A method, device and storage medium for optimizing mobile edge computing offloading delay

Abstract

The present invention discloses a method, device and storage medium for optimizing mobile edge computing offloading delay. The method comprises: using a hierarchical analysis model to comprehensively calculate and evaluate the task delay response priority of users in a communication system to obtain a priority weight; constructing a task processing time model to calculate the task processing time; using the priority weight and the task processing time to establish a problem model of an optimization target; based on the problem model of the optimization target, obtaining a global optimal target through iteration, and generating an optimal channel selection coefficient and an offloading coefficient parameter combination adopted by each user when processing a task. The present invention effectively avoids network congestion and data accumulation, fully utilizes the computing resources of heterogeneous edge servers while ensuring the transmission of priority users, and effectively optimizes and reduces system delay.

Description

Technical Field

[0001] The present invention relates to a method, device and storage medium for optimizing mobile edge computing offloading delay, and belongs to the technical field. Background Art

[0002] Task offloading means that users offload tasks that cannot be processed to the data center for processing. Resource allocation mainly refers to the allocation of wireless channels, wired channels, server computing and cache resources. Task offloading is closely related to the resource allocation problem. After making a task allocation decision, the reasonable allocation of resources must be considered. For example, when tasks are offloaded, how many wireless channels or time slots are allocated in the wireless access network to transmit them, how much bandwidth does the wired channel need to allocate in the backhaul network, and how much computing resources are finally allocated to the data center to process the task. These all need to be planned reasonably to effectively utilize resources, and the reasonable allocation of resources determines the efficiency of task execution and the user's service experience. In the process of data diversion, the transmission bandwidth depends on the minimum bandwidth of multiple paths. In general, wired transmission is easy to expand the bandwidth capacity by laying optical fibers, and it will not become a limiting factor in the transmission rate. However, wireless resources are relatively scarce and can easily become a bottleneck that limits the transmission speed. Summary of the invention

[0003] The purpose of the present invention is to overcome the shortcomings of the prior art and provide a mobile edge computing offloading delay optimization method, device and storage medium to effectively avoid network congestion and data accumulation, while ensuring priority user transmission, fully utilize the computing resources of heterogeneous edge servers, and effectively optimize and reduce system delays.

[0004] To achieve the above object, the present invention is implemented by adopting the following technical solutions:

[0005] In a first aspect, the present invention provides a method for optimizing mobile edge computing offloading latency, comprising:

[0006] The hierarchical analysis model is used to comprehensively calculate and evaluate the task delay response priority of users in the communication system to obtain the priority weight;

[0007] Build a task processing time model and calculate the task processing time;

[0008] Using the priority weights and task processing time, a problem model of an optimization goal is established;

[0009] Based on the problem model of optimization objectives, the global optimal objective is obtained through iteration, and the optimal channel selection coefficient and offloading coefficient parameter combination adopted by each user when processing tasks is generated.

[0010] Furthermore, the use of the hierarchical analysis model to comprehensively calculate and evaluate the task delay response priority of users in the communication system to obtain the priority weight includes:

[0011] Establishing the analytic hierarchy process model;

[0012] The judgment matrix is ​​constructed by comparing the importance of each factor at the same level with respect to a criterion at the previous level.

[0013] After constructing the judgment matrix, it is necessary to verify from a quantitative perspective whether it is significantly different from the consistent matrix;

[0014] Calculate the weight of each indicator and use the arithmetic mean method to find the weight;

[0015] The weight vectors of all criterion-scheme layer judgment matrices are combined into a new matrix, and then the task priority weight vector is calculated.

[0016] Furthermore, the task processing time model is constructed to calculate the task processing time, and the formula is as follows:

[0017]

[0018] Among them, t n represents the task processing time, d n represents the amount of task data, α represents the channel selection coefficient, k n represents the percentage of time slots occupied by user n when unloading to the base station, C S and C M denote the channel capacities of SBS and MBS respectively, β denotes the offloading coefficient, ζ S and M Respectively represent the time required for SBS and MBS to transmit a unit of data volume, and They represent the CPU frequencies allocated to user n by the corresponding servers in SBSm, MBS and cloud computing center respectively, where n represents the user number and m represents the MBS number.

[0019] Furthermore, the priority weight and task processing time are used to establish a problem model of the optimization target, and the formula is as follows:

[0020]

[0021] stC1:0≤α n,m ≤1, n∈N, m∈M

[0022] C2:

[0023] C3:

[0024] C4:

[0025] C5:

[0026] Among them, w n represents the priority weight of user n in the system per unit time in the model, t n represents the task processing time, C1 and C2 are the range constraints of the user channel selection coefficient and the offloading coefficient, indicating that one user can only be served by one node, but one node can serve multiple users; C3 represents the constraint on user delay, and the processing time cannot exceed the maximum tolerated completion time; C4 and C5 indicate that the wireless channel resources and node computing resources are limited, B k and C k They represent node channel resources and computing resources respectively.

[0027] Furthermore, based on the problem model of the optimization target, the global optimal target is obtained through iteration, and the optimal channel selection coefficient and unloading coefficient parameter combination adopted by each user when processing the task is generated, including:

[0028] Step 1: Determine that the population is a set of feasible solutions. The population consists of chromosomes. The rows where the chromosomes are located represent the corresponding service nodes. The number of chromosomes should be M+2. The genes on the chromosomes are binary numbers, 0 represents missing, and 1 represents the existence of the gene. The columns where the genes are located correspond to the users in the system. Therefore, the length of each chromosome, that is, the number of genes, is N;

[0029] Step 2: Calculate the population fitness value. First, calculate the fitness of each chromosome:

[0030]

[0031] Where N is the total number of terminal devices, δ ij Indicates whether the gene exists, and then calculates the fitness of the entire population:

[0032]

[0033] In this algorithm, the smaller the fitness value is set, the better its fitness is, the more adaptable the population is to survival, and the more likely it is to become the global optimum;

[0034] Step 3: Select excellent populations in order to prepare chromosomes for mating in the next step. The roulette method is used to select the population, and the probability of the population being selected is defined as where F i Represents the fitness value of the ith population. Repeat this operation to obtain two populations, retain the chromosome with the highest parental fitness, put it in the next population of the offspring, and select the chromosome in the same row in another population;

[0035] Step 4: Use single-point crossover to randomly select a crossover point, and perform interchromosomal gene swapping on the two chromosome parts selected in Step 3 before and after the crossover point to produce new offspring.

[0036] Step 5: Randomly obtain the crossover points on the chromosome and change the gene values ​​above.

[0037] Step6: Repeat Step2-5 until the number of evolutionary iterations is reached.

[0038] Furthermore, it also includes: for service requests issued by user nodes in heterogeneous networks within a unit time, while ensuring priority user transmission, obtaining the minimum average processing time of all tasks in the system model, and solving the optimal solution corresponding to the minimum optimization target.

[0039] In a second aspect, the present invention provides a mobile edge computing offloading latency optimization device, comprising:

[0040] The first calculation unit is used to use a hierarchical analysis model to comprehensively calculate and evaluate the task delay response priority of users in the communication system to obtain a priority weight;

[0041] The second calculation unit constructs a task processing time model and calculates the task processing time;

[0042] The problem model building unit of the optimization target uses the priority weight and the task processing time to build the problem model of the optimization target;

[0043] The generation unit is used to obtain the global optimal target through iteration based on the problem model of the optimization target, and to generate the optimal channel selection coefficient and unloading coefficient parameter combination adopted by each user when processing the task.

[0044] In a third aspect, the present invention provides a mobile edge computing offloading latency optimization device, including a processor and a storage medium;

[0045] The storage medium is used to store instructions;

[0046] The processor is configured to operate according to the instructions to execute the steps of any of the aforementioned methods.

[0047] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein the program, when executed by a processor, implements the steps of any of the aforementioned methods.

[0048] Compared with the prior art, the present invention has the following beneficial effects:

[0049] The present invention provides a method, device and storage medium for optimizing the offloading delay of mobile edge computing. According to the resource allocation optimization theory under the mobile edge computing application and in combination with the heterogeneous network model architecture, by formulating reasonable task allocation decisions and resource allocation strategies, namely, by using a hierarchical analysis model and an improved genetic algorithm, the computing offloading strategy is optimized to effectively avoid network congestion and data accumulation, and while ensuring the transmission of priority users, the computing resources of heterogeneous edge servers are fully utilized to effectively optimize and reduce system delay. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 It is a model diagram of heterogeneous network system;

[0051] Figure 2 It is the diagram of the hierarchical analysis model;

[0052] Figure 3 It is a schematic diagram of the genetic algorithm population;

[0053] Figure 4 It is the flow chart of genetic algorithm;

[0054] Figure 5 is the population fitness distribution diagram;

[0055] Figure 6 is the population average processing time evolution curve;

[0056] Figure 7 This is the curve of average task processing time changing with the number of users;

[0057] Figure 8 is the curve of average task processing time changing with the number of task cycles;

[0058] Fig. 9 This is the curve of average task processing time changing with the average task data volume. DETAILED DESCRIPTION

[0059] The present invention will be further described below in conjunction with the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and cannot be used to limit the protection scope of the present invention.

[0060] Example 1

[0061] The present invention provides a method for optimizing mobile edge computing offloading latency, comprising:

[0062] The hierarchical analysis model is used to comprehensively calculate and evaluate the task delay response priority of users in the communication system to obtain the priority weight;

[0063] Build a task processing time model and calculate the task processing time;

[0064] Using the priority weights and task processing time, a problem model of an optimization goal is established;

[0065] Based on the problem model of optimization objectives, the global optimal objective is obtained through iteration, and the optimal channel selection coefficient and offloading coefficient parameter combination adopted by each user when processing tasks is generated.

[0066] The contents designed in the above embodiment are described below in conjunction with a preferred embodiment.

[0067] The heterogeneous network system model used in this embodiment is as follows: Figure 1 As shown in the figure, the heterogeneous system consists of N user terminal devices, M small-cell base stations (SBS), 1 macro-cell base station (MBS) and 1 cloud computing center. MEC servers are deployed on both SBS and MBS to provide computing services to users. The cloud computing center has larger-scale computing equipment, which can provide powerful computing capabilities and centralized operation and maintenance management, and provide users with rich network and computing services. The terminal devices are represented by a set as N = 1, 2, ..., N, n∈N, where N is the total number of terminal devices and n represents the number of each user device. The small-cell base station is represented by a set as M = 1, 2, ..., M, m∈M, where M is the total number of SBSs and m represents the number of each SBS.

[0068] Assume that each device has a delay-sensitive or computationally intensive task to complete in unit time, so consider offloading the computation to a nearby base station or server in a cloud computing center. n It is expressed as: Among them, d n represents the amount of task data, λ n The number of CPU cycles required to execute the task, Indicates the maximum completion time tolerated by the task.

[0069] For the network model used by user devices, the time division multiple access (TDMA) system is considered here for wireless network communication. This system allows multiple user devices to use the same frequency in different time slots, so each device can use the time slot allocated by the base station to unload data. Considering the heterogeneity of device access, the terminal device is connected to the SBS or MBS wirelessly, while the SBS and MBS, and the MBS and the cloud computing center are connected by wire. In order to save costs, no wired connection is provided between the SBS and the cloud computing center. When the computing power of the terminal device is insufficient, the device can offload the task to the MEC server on the nearby SBS by wireless, and return the result data to the original device after calculation; or forward it to the MBS via the SBS, and then calculate it through the MEC server on the MBS and return it by the original route; in addition, the data is sent to the cloud computing center with stronger computing power for calculation through two forwardings of the SBS and MBS, and then the processing results are returned to the device. Generally speaking, the amount of data sent by users is much larger than the amount of data transmitted back by the link. Therefore, when considering the delay, the delay impact caused by the downlink data can be ignored. This model studies the uplink, that is, considering the channel data transmission time between the user and the base station and the execution time on each server.

[0070] According to the above system model and its related theories, a method for optimizing the mobile edge computing offloading latency for heterogeneous networks is provided. The present invention is divided into the following four steps:

[0071] Step 1: Use the hierarchical analysis model to comprehensively calculate and evaluate the task delay response priority of users in the communication system to obtain the weight.

[0072] In order to ensure the quality of experience (QoE) of different users, the Ayalytic Hierarchy Process (AHP) model is used to score the priority of all users appearing in the system within a unit time, which can solve complex decision-making problems with multiple indicators. It specifically includes the following five steps:

[0073] (1) Establish a hierarchical structure.

[0074] The hierarchical analysis model is shown in Figure Figure 2 As shown in the figure, the hierarchy is divided into three layers: 1. The target layer is the evaluation target, which is to distinguish the task priorities; 2. The criterion layer is the evaluation indicator selection task The three parameters contained in n∈N are used as indicators, namely, the indicator S 1 The corresponding task data volume d n 、Indicator S 2 The number of CPU cycles required by the corresponding server to process the task λn and index S 3 The maximum completion time tolerated by the corresponding task 3. The optional plans at the plan layer are the tasks that need to be unloaded in the system within a unit time. Through this model, the proportion of each task compared to the overall task priority can be obtained. The larger the proportion, the higher the priority, and vice versa.

[0075] (2) Construct a judgment matrix.

[0076] The importance of each factor at the same level to a criterion at the previous level is compared to construct a judgment matrix. It is difficult to directly give the weights assigned to each factor in the matrix, so the importance of the comparison between different factors is given, expressed by scale a. Table 1 gives the scale value judgment standard. This step first constructs the target layer-criterion layer judgment matrix A = (a ij ) 3×3 Then, we construct the criterion layer-scheme layer judgment matrix B corresponding to the three indicators k =(a ij ) n × n ,n∈N,k∈S 1 ,S 2 ,S 3 ,in

[0077]

[0078] Table 1 Scale value judgment standard

[0079]

[0080] (3)Consistency check

[0081] After constructing the judgment matrix, it is necessary to verify from a quantitative perspective whether it is significantly different from the consistent matrix. The general steps of consistency testing are divided into two steps:

[0082] 1) Calculate the consistency index CI:

[0083]

[0084] Among them, λ max It represents the maximum value of the eigenvalue of the judgment matrix, and n represents the order of the judgment matrix, that is, the number of indicators.

[0085] 2) Calculate the consistency ratio CR:

[0086]

[0087] Among them, RI is the average random consistency index, and the corresponding relationship is as follows in Table 2:

[0088] Table 2 Correspondence between RI and n

[0089]

[0090] If RI is less than 0.1, the consistency check is passed. If it is not passed, the elements in the judgment matrix need to be modified until they pass the verification. Here, the consistency check needs to be performed on the target layer-criterion layer judgment matrix A and the criterion layer-scheme layer judgment matrix B respectively.

[0091] (4) Calculate the weight of each indicator. The weight is calculated using the arithmetic mean method, which includes three steps:

[0092] 1) Normalize the judgment matrix by column.

[0093] 2) Add the normalized elements of the same row in each column.

[0094] 3) Divide each element of the vector obtained after addition by n to obtain the weight vector. S1 =(a ij ) nn For example, the weight vector obtained is in It is expressed as:

[0095]

[0096] According to the above calculation method, the weight vectors of the remaining two criterion layer-scheme layer judgment matrices can be obtained: And the target layer-criterion layer judgment matrix A = (a ij ) 3×3 The weight vector Λ=[λ S1 ,λ S2 ,λ S3 ] T .

[0097] (5) Calculate task priority.

[0098] The weight vectors of all criterion-scheme layer judgment matrices are combined into a new matrix Δ = (W S1 W S2 W S3 ), and then calculate the task priority weight vector W. The process can be expressed as:

[0099]

[0100] Among them, w n represents the priority weight of user n in the system per unit time in the model, w n The larger the value, the higher the task priority and the better the QoE.

[0101] Step 2: Construct a task processing time model, including channel transmission time and server execution time models.

[0102] Assuming that all tasks in the scenario of the present invention need to be offloaded to the MEC cloud computing center for calculation, define 0≤k n ≤1,n∈N represents the percentage of time slots occupied by user n when unloading to the base station. Define a channel selection coefficient α n,m ∈0,1,n∈N,m∈M represents the channel selection decision of user n, where n represents the user terminal number, N represents the terminal device set; m represents the SBS number, and M represents the SBS set. When user n transmits a task to SBSm, α n,m =1; when the user transfers the task to MBS, α n = 1. Due to the indivisibility of the task, the user can only select one of the two to the SBS or MBS through the wireless link. Therefore, for user n, its channel selection coefficient satisfies the constraint:

[0103]

[0104] Where M is the total number of SBSs.

[0105] (1) Channel transmission time

[0106] When α n,m = 1, considering the delay of user n accessing SBSm through the wireless channel, assuming that the bandwidth of all SBSs is the same, that is, B S According to Shannon's formula, the channel capacity C of SBS is obtained S :

[0107]

[0108] in, represents the channel gain between SBS and user n, P S represents the SBS forward link transmission power, σ 2 represents the Gaussian white noise power of the wireless channel, then the transmission rate between user n and SBS It is expressed as:

[0109]

[0110] Among them, k n Indicates the percentage of time slots occupied by user n when unloading to the base station, 0≤k n ≤1,n∈N. Then the wireless transmission delay connected to the SBS uplink is It is expressed as:

[0111]

[0112] Similarly, consider the latency of user n directly accessing the MBS, assuming that the bandwidth of all MBSs is the same, B M According to Shannon's formula, we can get the channel capacity C of MBS M :

[0113]

[0114] in, represents the channel gain between MBS and user n, P M represents the MBS forward link transmission power, then the transmission rate between user n and MBS It is expressed as:

[0115]

[0116] The wireless transmission delay connected to the MBS uplink can be obtained It is expressed as:

[0117]

[0118] Next, consider the delay of the wired channel transmission task. Since the SBS and MBS, MBS and cloud computing center are connected through wired media such as optical fiber, the signal interference is greatly reduced compared to wireless. Therefore, this model assumes that the transmission rate in the wired channel is a relatively large value. When calculating the wired transmission delay, it is considered to be proportional to the task data volume. Then the wired transmission delay generated by the task of transmitting user n from SBS to MBS is It is expressed as:

[0119]

[0120] Among them, S is a constant, which represents the time required by SBS to transmit a unit of data. Similarly, the wired transmission delay generated by the task of transmitting user n from MBS to the cloud computing center can be obtained as It is expressed as:

[0121]

[0122] Among them, M It is a constant, which indicates the time required for MBS to transmit a unit amount of data.

[0123] (2) Server execution time

[0124] Define an unloading factor represents the uninstallation execution decision of user n, SB m, MB, CE represent the servers deployed on SBSm, MBS, and cloud computing centers respectively. When the coefficient value is 1, it means that the task is offloaded to the server on the corresponding node to execute the task; the coefficient value is 0, which means that it is not executed on the server of this node. Since the task cannot be split, it can only be offloaded to one server for execution, and each task is successfully offloaded. Therefore, for each user n, the following constraints must be met:

[0125]

[0126] Defining tasks The execution time on the server is expressed as:

[0127]

[0128] in, and Represents the task τ n Corresponding server execution time on SBSm, MBS and cloud computing center, λ n Indicates the number of CPU cycles required to execute a task. and They represent the CPU frequencies allocated to user n by the corresponding servers in SBS m, MBS and cloud computing center respectively.

[0129] (3) Task processing time

[0130] According to the different task transmission paths and execution endpoints, this model proposes five different types of task offloading modes. n represents the user number, and m represents the SBS number.

[0131] 1) When the channel selection coefficient α n,m =1 and unloading coefficient When , it means that the task is directly unloaded on SBSm through the wireless link. This mode is named mode 1, and the task processing time calculation expression is:

[0132]

[0133] The processing time consists of two parts: the wireless transmission delay from the terminal to the SBS uplink and MEC server execution time on SBSm

[0134] 2) When the channel selection coefficient α n,m =1 and unloading coefficient When , it means that after the task reaches SBS through the wireless link, it reaches MBS through the wired link and is finally unloaded on MBS. This mode is named mode 2, and its task processing time calculation expression is:

[0135]

[0136] The processing time consists of three parts: the wireless transmission delay from the terminal to the SBS uplink Wired transmission delay of user n tasks from SBS to MBS and MEC server execution time on MBS

[0137] 3) When the channel selection coefficient α n =1 and unloading coefficient When , it means that the task is directly unloaded on the MBS through the wireless link. This mode is named mode 3, and the task processing time calculation expression is:

[0138]

[0139] The processing time consists of two parts: the wireless transmission delay from the terminal to the MBS uplink and MEC server execution time on SBS m

[0140] 4) When the channel selection coefficient α n =1 and unloading coefficient When , it means that after the task reaches MBS through the wireless link, it reaches the cloud computing center through the wired link and is finally unloaded to the cloud computing center. This mode is named mode 4, and its task processing time calculation expression is:

[0141]

[0142] The processing time consists of three parts: the wireless transmission delay from the terminal to the MBS uplink Wired transmission delay of user n tasks from MBS to cloud computing center and the execution time of MEC servers in the cloud computing center

[0143] 5) When the channel selection coefficient α n,m =1 and unloading coefficient When , it means that after the task reaches SBS through the wireless link, it reaches MBS through the wired link, is forwarded to the cloud computing center through the wired link again, and is finally unloaded from the cloud computing center. This mode is named mode 5, and its task processing time calculation expression is:

[0144]

[0145] The processing time consists of four parts: the wireless transmission delay from the terminal to the SBS uplink Wired transmission delay of user n tasks from SBS to MBS Wired transmission delay of user n tasks from MBS to cloud computing center and the execution time of MEC servers in the cloud computing center

[0146] In order to ensure the uniqueness of mode selection, combined with the channel selection coefficient and the unloading coefficient, the task processing time obtained by the above five modes can be summarized as:

[0147]

[0148] And because the channel selection coefficient satisfies the constraint And the unloading coefficient satisfies the constraint The above formula can be expanded and simplified to

[0149]

[0150] Step 3: Combine the hierarchical analysis model and the task processing time model to establish a problem model of the optimization goal.

[0151] The problem model that the present invention intends to solve is: for the service requests issued by user nodes in the heterogeneous network within a unit time, while ensuring the transmission of priority users, the minimum value of the average processing time of all tasks in the system model is obtained, and the optimal solution corresponding to the minimum optimization target is solved, that is, the optimal channel selection coefficient and unloading coefficient parameter combination adopted by each user when processing tasks is obtained. The problem function is expressed as:

[0152]

[0153] stC1:0≤α n,m ≤1, n∈N, m∈M

[0154] C2:

[0155] C3:

[0156] C4:

[0157] C5:

[0158] Among them, C1 and C2 are the range constraints of user channel selection coefficient and offloading coefficient, indicating that one user can only be served by one node, but one node can serve multiple users; C3 represents the constraint on user delay, and the processing time cannot exceed the maximum tolerated completion time; C4 and C5 indicate that wireless channel resources and node computing resources are limited. k and C kThey represent node channel resources and computing resources respectively.

[0159] Step 4: Use the improved genetic algorithm to solve the load balancing scheduling problem, obtain the global optimal target through iteration, and generate the optimal channel selection coefficient and unloading coefficient parameter combination for each user to process the task.

[0160] This problem can be regarded as a load balancing scheduling problem. This problem is non-convex, that is, there will be many local optimal solutions in the global scope. In order to find the global optimal solution in the global scope, the present invention uses an improved genetic algorithm (GA) to solve it. Compared with the traditional genetic algorithm, it is an Indian function that considers the priority transmission of tasks according to weights. The parameter set of the problem includes: task priority weight matrix, task cycle number matrix, task data volume matrix, node processing speed matrix, and node processing time matrix. The algorithm flow is as follows:

[0161] Step 1: Population initialization. Determine the population as a set of feasible solutions, set the population size to 100, the population is composed of chromosomes, the rows of chromosomes represent the corresponding service nodes, the number of chromosomes should be M+2, the genes on the chromosomes are binary numbers, 0 represents missing, 1 represents the existence of the gene, the column where the gene is located corresponds to the user in the system, so the length of each chromosome, that is, the number of genes, is N. The population diagram is as follows Figure 3 shown.

[0162] Step 2: Calculate the population fitness value. First calculate the fitness of each chromosome:

[0163]

[0164] Where N is the total number of terminal devices, δ ij Indicates whether the gene exists, and then calculates the fitness of the entire population:

[0165]

[0166] In this algorithm, the smaller the fitness value is set, the better its fitness is, the more adaptable the population is to survival, and the more likely it is to become the global optimum.

[0167] Step 3: Selection. The purpose of selecting excellent populations is to prepare chromosomes for mating in the next step. The roulette method is used to select populations, and the probability of a population being selected is defined as where F i Represents the fitness value of the ith population. Repeat this operation to obtain two populations, retain the chromosome with the highest parental fitness, put it in the next population of the offspring, and select the chromosome in the same row of another population.

[0168] Step 4: Crossover. Use single-point crossover to randomly select a crossover point, and swap the two chromosome parts selected in Step 3 before and after the crossover point to produce new offspring. The crossover probability is set to 0.6.

[0169] Step 5: Mutation. Randomly obtain the crossover points on the chromosome and change the gene values ​​above. In order to prevent the algorithm from converging to the local optimal solution too quickly, the mutation probability is set to 0.1.

[0170] Step6: Repeat Step2-5 until the number of evolutionary iterations is reached.

[0171] The above algorithm process is as follows Figure 4 As shown in the figure, the feasible solution corresponding to the population with the minimum fitness value is the global optimal solution, that is, the decision on the channel selection coefficient and the offloading coefficient for the user task that minimizes the average task processing time of the system while ensuring the transmission of priority users.

[0172] To further illustrate the effectiveness of the solution proposed in the present invention, the present invention has conducted simulation verification, and the simulation tools used include Python and MATLAB. In the simulation, the number of user terminals is set to N = 1000, the total number of SBSs is set to M = 8, and the total frequency of MEC servers on all SBSs is The total frequency f of MEC servers on MBS MB =80GHz,,total frequency of cloud computing center server f CE = 200GHz. The number of task cycles that need to be offloaded follows a normal distribution with an average value of 0.5 Gigacycles, and the amount of task data follows a normal distribution with an average value of 18MB. Figure 5 The distribution diagram of the fitness values ​​of 100 populations as the number of iterations changes is given. It can be seen that with the increase of the number of iterations, the chromosome fitness is getting better and better, and the total task processing time gradually tends to converge. Figure 6 The evolution curve of the population average processing time is given. It can be seen that with the increase of the number of iterations, the average task processing time value is getting smaller and smaller, and when the number of iterations is 80, the system performance tends to be stable. In order to verify the superiority of the proposed scheme in mobile edge computing latency optimization, a comparison with two other heuristic algorithms is given, Particle Swarm Optimization (PSO) and Ant Colony Optimization (ACO). The number of users, the average number of task cycles, and the amount of task data are changed by the control variable method. The simulation comparison is as follows: Figure 7 , Figure 8 and Fig. 9As shown in the figure, it can be seen that the genetic algorithm proposed in this paper is superior to the two traditional heuristic algorithms in performance. In addition, in order to verify the necessity of processing task priorities, the traditional genetic algorithm that does not consider task priorities is also compared with the improved genetic algorithm in this paper. Figure 7 It can be seen that with the increase in the number of users, the performance of the improved genetic algorithm proposed in the present invention is always better than that of the traditional genetic algorithm, and when the number of users exceeds 1400, the average task processing time of users increases faster and the performance begins to deteriorate, indicating the advantage of the algorithm proposed in the present invention in optimizing the latency of offloading tasks by using edge computing when processing large-scale user groups.

[0173] Example 2

[0174] This embodiment provides a mobile edge computing offloading latency optimization device, including:

[0175] The first calculation unit is used to use a hierarchical analysis model to comprehensively calculate and evaluate the task delay response priority of users in the communication system to obtain a priority weight;

[0176] The second calculation unit constructs a task processing time model and calculates the task processing time;

[0177] The problem model building unit of the optimization target uses the priority weight and the task processing time to build the problem model of the optimization target;

[0178] The generation unit is used to obtain the global optimal target through iteration based on the problem model of the optimization target, and to generate the optimal channel selection coefficient and unloading coefficient parameter combination adopted by each user when processing the task.

[0179] Example 3

[0180] This embodiment provides a mobile edge computing offloading latency optimization device, including a processor and a storage medium;

[0181] The storage medium is used to store instructions;

[0182] The processor is used to operate according to the instructions to execute the steps of the method according to any one of Embodiment 1.

[0183] Example 4

[0184] This embodiment provides a computer-readable storage medium on which a computer program is stored, characterized in that when the program is executed by a processor, the steps of any one of the methods described in Embodiment 1 are implemented.

[0185] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the technical principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A method for optimizing mobile edge computing offloading latency, characterized in that: include: The hierarchical analysis model is used to comprehensively calculate and evaluate the task delay response priority of users in the communication system to obtain the priority weight; Build a task processing time model and calculate the task processing time; Using the priority weights and task processing time, a problem model of an optimization goal is established; Based on the problem model of the optimization target, the global optimal target is obtained through iteration, and the optimal channel selection coefficient and unloading coefficient parameter combination adopted by each user when processing the task is generated; The task processing time model is constructed to calculate the task processing time. The formula is as follows: Among them, t n represents the task processing time, d n represents the amount of task data, α represents the channel selection coefficient, k n represents the percentage of time slots occupied by user n when unloading to the base station, C S and C M denote the channel capacities of SBS and MBS respectively, β denotes the offloading coefficient, ζ S and M Respectively represent the time required for SBS and MBS to transmit a unit of data volume, and They represent the CPU frequencies of SBSm, MBS and the corresponding servers in the cloud computing center allocated to user n, n represents the user number, and m represents the MBS number; SB m , MB, and CE represent the servers deployed in SBSm, MBS, and cloud computing centers, respectively; n Indicates the number of CPU cycles required to execute a task; The problem model of the optimization target is established by using the priority weight and task processing time. The formula is as follows: Among them, w n represents the priority weight of user n in the system per unit time in the model, t n represents the task processing time, C1 and C2 are the range constraints of the user channel selection coefficient and the offloading coefficient, indicating that one user can only be served by one node, but one node can serve multiple users; C3 represents the constraint on user delay, and the processing time cannot exceed the maximum tolerated completion time; C4 and C5 indicate that the wireless channel resources and node computing resources are limited, B k and C k They represent node channel resources and computing resources respectively; n represents the user terminal number, N represents the terminal device set; m represents the SBS number, and M represents the SBS set.

2. The mobile edge computing offloading latency optimization method according to claim 1 is characterized in that: The use of the hierarchical analysis model to comprehensively calculate and evaluate the task delay response priority of users in the communication system to obtain the priority weight includes: Establishing the analytic hierarchy process model; The judgment matrix is ​​constructed by comparing the importance of each factor at the same level with respect to a criterion at the previous level. After constructing the judgment matrix, it is necessary to verify from a quantitative perspective whether the consistency test has passed; Calculate the weight of each indicator and use the arithmetic mean method to find the weight; The weight vectors of all criterion-scheme layer judgment matrices are combined into a new matrix, and then the task priority weight vector is calculated.

3. The mobile edge computing offloading latency optimization method according to claim 1, characterized in that: Based on the problem model of the optimization target, the global optimal target is obtained through iteration, and the optimal channel selection coefficient and unloading coefficient parameter combination adopted by each user when processing the task is generated, including: Step 1: Determine that the population is a set of feasible solutions. The population consists of chromosomes. The rows where the chromosomes are located represent the corresponding service nodes. The number of chromosomes should be M+2. The genes on the chromosomes are binary numbers, 0 represents missing, and 1 represents the existence of the gene. The columns where the genes are located correspond to the users in the system. Therefore, the length of each chromosome, that is, the number of genes, is N; Step 2: Calculate the population fitness value. First, calculate the fitness of each chromosome: Where N is the total number of terminal devices, δ ij Indicates whether the gene exists, and then calculates the fitness of the entire population: In this algorithm, the smaller the fitness value is set, the better its fitness is, the more adaptable the population is to survival, and the more likely it is to become the global optimum; Step 3: Select excellent populations in order to prepare chromosomes for mating in the next step. The roulette method is used to select the population, and the probability of the population being selected is defined as where F y Represents the fitness value of the yth population. Repeat this operation to obtain two populations. Keep the chromosome with the highest parental fitness and put it into the next population after the offspring. And select the chromosome in the same row and another population. Step 4: Use single-point crossover to randomly select a crossover point, and perform interchromosomal gene swapping on the two chromosome parts selected in Step 3 before and after the crossover point to produce new offspring; Step 5: Randomly obtain the crossover points on the chromosome and change the gene values ​​above; Step6: Repeat Step2-5 until the number of evolutionary iterations is reached.

4. The mobile edge computing offloading latency optimization method according to claim 1, characterized in that: Also includes: For the service requests sent by user nodes in heterogeneous networks within a unit time, while ensuring the transmission of priority users, the minimum average processing time of all tasks in the system model is obtained, and the optimal solution corresponding to the minimum optimization target is solved.

5. A mobile edge computing offloading delay optimization device, characterized in that: include: The first calculation unit is used to use a hierarchical analysis model to comprehensively calculate and evaluate the task delay response priority of users in the communication system to obtain a priority weight; The second calculation unit constructs a task processing time model and calculates the task processing time; The problem model building unit of the optimization target uses the priority weight and the task processing time to build the problem model of the optimization target; A generation unit, which is used to obtain the global optimal target through iteration based on the problem model of the optimization target, and to generate the optimal channel selection coefficient and unloading coefficient parameter combination adopted by each user when processing the task; The task processing time model is constructed to calculate the task processing time, and the formula is as follows: Among them, t n represents the task processing time, d n represents the amount of task data, α represents the channel selection coefficient, k n represents the percentage of time slots occupied by user n when unloading to the base station, C S and C M denote the channel capacities of SBS and MBS respectively, β denotes the offloading coefficient, ζ S and M Respectively represent the time required for SBS and MBS to transmit a unit of data volume, and They represent the CPU frequencies of SBSm, MBS and the corresponding servers in the cloud computing center allocated to user n, n represents the user number, and m represents the MBS number; SB m , MB, and CE represent the servers deployed in SBSm, MBS, and cloud computing centers, respectively; n Indicates the number of CPU cycles required to execute a task; The problem model of the optimization target is established by using the priority weight and task processing time. The formula is as follows: Among them, w n represents the priority weight of user n in the system per unit time in the model, t n represents the task processing time, C1 and C2 are the range constraints of the user channel selection coefficient and the offloading coefficient, indicating that one user can only be served by one node, but one node can serve multiple users; C3 represents the constraint on user delay, and the processing time cannot exceed the maximum tolerated completion time; C4 and C5 indicate that the wireless channel resources and node computing resources are limited, B k and C k They represent node channel resources and computing resources respectively; n represents the user terminal number, N represents the terminal device set; m represents the SBS number, and M represents the SBS set.

6. A mobile edge computing offloading delay optimization device, characterized in that: including processor and storage medium; The storage medium is used to store instructions; The processor is used to operate according to the instructions to execute the steps of the method according to any one of claims 1 to 4.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of the method according to any one of claims 1 to 4 are implemented.

Images