching the maximum number of iterations, the large-scale multi-objective optimization algorithm outputs an optimal deployment solution set, in which all solutions are non-dominated.

[0166] The optimal deployment solution set generated by S4 is used as input to S5, and the optimal deployment scheme is selected according to the dual-index selection strategy of S5. S6 is used to sort the optimal deployment schemes of S5 by preference. This method selects a flexible resource allocation scheme that can change with the computational resource cost from the optimal deployment solution set generated by the large-scale multi-objective optimization algorithm.

[0167] By comparing performance using digital examples, we demonstrate the effectiveness of the proposed algorithm in solving the resource allocation problem in vehicle-to-everything (V2X) networks. Experimental results for the four algorithms are shown below. Figure 2-5 It is worth noting that, Figure 2 and Figure 5 This indicates that the improved algorithm outperforms other algorithms in terms of time latency and user satisfaction. For example... Figure 4 The energy consumption results shown indicate that the increased energy consumption is acceptable given the significantly shorter service time of the improved algorithm. Regarding load balancing, Figure 3 The results show that the improved algorithm and the covariance matrix adaptive evolution strategy perform similarly when handling a small number of tasks, but the improved algorithm gradually outperforms the covariance matrix adaptive evolution strategy when handling a large number of tasks. Experimental results demonstrate the effectiveness of the improved algorithm in solving the resource allocation problem in vehicle-to-everything (V2X) networks.

Claims

1. A method for allocating large-scale, multi-objective edge computing resources while considering user satisfaction, characterized in that: First, user satisfaction, time delay, load balancing, and energy consumption are regarded as objective functions. Mathematical modeling is performed on the multi-objective resource allocation optimization model of the Internet of Vehicles, and a node association matrix is ​​constructed based on the mathematical modeling. The adaptive covariance matrix evolution strategy is an evolutionary strategy that generates the next generation of the population by sampling in each generation. This strategy is improved using a dual-index selection strategy to obtain a set of optimal resource allocation schemes for the Internet of Vehicles (IoV). Finally, a dual-weighted analysis method is used to select a suitable edge computing resource allocation scheme from this set of optimal resource allocation schemes. The steps are as follows: (1) Construct the objective function with the goal of minimizing energy consumption to reduce costs. On the server The energy consumption is calculated using the following formula: in, Indicates task Size, Indicates server The calculation speed Indicates energy factor; Utilizing energy factors The total energy consumption is determined by the sum of the energy used to perform all computational tasks, which can be expressed mathematically as follows: in, Indicates the computational task in Energy consumption on the server; (2) Construct the objective function with the minimum time delay in executing the computation task, specifically as follows: ; When a task is sent to the cloud for processing, which is far from the user, the transmission time of the task needs to be taken into account. The total service completion time is calculated as follows: in, The value is either 0 or 1, used to identify the task. Is it on the edge server? Up operation; When the task On the edge server During the above operation, Not on the edge server During the above operation, When the task On cloud servers During the above operation, Not on cloud server During the above operation, ; The total task delay is calculated as follows: (3) Construct the objective function to maximize the load balancing of computing resources, specifically: Average workload is calculated using the workload per server obtained from the following formula, expressed as follows: in, Indicates the energy coefficient. Indicates server The workload; The variance of server workload in the system is a metric used to evaluate the effectiveness of load balancing strategies, and it is calculated using the following formula: Load balancing index, using This indicates the average workload of servers performing computing tasks. express; (4) Construct the objective function with the goal of maximizing user satisfaction by ensuring that the expected time delay is no greater than the maximum tolerable time delay, as follows: Among them, the maximum service time delay that user applications can tolerate is used express, It is a constant; When the expected time delay No greater than the maximum tolerable time delay hour, When the expected time delay Greater than the maximum tolerable time delay At that time, user satisfaction dropped to 0; The task Send to server The service time required for processing, using Indicates; if from Subtract the expected processing time from the middle. And take the maximum value of this result and 0, the resulting quantity is When the actual service time is less than or equal to the expected service time, this quantity is... In this case, user satisfaction This indicates complete satisfaction; If the actual service time falls on Within a certain range, user satisfaction decreases as service time increases, with the degree of impact varying depending on the factors. Decide; The value represents the tolerable range of actual service time for a given task; when the actual service time exceeds... At the maximum tolerance level, user satisfaction drops to 0; (5) Based on the above objective function, a mathematical model containing i tasks and j server nodes was constructed. Next, the node association matrix of the system was constructed as follows: let Let j represent the j-th individual in the population. This indicates that the i-th task is executed on the j-th node; (6) For the node association matrix of the above-mentioned vehicle-to-everything (V2X) multi-objective resource allocation optimization model, an improved algorithm based on a dual-index selection strategy is proposed, and the improved large-scale multi-objective optimization algorithm is used to solve the above four-objective optimization problem. The dual-index selection strategy combines Pareto dominance and uniformity calculation strategies, specifically: The first part of the index is calculated for each individual based on the Pareto dominance relationship. : Among them, the indicator index Representing an individual The number of times it is dominated by all other individuals; Based on the uniformity calculation strategy, the second part of the index is calculated for further individual selection: Where L is a constant greater than 1; finally, the individual fitness is calculated, and the individuals with the highest fitness are selected and retained; the assessment of individual fitness is carried out using the following methods: The value of is a positive integer or 0. Using this method, first use When making individual choices, If they are equal, Pareto will not work at this point. Then use... Make individual choices; (7) After obtaining the optimal resource allocation scheme set for the Internet of Vehicles (IoV) through a dual-index selection strategy, a dual-weight analysis method is used to select a final edge computing resource allocation scheme from the optimal resource allocation scheme set for the IoV. The dual-weight analysis method combines the first weight coefficient and the second weight coefficient to obtain the preference ranking calculation method, specifically: First, determine the first weighting coefficient. This indicates the importance of the x-th objective function to the y-th objective function, and Then, the first weight vector is calculated according to the following formula, and the weight vectors corresponding to each target are sorted: Then, determine the second weighting coefficient and establish a... The solution set, where n is the number of optimal solutions and i is the objective number. yes A solution in the solution set, where This represents the o-th objective value in the a-th solution; Before calculating the weight values, the solution is normalized according to the following formula: The importance of the normalized 0th objective value is calculated using the following formula: The second weighting coefficient for the 0th objective value is calculated using the following formula: Finally, to calculate the individual preference value, the first and second weighting coefficients are combined according to the following formula: in, Used to represent the difference between the first weighting coefficient and the second weighting coefficient; Next, using the merged weight coefficients, construct a normalized matrix according to the following formula. ; in In evolutionary algorithms, when the objective value is as small as possible, the ideal solution is represented as: When the objective value is as large as possible, the ideal solution is expressed as: so The distance to the ideal solution is: or Calculate the preference values ​​for all individuals: The higher an individual's preference value, the higher their ranking. Based on the above formula, a suitable edge computing resource allocation scheme is selected from the set of optimal resource allocation schemes for the Internet of Vehicles.