Joint optimization method, device and medium for task offloading and resource allocation in 5G ultra-dense networks

Abstract

The present invention aims to solve the high latency problem in 5G ultra-dense networks caused by users with insufficient computing power processing low-latency, high-reliability applications, and implements a joint optimization strategy for offloading strategy and resource allocation under the condition of limited computing resources and channel resources. First, a system model of MEC and local computing in a 5G ultra-dense network scenario is constructed, and a mixed integer nonlinear optimization problem that minimizes task completion time is constructed. Then, a joint optimization strategy for task offloading decision and resource allocation is proposed for the optimization problem. This strategy first uses variable substitution to simplify the problem, and then solves it by decomposing the sub-problems. The original problem is decomposed into two sub-problems: computing resource allocation and channel resource allocation. The Lagrange multiplier method is first used to obtain the optimal solution for computing resources, and then a channel resource allocation algorithm based on the idea of differential evolution is used to perform channel resource allocation under the condition of determining the optimal solution for each computing resource allocation.

Description

Technical Field

[0001] The present invention relates to a joint optimization method for task offloading and resource allocation in a 5G ultra-dense network, and belongs to the technical field. Background Art

[0002] With the rapid development and advancement of the mobile internet and the Internet of Things (IoT), global mobile data traffic is experiencing explosive growth. Global mobile data traffic in the 5G era is expected to reach over 1,000 times that of the 4G era. Furthermore, mobile devices are becoming increasingly intelligent. These devices are driving a variety of emerging services, including autonomous driving, real-time interactive gaming, virtual reality, and augmented reality. These emerging services place high demands on network latency. For example, some real-time interactive gaming and autonomous driving require latency of less than 1ms. However, existing fourth-generation mobile communication systems (4G) can only offer a maximum latency of 20ms-70ms, far from meeting these requirements. Furthermore, these emerging applications place high demands on user devices (UEs). While current UE CPUs offer robust computing power, these computationally intensive applications consume significant energy, which is incompatible with the battery life of UEs, significantly reducing their battery life.

[0003] To address these challenges, traditional cloud computing is no longer able to meet the needs of the explosive growth of massive data. In 2014, the European Telecommunications Standards Institute (ETSI) launched the LTE-based edge computing standards project, Mobile Edge Computing (MEC), and began standardization work. According to ETSI, MEC is defined as follows: "Mobile edge computing refers to the provision of IT service environments and cloud computing capabilities at the edge of mobile networks, within the Radio Access Network (RAN), and near mobile users." Its initial purpose was to reduce network latency and alleviate network congestion. With the continuous expansion of services and deepening research, the meaning of mobile edge computing has been further expanded to multi-access edge computing (MEC), with further enhanced capabilities and access technologies not limited to LTE but also encompassing 5G, Wi-Fi, fixed networks, and other access technologies.

[0004] With the massive growth of mobile devices and the access of densely deployed network equipment, including wireless access points (APs), small base stations (SBSs), and macro base stations (MBSs), network data traffic is trending towards hotspots, placing extremely high demands on network capacity. Traditionally, increasing spectrum bandwidth and improving spectrum reuse are two important approaches to improving network capacity. However, due to the limited spectrum resources within a network, increasing bandwidth alone cannot significantly increase network capacity. Ultra-Dense Networks (UDNs) are expected to be an effective solution, significantly improving network spectrum utilization and capacity by reusing limited spectrum resources between cells.

[0005] Therefore, UDN can meet the access needs of massive UEs (User Equipment) and improve network capacity. MEC can process computing-intensive and data-intensive tasks in real time. The combination of UDN and MEC can provide more terminals with instant computing capabilities, meeting the computing power requirements of tasks while avoiding the disadvantage of high latency caused by offloading tasks to remote clouds. However, in UDN, a large number of UEs sharing limited channel resources will cause serious interference, resulting in a decrease in transmission rate. The allocation of channel resources and computing resources is particularly important for MEC systems, with the former affecting the data transmission rate and the latter affecting the computing delay of tasks. Therefore, there is an urgent need to invent a joint optimization method for task offloading and resource allocation in 5G ultra-dense networks that can effectively reduce the task completion delay of the entire system. Summary of the Invention

[0006] The purpose of the present invention is to overcome the deficiencies in the prior art and provide a joint optimization method, device and medium for task offloading and resource allocation in 5G ultra-dense networks, which can minimize the task delay of all terminals and effectively reduce the task completion delay of the entire system.

[0007] To achieve the above object, the present invention is implemented by adopting the following technical solutions:

[0008] In a first aspect, the present invention provides a joint optimization method for task offloading and resource allocation in a 5G ultra-dense network, the method comprising the following steps:

[0009] Establish an OFDMA-based ultra-dense network scenario consisting of a macro base station and multiple small base stations, where MEC is deployed in the macro base station to support multi-user access;

[0010] Obtain basic network information, including: local terminal computing power, average task computing amount, data volume, MEC computing power, number of channels, and channel bandwidth; calculate the computing delay of the user-selected task locally and offloaded to the MEC based on the basic network information, and calculate the system model and formula for the user's transmission rate, the transmission delay when the task is offloaded, and the total delay when the task is offloaded to the MEC server for processing based on the Shannon formula;

[0011] Based on the ultra-dense network scenario, system model and formula, an optimization objective function for task offloading and resource allocation in 5G ultra-dense networks is proposed;

[0012] According to the relationship between the offloading decision variables and the channel resource allocation variables, the variables are replaced to simplify the optimization objective function. The simplified optimization objective function is then decomposed into subproblems to obtain the computing resource allocation problem and the channel resource allocation problem.

[0013] The computing resource allocation problem and channel resource allocation problem are solved respectively, and the channel resource allocation results and the total task completion delay are obtained and output.

[0014] Furthermore, in a multi-cell, multi-user UDN scenario, based on the different resources required for different tasks, the optimization objective function for offloading decision X, channel allocation C, and MEC computing resource allocation F under resource constraints can be expressed as:

[0015]

[0016] st(C1)

[0017] (C2)

[0018] (C3)

[0019] (C4)

[0020] (C5)

[0021] (C6)

[0022] (C7)

[0023] Where X is the offloading decision matrix, C is the channel allocation matrix, F is the MEC computing resource allocation matrix, S = {1, 2, ..., S} represents the set of SBSs, N = {1, 2, ..., N s}, s∈S represents the set of UEs served by each SBS, represents the set of offloading users, K = {1, 2, ..., K} represents the set of subchannels; represents the total completion delay of the task, Indicates the UE's offloading decision, if Indicates a task Offload computing to edge servers, Indicates a task Process locally; Indicates the channel allocation value, Represents UEn s Use subchannel k for offloading, Represents UEn s Subchannel k is not used for offloading; Indicates that MEC is UEn s Allocated computing resources; f m Indicates the maximum computing capacity of MEC; Indicates a task The maximum tolerable delay.

[0024] Furthermore, the present invention adopts a method of performing variable replacement based on the relationship between the offloading decision variable and the channel resource allocation variable to simplify the optimization objective function, and decomposing the simplified optimization objective function into subproblems, including the following steps:

[0025] according to The relationship between , we get the following equation:

[0026]

[0027] Where K = {1, 2, ..., K} represents the set of subchannels, and K represents the number of subchannels; Indicates the UE's offloading decision, if Indicates a task Offload computing to edge servers, Indicates a task Process locally; Indicates the channel allocation value, Represents UEn s Use subchannel k for offloading, Represents UEn s Subchannel k is not used for offloading;

[0028] Get the new objective function P1:

[0029]

[0030] st(C1)

[0031] (C2)

[0032] (C3)

[0033] (C4)

[0034] (C5)

[0035] (C6)

[0036] Where C is the channel allocation matrix, F is the MEC computing resource allocation matrix, S = {1, 2, ..., S} represents the set of SBSs, and N = {1, 2, ..., N s}, s∈S represents the set of UEs served by each SBS, represents the set of offloading users, K = {1, 2, ..., K} represents the set of subchannels; Indicates the total completion delay of the task; Indicates the channel allocation value, Represents UEn s Use subchannel k for offloading, Represents UEn s Subchannel k is not used for offloading; Indicates that MEC is UEn s Allocated computing resources; f m Indicates the maximum computing capacity of MEC; Indicates a task The maximum tolerable delay;

[0037] Transform problem P into problem P1 of allocating MEC computing resources and channel resources. If the user terminal chooses to reuse channel resources, the computing task is offloaded to the MEC server, and the MEC server allocates the corresponding computing resources for task computing. If the terminal does not choose to reuse channel resources, the computing task is performed locally on the terminal.

[0038] Using the indirect optimization method, problem P1 is divided into computing resource allocation problem and channel resource allocation problem for solution.

[0039] Furthermore, methods for solving the computing resource allocation problem and the channel resource allocation problem include:

[0040] The computing resource allocation problem is converted into a convex function about the allocation of server computing power. The Lagrangian function and KKT condition are introduced to obtain the optimal computing resource allocation value.

[0041] According to the optimal computing resource allocation value, the channel resource allocation algorithm based on differential evolution is used to solve the channel resource allocation problem, obtain the optimal channel resource allocation, and output the channel resource allocation result and the total task completion delay.

[0042] Furthermore, the computing resource allocation problem is converted into a convex function about the allocation of server computing power. The Lagrangian function and KKT condition are introduced to obtain the optimal computing resource allocation value, which includes the following steps:

[0043] The task Offload the calculation to the MEC server and give the channel allocation scheme C 0 ,at this time The set of UEs selected for offloading by each SBS is expressed as Then the original objective function is transformed into The convex function P1(1):

[0044]

[0045] stC2,C3

[0046] in Indicates that under a given channel allocation scheme C 0 Then we can get the uplink transmission rate of UEs; find f(C * ,F)About The Hessian matrix is:

[0047]

[0048] Each item of the matrix is expressed as:

[0049]

[0050] Construct about f(C 0 ,F)’s Lagrangian function:

[0051]

[0052] Where μ is the Lagrange multiplier corresponding to the constraint condition C3, and λ,μ≥0; the optimal value of the improved problem can be obtained by the Karush-Kuhn-Tucker (KKT) condition; using the KKT condition, the optimal value The conditions that μ must satisfy are:

[0053]

[0054] Using the KKT condition, the optimal solution can be obtained for:

[0055]

[0056] According to the optimal solution Substituting into problem P1(1), we can get the optimal allocation of MEC computing resources f(C 0 ,F * ):

[0057]

[0058] Where C 0 Represents the given channel allocation matrix, F * represents the calculated optimal MEC computing resource allocation matrix, S = {1, 2, ..., S} represents the set of SBSs, Indicates the collection of uninstalled users; f m Indicates the maximum computing capacity of MEC; Indicates completion of the task The total number of CPU cycles required; Indicates a task The amount of data; Represents a given channel allocation matrix C 0 The task after Uplink transmission delay.

[0059] Furthermore, according to the optimal computing resource allocation value, a channel resource allocation algorithm based on differential evolution is used to solve the channel resource allocation problem, including the following steps:

[0060] N individuals are randomly and uniformly generated within the solution vector value range to form the initial population. The population set is represented by X i ={x i,1 ,x i,2 ,...,x i,j}, where i = 1, 2, ..., N represents the serial number of the individual in the population, j = 1, 2, ..., D (D = N s ×S×K) represents the dimension of the solution vector; then the value of the j-th dimension of the i-th individual is as follows:

[0061] x i,j =rand(0,1)(U i,j -L i,j )+L i,j

[0062] Among them L i,j represents the lower bound of the j-th dimension of the i-th solution vector, U i,j Indicates the upper bound of the j-th dimension of the i-th solution vector; rand(0,1) indicates that the random value is "0" or "1";

[0063] The differential evolution algorithm randomly selects two different individuals in the population, scales their difference, and adds other individuals to generate new mutant individuals; in the mth iteration, the mutant individual V i (m) is as follows:

[0064] V i (m) = F(X p (m-1)-X q (m-1))+X k (m-1)

[0065] where X p (m-1), X q (m-1), X k (m-1) is the three different individual solution vectors of the generated population in the parent generation, and F is the scaling factor; in the mutation operation, it is necessary to determine whether each individual in the population meets the boundary conditions, that is, L i,j ≤v i,j (m)≤U i,j ;

[0066] Exchange the weight of each individual in the parent population with the weight of the corresponding mutant individual. The operation is as follows:

[0067]

[0068] Where randi(1,n) represents a randomly selected integer from 1 to n, ensuring that at least one gene in the crossover individual is provided by the corresponding individual in the parent generation. CR is the crossover probability, which affects the convergence speed of the algorithm and the richness of the population. A larger CR is conducive to global optimization, while a smaller CR is conducive to achieving local optimality.

[0069] Get the fitness function:

[0070]

[0071] in is the penalty function, which is expressed as follows:

[0072]

[0073] in and is a penalty factor greater than zero;

[0074] After mutation and crossover operations, a group of crossover individuals are obtained. These individuals are compared with the corresponding individuals in the parent population by entering the fitness function. Following the greedy strategy, the next generation parent individuals with high fitness are retained. The greedy selection strategy is as follows:

[0075]

[0076] Among them F * (m) is the optimal MEC computing resource allocation matrix corresponding to the mth generation of individuals; the selection operation ensures that the new generation of population is better than the previous generation of population, eliminates the poor individuals, retains the original excellent individuals, and guides the algorithm to approach the optimal solution.

[0077] In a second aspect, the present invention provides a joint optimization device for task offloading and resource allocation in a 5G ultra-dense network, comprising:

[0078] Network establishment module: used to establish an OFDMA-based ultra-dense network scenario consisting of a macro base station and multiple small base stations, where MEC is deployed in the macro base station to support multi-user access;

[0079] Modeling module: used to obtain basic network information, including: local terminal computing power, average task computing amount, data volume, MEC computing power, number of channels, and channel bandwidth; based on this basic network information, calculate the computing delay of the user-selected task locally and offloaded to the MEC, and calculate the system model and formula for the user's transmission rate, the transmission delay when the task is offloaded, and the total delay of the task offloaded to the MEC server for processing based on the Shannon formula;

[0080] Optimization function module: used to propose the optimization objective function for task offloading and resource allocation in 5G ultra-dense networks based on ultra-dense network scenarios, system models, and formulas;

[0081] Decomposition module: used to replace variables based on the relationship between offloading decision variables and channel resource allocation variables to simplify the optimization objective function, and then decompose the simplified optimization objective function into sub-problems to obtain computing resource allocation problems and channel resource allocation problems;

[0082] Solution output module: used to solve the computing resource allocation problem and the channel resource allocation problem respectively, obtain and output the channel resource allocation results and the total task completion delay.

[0083] Furthermore, the method for the solution output module to solve the computing resource allocation problem and the channel resource allocation problem respectively includes:

[0084] The computing resource allocation problem is converted into a convex function about the allocation of server computing power. The Lagrangian function and KKT condition are introduced to obtain the optimal computing resource allocation value.

[0085] According to the optimal computing resource allocation value, the channel resource allocation algorithm based on differential evolution is used to solve the channel resource allocation problem, obtain the optimal channel resource allocation, and output the channel resource allocation result and the total task completion delay.

[0086] In a third aspect, the present invention provides a joint optimization device for task offloading and resource allocation in a 5G ultra-dense network, including a processor and a storage medium;

[0087] The storage medium is used to store instructions;

[0088] The processor is configured to operate according to the instructions to execute the steps of the method according to the first aspect.

[0089] In a fourth aspect, the present invention provides 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 described in the first aspect are implemented.

[0090] Compared with the prior art, the present invention has the following beneficial effects:

[0091] 1. This invention addresses the high latency problem in 5G ultra-dense networks caused by users with insufficient computing power processing low-latency, high-reliability applications. Under the conditions of limited computing resources and channel resources, it implements a joint optimization strategy of offloading strategy and resource allocation.

[0092] 2. The present invention constructs a system model of MEC and local computing in a 5G ultra-dense network scenario, constructs a mixed integer nonlinear optimization problem to minimize the task completion time, and then proposes a joint optimization strategy for task offloading decision-making and resource allocation for the optimization problem. This strategy first uses variable replacement to simplify the problem, and then solves it by decomposing the sub-problems. The original problem is decomposed into two sub-problems: computing resource allocation and channel resource allocation. The Lagrange multiplier method is first used to obtain the optimal solution for computing resources, and then a channel resource allocation algorithm based on the idea of differential evolution is used to perform channel resource allocation under the condition of determining the optimal solution for each computing resource allocation. Simulation results show that the algorithm proposed in the present invention can effectively reduce the total task completion delay of the system under the condition of limited resources, thereby improving the user's service experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0093] Figure 1 It is a system model diagram of the present invention;

[0094] Figure 2 This is a comparison chart of the total completion delay of different algorithms under different numbers of tasks;

[0095] Figure 3 This is a comparison chart of the total latency of different algorithms at different task computational loads. DETAILED DESCRIPTION

[0096] 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 solutions of the present invention and are not intended to limit the scope of protection of the present invention.

[0097] Example 1:

[0098] The technical problem to be solved by the present invention is the task completion delay problem based on edge computing in 5G ultra-dense networks. It mainly solves the resource allocation and offloading strategy problems faced by user devices when offloading tasks to edge servers.

[0099] Consider a UDN consisting of a macro base station (MBS) and N small base stations (SBS). The set of SBSs is represented by S = {1, 2, ..., S}. A MEC server is deployed in the macro base station. The MEC server has a certain computing power and can process multiple computing tasks in parallel. The set of UEs served by each SBS is represented by N = {1, 2, ..., N s}, s∈S. The UE and MEC server are connected through a wireless channel, and Orthogonal Frequency Division Multiple Access (OFDMA) is used as the user's multiple access method. The bandwidth is divided into K orthogonal sub-channels. The set of sub-channels is represented by K = {1, 2, ..., K}. The bandwidth of each sub-channel is represented by w. Each UE has a computing task to be processed, and each task is indivisible. The specific analysis is as follows:

[0100] 1) Local computing model

[0101] Assume that represents a computing task, where Indicates completion of the task The total number of CPU cycles required, Indicates the amount of data for the task, It represents the maximum time delay that the task can tolerate. The UE's offloading decision is expressed as like Indicates a task Will be offloaded to the edge server through the channel, Indicates a task Process locally. When computing tasks When calculating on the local terminal: the total delay is the local processing time of the task. Assume is the CPU frequency of the local terminal, that is, the CPU cycles executed per unit time, then the task The calculation delay on the local device is:

[0102]

[0103] 2) MEC computing model

[0104] when Indicates that the task is offloaded to MEC for calculation. At this time, it is mainly divided into two stages: the first stage is the task data upload stage, and the second stage is the task execution stage. In the first stage, the UE offloads the calculation to MEC through the wireless channel. s When using subchannel k for offloading, the channel allocation binary variable on the contrary Considering the interference caused by frequency reuse, UEn s Signal-to-interference-and-noise ratio on subchannel k It can be expressed as:

[0105]

[0106] in Represents UEn s The uplink transmit power, represents the channel gain, σ 2 represents the Gaussian noise of the channel, represents the inter-cell interference caused by other UEs in the channel. Then according to Shannon’s formula, we can get UEn s The rate of uplink transmission through subchannel k is:

[0107]

[0108] Then, we can get UEn s Uplink transmission rate and task when offloading tasks Uplink transmission delay:

[0109]

[0110]

[0111] In the second stage, when the task After unloading to the MEC server, the delay of this part is the task execution delay of the MEC server. The set of tasks unloaded to the MEC server in SBS is defined as Assumptions Indicates that the MEC server is a task The allocated computing resources, then the task The computation delay on the MEC server is:

[0112]

[0113] After the MEC server completes the task, it will return the execution result to the UE. However, since the amount of data returned is usually small and can be ignored, the task Total latency of offloading to MEC server processing Expressed as:

[0114]

[0115] Then the task Total completion delay for:

[0116]

[0117] Based on the above system model and related theories, we propose a joint optimization strategy for task offloading and resource allocation in 5G ultra-dense networks. This invention is divided into the following steps:

[0118] Step 1: Establish an OFDMA-based ultra-dense network consisting of a macro base station and multiple small base stations, where MEC is deployed in the macro base station to support multi-user access;

[0119] Consider an ultra-dense network consisting of a macro base station and N small base stations (SBS). The set of SBSs is represented by S = {1, 2, ..., S}. A MEC server is deployed in the macro base station. The MEC server has a certain computing power and can process multiple computing tasks in parallel. The set of UEs served by each SBS is represented by N = {1, 2, ..., N s}, s∈S. The UE and MEC server are connected through a wireless channel, and Orthogonal Frequency Division Multiple Access (OFDMA) is used as the user's multiple access method. The bandwidth is divided into K orthogonal sub-channels. The set of sub-channels is represented by K = {1, 2, ..., K}. The bandwidth of each sub-channel is represented by w. Each UE has a computing task to be processed, and each task is indivisible.

[0120] Step 2: Calculate the computational latency of the user-selected task locally and when it is offloaded to the MEC. Based on the Shannon formula, calculate the user's transmission rate, the transmission latency when the task is offloaded, and the total latency of the task being offloaded to the MEC server for processing.

[0121] Assume that represents a computing task, where Indicates completion of the task The total number of CPU cycles required, Indicates the amount of data for the task, It represents the maximum time delay that the task can tolerate. The UE's offloading decision is expressed as like Indicates a task Will be offloaded to the edge server through the channel, Indicates a task Process locally. When computing tasks When calculating on the local terminal: the total delay is the local processing time of the task. Assume is the CPU frequency of the local terminal, that is, the CPU cycles executed per unit time, then the task The calculation delay on the local device is:

[0122]

[0123] when Indicates that the task is offloaded to MEC for calculation. At this time, it is mainly divided into two stages: the first stage is the task data upload stage, and the second stage is the task execution stage. In the first stage, the UE offloads the calculation to MEC through the wireless channel. s When using subchannel k for offloading, the channel allocation binary variable on the contrary Considering the interference caused by frequency reuse, UEn s Signal-to-interference-and-noise ratio on subchannel k It can be expressed as:

[0124]

[0125] in Represents UEn s The uplink transmit power, represents the channel gain, σ 2 represents the Gaussian noise of the channel, represents the inter-cell interference caused by other UEs in the channel. Then according to Shannon’s formula, we can get UEn s The rate of uplink transmission through subchannel k is:

[0126]

[0127] Then, we can get UEn s Uplink transmission rate and task when offloading tasks Uplink transmission delay:

[0128]

[0129]

[0130] In the second stage, when the task After offloading to the MEC server, the delay of this part is the task execution delay of the MEC server. The set of tasks offloaded to the MEC server in SBSs is defined as Assumptions Indicates that the MEC server is a task The allocated computing resources, then the task The computation delay on the MEC server is:

[0131]

[0132] After the MEC server completes the task, it will return the execution result to the UE. However, since the amount of data returned is usually small and can be ignored, the task Total latency of offloading to MEC server processing Expressed as:

[0133]

[0134] Then the task Total completion delay for:

[0135]

[0136] Step 3: Based on steps 1 and 2, an optimization objective function for task offloading and resource allocation in 5G ultra-dense networks is proposed;

[0137] The main goal of this invention is to minimize the task latency of all terminals. In a multi-cell, multi-user UDN scenario, based on the different resources required for different tasks, the optimization objective for offloading decision X, channel allocation C, and MEC computing resource allocation F under resource constraints can be expressed as:

[0138]

[0139] st(C1)

[0140] (C2)

[0141] (C3)

[0142] (C4)

[0143] (C5)

[0144] (C6)

[0145] (C7)

[0146] Among them, constraint C1 indicates that each task can be executed locally or offloaded to the MEC server; constraint C2 indicates the channel allocation status; C3 and C4 ensure that the total computing resources allocated to all users do not exceed the maximum computing capacity of the MEC server; constraint C5 indicates that during the offloading process, each UE can only be allocated at most one channel for task transmission; constraint C6 indicates that each UE in the same base station multiplexes different channels; constraint C7 ensures that the total delay of each task does not exceed its maximum tolerable delay.

[0147] Step 4: Replace variables based on the relationship between the offloading decision variables and the channel resource allocation variables to simplify the problem, and decompose the simplified objective function into sub-problems.

[0148] The difficulty in solving the above problem lies in the existence of integer constraints and and continuous variables This is an NP-hard problem. The present invention first replaces the two decision variables by variables. Transformed into a single decision variable. The relationship between them can be obtained as follows:

[0149]

[0150] This is because if UEn s Offload the computing tasks to the MEC server, then During the task upload phase, the user must use the channel for transmission. Similarly, if UEn s If you choose to process the computing task locally, At this time, the user does not need to use channel resources, that is, therefore

[0151] Now, we can get the new objective function:

[0152]

[0153] st(C1)

[0154] (C2)

[0155] (C3)

[0156] (C4)

[0157] (C5)

[0158] (C6)

[0159] Now, let's transform Problem P into Problem P1, which involves allocating MEC computing resources and channel resources. If the user terminal chooses to reuse channel resources, the computing task is offloaded to the MEC server, which allocates the corresponding computing resources for task computation. If the terminal does not choose to reuse channel resources, the computing task is performed locally on the terminal.

[0160] Step 5: Convert subproblem 1 into a convex function regarding server computing capacity allocation, introduce the Lagrangian function and KKT condition to obtain the optimal computing resource allocation value;

[0161] Assume that the task Offload the calculation to the MEC server and give the channel allocation scheme C 0 ,at this time The set of UEs selected for offloading by each SBS is expressed as Then the original objective function is transformed into The convex function P1(1):

[0162]

[0163] stC2,C3

[0164] in Indicates that under a given channel allocation scheme C 0 Then we can get the uplink transmission rate of UEs. * ,F)About The Hessian matrix is:

[0165]

[0166] Each item of the matrix is expressed as:

[0167]

[0168] From the second-order derivative above, we can see that the diagonal elements in the Hessian matrix of formula (10) are positive (that is, each eigenvalue of the Hessian matrix is greater than 0), so the Hesse matrix is a symmetric positive definite matrix, and we can conclude that f(C 0 ,F) is a convex function. Since the constraints are linear, it can be concluded that this problem is a convex optimization problem. To this end, first construct the equation about f(C 0 ,F)’s Lagrangian function:

[0169]

[0170] Where μ is the Lagrange multiplier corresponding to the constraint C3, and λ,μ≥0. The optimal value of the improved problem can be obtained by the Karush-Kuhn-Tucker (KKT) condition. Using the KKT condition, the optimal value The conditions that μ must satisfy are:

[0171]

[0172] Using the KKT condition, the optimal solution can be obtained for:

[0173]

[0174] According to the optimal solution Substituting into problem P1(1), we can get the optimal allocation of MEC computing resources f(C 0 ,F * ):

[0175]

[0176] Step 6: Based on the optimal computing resource allocation obtained in step 5, a channel resource allocation algorithm based on differential evolution is used to solve subproblem 2 to obtain the optimal channel resource allocation.

[0177] The specific steps of the channel resource allocation algorithm based on differential evolution proposed in the present invention are as follows:

[0178] (1) Population initialization

[0179] N individuals are randomly and uniformly generated within the solution vector value range to form the initial population. The population set is represented by X i ={x i,1 ,x i,2 ,...,x i,j}, where i = 1, 2, ..., N represents the serial number of the individual in the population, j = 1, 2, ..., D (D = N s ×S×K) represents the dimension of the solution vector. Then the value of the j-th dimension of the i-th individual is as follows:

[0180] x i,j =rand(0,1)(U i,j -L i,j )+L i,j

[0181] Among them L i,j represents the lower bound of the j-th dimension of the i-th solution vector, U i,jrepresents the upper bound of the jth dimension of the ith solution vector. The solution vector in this paper is a binary discrete variable, so rand(0,1) represents a random value of "0" or "1".

[0182] (2) Variation

[0183] The differential evolution algorithm randomly selects two different individuals in the population, scales their difference, and adds other individuals to generate new mutant individuals, which can enhance the diversity of the population and reduce the risk of falling into a local optimal solution during the algorithm. i (m) is as follows:

[0184] V i (m) = F(X p (m-1)-X q (m-1))+X k (m-1)

[0185] where X p (m-1), X q (m-1), X k (m-1) is the three different individual solution vectors of the generated population in the parent generation, and F is the scaling factor. In the mutation operation, it is necessary to determine whether each individual in the population meets the boundary conditions, that is, L i,j ≤v i,j (m)≤U i,j .

[0186] (3) Cross

[0187] Crossover is to exchange the components of each individual in the parent population with the components of the corresponding mutant individual, which can enhance population diversity and avoid premature algorithm maturation. The crossover operation is as follows:

[0188]

[0189] Where randi(1,n) represents a randomly selected integer from 1 to n, ensuring that at least one gene in the crossover individual is provided by the corresponding individual in the parent generation. CR is the crossover probability, which affects the convergence speed of the algorithm and the richness of the population. A larger CR is conducive to global optimization, and a smaller CR is conducive to achieving local optimality.

[0190] (4)Select

[0191] First, the fitness function can be obtained:

[0192]

[0193] in is the penalty function, which is expressed as follows:

[0194]

[0195] in and is a penalty factor greater than zero.

[0196] After mutation and crossover operations, a set of crossover individuals are obtained. These individuals are compared with the corresponding individuals in the parent population by entering the fitness function. Following the greedy strategy, the next generation parent individuals with high fitness are retained. The greedy selection strategy is as follows:

[0197]

[0198] Among them F * (m) is the optimal MEC computing resource allocation matrix corresponding to the mth generation of individuals. The selection operation ensures that the new generation of population is better than the previous generation of population, eliminates poor individuals, retains the original excellent individuals, and guides the algorithm to approach the optimal solution.

[0199] To further illustrate the algorithm proposed in this patent, this patent has been simulated and verified.

[0200] Figure 2 The total delay variation diagram of different task numbers under different algorithms is shown in Figure 2. The joint optimization strategy of computation offloading and resource allocation proposed in this invention is compared and analyzed with the local offloading algorithm (LOC) and the random offloading algorithm (Random Offloading Completely, ROC). Figure 2 As shown in the figure, the performance of the algorithm proposed in this invention is better than that of the LOC and ROC algorithms, and the advantage of the algorithm of this invention becomes more obvious as the number of tasks increases. This is because under the condition of limited computing and communication resources, the algorithm of this invention can better allocate the optimal computing resources and channel resources to different user devices, so the total completion delay of all tasks is the lowest. Figure 2 It can be seen that when all users' computing tasks are executed locally, the size of the computing task delay is only related to the computing power of the user itself. Therefore, as the number of tasks increases, the total delay of the LOC algorithm increases linearly. When a random offloading strategy is adopted, some user devices will offload tasks to the edge server while other user devices execute computing tasks locally. Therefore, the total delay of all user devices is lower than that of the LOC algorithm and fluctuates.

[0201] Figure 3 The total delay comparison of different algorithms under different task computation loads is shown in the simulation results. Figure 3As shown in the figure, when the computation task is executed locally, the LOC algorithm shows no significant change, remaining at approximately 20 seconds. This is because the total latency of the LOC algorithm is only related to the computational resources required for the task. As the input data of the computation task increases, the total latency of this algorithm is lower than that of the LOC, ROC, and JOR algorithms. This is because the algorithm can allocate optimal channel and computational resources based on the computational complexity of the computation task, thereby minimizing the total latency for all user devices.

[0202] The method of this embodiment effectively solves the delay problem of the edge computing task offloading process in the 5G ultra-dense network, can minimize the task completion delay of the user equipment under the condition of limited resources, and improve the user's service experience.

[0203] Example 2

[0204] Provided is a joint optimization device for task offloading and resource allocation in a 5G ultra-dense network, comprising:

[0205] Network establishment module: used to establish an OFDMA-based ultra-dense network scenario consisting of a macro base station and multiple small base stations, where MEC is deployed in the macro base station to support multi-user access;

[0206] Modeling module: used to obtain basic network information, including: local terminal computing power, average task computing amount, data volume, MEC computing power, number of channels, and channel bandwidth; based on this basic network information, calculate the computing delay of the user-selected task locally and offloaded to the MEC, and calculate the system model and formula for the user's transmission rate, the transmission delay when the task is offloaded, and the total delay of the task offloaded to the MEC server for processing based on the Shannon formula;

[0207] Optimization function module: used to propose the optimization objective function for task offloading and resource allocation in 5G ultra-dense networks based on ultra-dense network scenarios, system models, and formulas;

[0208] Decomposition module: used to replace variables based on the relationship between offloading decision variables and channel resource allocation variables to simplify the optimization objective function, and then decompose the simplified optimization objective function into sub-problems to obtain computing resource allocation problems and channel resource allocation problems;

[0209] Solution output module: used to solve the computing resource allocation problem and the channel resource allocation problem respectively, obtain and output the channel resource allocation results and the total task completion delay.

[0210] Specifically, the method for the solution output module to solve the computing resource allocation problem and the channel resource allocation problem respectively includes:

[0211] The computing resource allocation problem is converted into a convex function about the allocation of server computing power. The Lagrangian function and KKT condition are introduced to obtain the optimal computing resource allocation value.

[0212] According to the optimal computing resource allocation value, the channel resource allocation algorithm based on differential evolution is used to solve the channel resource allocation problem, obtain the optimal channel resource allocation, and output the channel resource allocation result and the total task completion delay.

[0213] The device of this embodiment can be used to implement the method described in the first embodiment.

[0214] Example 3:

[0215] This embodiment provides a joint optimization device for task offloading and resource allocation in a 5G ultra-dense network, including a processor and a storage medium;

[0216] The storage medium is used to store instructions;

[0217] The processor is configured to operate according to the instructions to execute the steps of the method according to the first aspect.

[0218] The device of this embodiment can be used to implement the method described in the first embodiment.

[0219] Example 4:

[0220] This embodiment provides a computer-readable storage medium having a computer program stored thereon, wherein the program, when executed by a processor, implements the steps of the method described in the first embodiment.

[0221] Those skilled in the art will appreciate that the embodiments of the present application can be provided as methods, systems, or computer program products. Therefore, the present application can adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application can adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code.

[0222] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the steps in the flowchart and / or block diagram. Figure 1a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0223] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0224] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0225] 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 joint optimization method for task offloading and resource allocation in 5G ultra-dense networks, characterized in that: The method comprises the following steps: Establish an OFDMA-based ultra-dense network scenario consisting of a macro base station and multiple small base stations, where MEC is deployed in the macro base station to support multi-user access; Obtain basic network information, including: local terminal computing power, average task computing amount, data volume, MEC computing power, number of channels, and channel bandwidth; calculate the computing delay of the user-selected task locally and offloaded to the MEC based on the basic network information, and calculate the system model and formula for the user's transmission rate, the transmission delay when the task is offloaded, and the total delay when the task is offloaded to the MEC server for processing based on the Shannon formula; Based on the ultra-dense network scenario, system model and formula, an optimization objective function for task offloading and resource allocation in 5G ultra-dense networks is proposed; According to the relationship between the offloading decision variables and the channel resource allocation variables, the variables are replaced to simplify the optimization objective function. The simplified optimization objective function is then decomposed into subproblems to obtain the computing resource allocation problem and the channel resource allocation problem. Solve the computing resource allocation problem and channel resource allocation problem respectively, obtain and output the channel resource allocation result and the total task completion delay; Methods for solving the computing resource allocation problem and the channel resource allocation problem include: The computing resource allocation problem is converted into a convex function about the allocation of server computing power. The Lagrangian function and KKT condition are introduced to obtain the optimal computing resource allocation value. According to the optimal computing resource allocation value, a channel resource allocation algorithm based on differential evolution is used to solve the channel resource allocation problem, obtain the optimal channel resource allocation, and output the channel resource allocation result and the total task completion delay; According to the optimal computing resource allocation value, a channel resource allocation algorithm based on differential evolution is used to solve the channel resource allocation problem, including the following steps: N individuals are randomly and uniformly generated within the solution vector value range to form the initial population. The population set is represented by X i ={x i,1 ,x i,2 ,...,x i,j }, where i = 1, 2, ..., N represents the serial number of the individual in the population, j = 1, 2, ..., D (D = N s ×S×K) represents the dimension of the solution vector; then the value of the j-th dimension of the i-th individual is as follows: x i,j =rand(0,1)(U i,j -L i,j )+L i,j Among them L i,j represents the lower bound of the j-th dimension of the i-th solution vector, U i,j Indicates the upper bound of the j-th dimension of the i-th solution vector; rand(0,1) indicates that the random value is "0" or "1"; The differential evolution algorithm randomly selects two different individuals in the population, scales their difference, and adds other individuals to generate new mutant individuals; in the mth iteration, the mutant individual V i (m) is as follows: V i (m)=F(X p (m-1)-X q (m-1))+X k (m-1) where X p (m-1), X q (m-1), X k (m-1) is the three different individual solution vectors of the generated population in the parent generation, and F is the scaling factor; in the mutation operation, it is necessary to determine whether each individual in the population meets the boundary conditions, that is, L i,j ≤v i,j (m)≤U i,j ; Exchange the weight of each individual in the parent population with the weight of the corresponding mutant individual. The operation is as follows: Where randi(1,n) represents a randomly selected integer from 1 to n, ensuring that at least one gene in the crossover individual is provided by the corresponding individual in the parent generation. CR is the crossover probability, which affects the convergence speed of the algorithm and the richness of the population. A larger CR is conducive to global optimization, while a smaller CR is conducive to achieving local optimality. Get the fitness function: in is the penalty function, which is expressed as follows: in and is a penalty factor greater than zero; Indicates completion of the task The total number of CPU cycles required; Indicates a task The amount of data; Represents a given channel allocation matrix C 0 The task after Uplink transmission delay, Represents the set of SBS components, Represents a collection of uninstalled users. represents a set of subchannels; represents the total completion delay of the task, Indicates the UE's offloading decision, if Indicates a task Offload computing to edge servers, Indicates a task Process locally; Indicates the channel allocation value, Represents UEn s Use subchannel k for offloading, Represents UEn s Subchannel k is not used for offloading; Indicates that MEC is UEn s Allocated computing resources; f m Indicates the maximum computing capacity of MEC; Indicates a task The maximum tolerable delay; After mutation and crossover operations, a group of crossover individuals are obtained. These individuals are compared with the corresponding individuals in the parent population by entering the fitness function. Following the greedy strategy, the next generation parent individuals with high fitness are retained. The greedy selection strategy is as follows: Among them F * (m) is the optimal MEC computing resource allocation matrix corresponding to the mth generation individual.

2. The joint optimization method according to claim 1, characterized in that: In a multi-cell, multi-user UDN scenario, based on the different resources required by different tasks, the optimization objective function for offloading decision X, channel allocation C, and MEC computing resource allocation F under resource constraints can be expressed as: Where X is the offloading decision matrix, C is the channel allocation matrix, and F is the MEC computing resource allocation matrix. Represents the set of SBS components, Represents the set of UEs served by each SBS, Represents a collection of uninstalled users. represents a set of subchannels; represents the total completion delay of the task, Indicates the UE's offloading decision, if Indicates a task Offload computing to edge servers, Indicates a task Process locally; Indicates the channel allocation value, Represents UEn s Use subchannel k for offloading, Represents UEn s Subchannel k is not used for offloading; Indicates that MEC is UEn s Allocated computing resources; f m Indicates the maximum computing capacity of MEC; Indicates a task The maximum tolerable delay.

3. The joint optimization method according to claim 1, characterized in that: The present invention uses variable replacement based on the relationship between the offloading decision variable and the channel resource allocation variable to simplify the optimization objective function, and decomposes the simplified optimization objective function into subproblems. The method includes the following steps: according to The relationship between , we get the following equation: Where, represents the set of subchannels, K represents the number of subchannels; Indicates the UE's offloading decision, if Indicates a task Offload computing to edge servers, Indicates a task Process locally; Indicates the channel allocation value, Represents UEn s Use subchannel k for offloading, Represents UEn s Subchannel k is not used for offloading; Get the new objective function P1: Where C is the channel allocation matrix, F is the MEC computing resource allocation matrix, Represents the set of SBS components, Represents the set of UEs served by each SBS, Represents a collection of uninstalled users. represents a set of subchannels; Indicates the total completion delay of the task; Indicates the channel allocation value, Represents UEn s Use subchannel k for offloading, Represents UEn s Subchannel k is not used for offloading; Indicates that MEC is UEn s Allocated computing resources; f m Indicates the maximum computing capacity of MEC; Indicates a task The maximum tolerable delay; Problem P is transformed into problem P1 of allocating MEC computing resources and channel resources. If the user terminal chooses to reuse channel resources, the computing task is offloaded to the MEC server, and the MEC server allocates the corresponding computing resources for task computing. If the terminal does not choose to reuse channel resources, the computing task is performed locally on the terminal. Using the indirect optimization method, problem P1 is divided into computing resource allocation problem and channel resource allocation problem for solution.

4. The joint optimization method according to claim 1, characterized in that: The computing resource allocation problem is converted into a convex function about the allocation of server computing power. The Lagrangian function and KKT condition are introduced to obtain the optimal computing resource allocation value, including the following steps: The task Offload the calculation to the MEC server and give the channel allocation scheme C 0 ,at this time The set of UEs selected for offloading by each SBS is expressed as Then the original objective function is transformed into The convex function P1(1): stC2,C3 in Indicates that under a given channel allocation scheme C 0 Then we can get the uplink transmission rate of UEs; find f(C * ,F)About The Hessian matrix is: Each item of the matrix is expressed as: Construct about f(C 0 ,F)’s Lagrangian function: Where μ is the Lagrange multiplier corresponding to the constraint condition C3, and λ,μ≥0; the optimal value of the improved problem can be obtained by the Carlo-Kuhn-Tucker condition; using the Carlo-Kuhn-Tucker condition, the optimal value The conditions that μ must satisfy are: Using the KKT condition, the optimal solution can be obtained for: According to the optimal solution Substituting into the convex function P1(1), we can get the optimal allocation of MEC computing resources f(C 0 ,F * ): Where C 0 Represents the given channel allocation matrix, F * represents the calculated optimal MEC computing resource allocation matrix, Represents the set of SBS components, Indicates the collection of uninstalled users; f m Indicates the maximum computing capacity of MEC; Indicates completion of the task The total number of CPU cycles required; Indicates a task The amount of data; Represents a given channel allocation matrix C 0 The task after Uplink transmission delay.

5. A joint optimization device for task offloading and resource allocation in a 5G ultra-dense network for executing the method according to claim 1, characterized in that: include: Network establishment module: used to establish an OFDMA-based ultra-dense network scenario consisting of a macro base station and multiple small base stations, where MEC is deployed in the macro base station to support multi-user access; Modeling module: used to obtain basic network information, including: local terminal computing power, average task computing amount, data volume, MEC computing power, number of channels, and channel bandwidth; based on this basic network information, calculate the computing delay of the user-selected task locally and offloaded to the MEC, and calculate the system model and formula for the user's transmission rate, the transmission delay when the task is offloaded, and the total delay of the task offloaded to the MEC server for processing based on the Shannon formula; Optimization function module: used to propose the optimization objective function for task offloading and resource allocation in 5G ultra-dense networks based on ultra-dense network scenarios, system models, and formulas; Decomposition module: used to replace variables based on the relationship between offloading decision variables and channel resource allocation variables to simplify the optimization objective function, and then decompose the simplified optimization objective function into sub-problems to obtain computing resource allocation problems and channel resource allocation problems; Solution output module: used to solve the computing resource allocation problem and the channel resource allocation problem respectively, obtain and output the channel resource allocation results and the total task completion delay.

6. The joint optimization device according to claim 5, wherein the method in which the solution output module solves the computing resource allocation problem and the channel resource allocation problem respectively comprises: The computing resource allocation problem is converted into a convex function about the allocation of server computing power. The Lagrangian function and KKT condition are introduced to obtain the optimal computing resource allocation value. According to the optimal computing resource allocation value, the channel resource allocation algorithm based on differential evolution is used to solve the channel resource allocation problem, obtain the optimal channel resource allocation, and output the channel resource allocation result and the total task completion delay.

7. A joint optimization device for task offloading and resource allocation in 5G ultra-dense networks, characterized in that: including processors and storage media; The storage medium is used to store instructions; The processor is configured to operate according to the instructions to execute the steps of the method according to any one of claims 1 to 4.

8. 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