Task unloading scheduling method and device for multi-user multi-access point MEC

Abstract

The invention relates to a task unloading scheduling method and device for a multi-user multi-access point MEC. The task unloading scheduling method comprises the following steps: initializing the individual number of a particle swarm, the speed vector of particles and a task unloading scheme; according to the cascade channel information and the noise power, fitness calculation is carried out in combination with the position vector of the particle swarm population, the current particle swarm population and the corresponding fitness are calculated, particle swarm updating is carried out, and the updated particle swarm population is obtained; thirdly, performing constraint term check on the updated position vector of the particle to obtain a particle swarm population conforming to constraints, and judging whether the position vector of the particle swarm population meets an iteration threshold or not: if yes, performing unloading allocation according to the optimal position vector to complete scheduling; and if not, carrying out fitness calculation again. According to the task unloading scheduling method, the boundary crossing problem in the particle swarm optimization process is remarkably inhibited, and the scheduling robustness and the service quality of multi-user multi-access-point task unloading are improved.

Description

Technical Field

[0001] The present invention relates to the field of resource scheduling for mobile edge computing, and in particular to a task offloading scheduling method, a task offloading scheduling device, and a mobile edge computing system for multi-user multi-access point MEC. Background Art

[0002] Under the impact of the digital wave, applications such as high-definition video, semantic recognition, virtual reality, and smart home are booming, and the demand for computing power on terminal devices is growing exponentially. However, terminal devices, especially mobile devices, are limited by battery capacity, computing resources, and network transmission speeds, making it difficult to carry out complex computing tasks. Therefore, Mobile Edge Computing (MEC) technology came into being. MEC deploys computing resources (such as edge servers) at the edge of the network, allowing users' mobile devices to offload all or part of the computing tasks to nearby edge servers, which complete the computing or processing tasks and return the results to the user, thereby effectively reducing the energy consumption and response latency of the user device. However, when multiple users are concurrent and the edge server has multiple access points, it is difficult to determine the optimal low-energy task allocation solution through simple preset rules.

[0003] To this end, traditional technologies typically model the multi-user, multi-access point energy consumption optimization problem in MEC using mathematical programming methods, including integer programming and convex optimization, to minimize system energy consumption while satisfying resource constraints and latency requirements. However, because this type of problem is NP-hard, traditional methods often struggle to obtain feasible solutions within a limited time when faced with a large number of users and access points. They also suffer from slow convergence and are prone to falling into local optimality, making them difficult to meet the real-time and computational efficiency requirements of practical systems.

[0004] Based on this, existing technologies generally use heuristic algorithms, such as genetic algorithms, ant colony algorithms, and particle swarm optimization algorithms (PSO), to deal with energy consumption optimization problems in MEC systems. Among them, PSO simulates particles in the search space and continuously adjusts the flight speed and position based on individual experience and group experience, thereby gradually approaching the optimal solution. In the MEC scenario, each particle represents a task offloading and access point allocation strategy. The algorithm evaluates the total energy consumption of the system through the fitness function to judge the quality of the particles, and continuously optimizes the task allocation plan through operations such as initializing the population, recording individual and global optimal values, and iteratively updating speed and position, and finally outputs the task offloading and access strategy with the lowest energy consumption.

[0005] However, when applying the particle swarm optimization (PSO) algorithm to MEC systems, existing technologies face difficulties in applying complex optimization scenarios such as multi-user and multi-access point optimization. Because task completion delay is typically a composite function composed of multiple parameters such as computing resource allocation and transmission rate, it is difficult to directly and explicitly model it as a feasible boundary constraint, and it is also difficult to directly perform boundary corrections. Furthermore, existing technologies typically only optimize the offloading relationship between users and access points. To simplify the design of optimization variables and constraints, task allocation behavior is generally modeled only in the form of discrete variables, without considering continuous variable factors such as the dynamic allocation of computing resources. This makes it difficult to accurately reflect the coupling relationship between task completion delay and computing resource allocation. Consequently, when dealing with application scenarios with high service quality requirements or strong task heterogeneity, the generated offloading strategies have hidden dangers such as excessive execution delay and unstable task response.

[0006] Furthermore, when applying the existing particle swarm algorithm to the joint optimization problem of multi-user, multi-access point task offloading, the discrete variables in the two-dimensional scheduling matrix are subject to dual row and column constraints, which can easily lead to over-allocation and over-allocation of continuous variables. This can cause the particle swarm's corresponding position to exceed the search space boundary after updating, resulting in fitness failure and infeasible solutions, seriously affecting the convergence and stability of the optimization process. Consequently, when applying the particle swarm algorithm to the multi-user, multi-access point MEC offloading optimization problem, the existing technology faces the problem of difficulty in converting task completion delays into explicit optimization constraints and the failure to suppress particle over-allocation. Summary of the Invention

[0007] Based on this, the object of the present invention is to provide a task offloading scheduling method for multi-user multi-access point MEC.

[0008] A multi-user multi-access point MEC task offloading scheduling method includes the following steps:

[0009] S1. Initialize the number of particle swarm individuals, the particle velocity vector, and the task offloading scheme to obtain an initial particle swarm population; wherein the task offloading scheme includes a user selection matrix and a computing resource allocation matrix;

[0010] S2. Calculate the objective function value consisting of energy consumption and delay penalty based on the cascade channel information and noise power between the user and the access point, and in combination with the current task offloading scheme of the particle swarm population, to obtain the current particle swarm population and the corresponding fitness; wherein the objective function consisting of energy consumption and delay penalty is subject to several constraints, including computing resource limitation constraint, task unique access constraint, access point capacity constraint, and task offloading variable binary constraint;

[0011] S3, update the particle swarm according to the current particle swarm population and the corresponding fitness to obtain the updated particle swarm population;

[0012] S4. Performing a constraint check on the task offloading scheme in the updated particle swarm population to obtain a particle swarm population that meets the constraints; wherein the constraint check is used to sequentially perform out-of-bounds checks and corrections on the computing resource allocation constraint, the task unique access constraint, and the access point capacity constraint;

[0013] S5. Determine whether the global optimal task offloading scheme of the particle swarm population that meets the constraints meets the iteration threshold: if so, complete the iteration, offload the user's computing tasks according to the global optimal task offloading scheme, and complete the offloading scheduling of the user's tasks; if not, re-execute step S2.

[0014] Compared with the prior art, the task offloading scheduling method for multi-user, multi-access point MEC described in the present invention constructs an objective function consisting of the total system energy consumption and the delay penalty term, introduces the user selection matrix (discrete variable) and the computing resource allocation matrix (continuous variable) as collaborative optimization variables, and replaces the maximum task delay constraint with a model by introducing a penalty term. This avoids the problem of delay constraints being difficult to directly correct their boundaries during the optimization process, thereby enabling the optimization process to simultaneously focus on task allocation and resource scheduling, significantly enhancing the offloading strategy's adaptability to delay-sensitive tasks, and improving scheduling robustness and service quality in complex network environments with multiple users and multiple access points.

[0015] At the same time, the present invention addresses the problems of discrete variable logic overlap and conflict caused by the simultaneous restriction of rows and columns of the user selection matrix in the task offloading scheme, as well as the problem of boundary crossing of the computing resource allocation matrix during the particle update process. By sequentially checking and correcting the computing resource limitation constraints, task unique access constraints, and access point capacity constraints, the conflict relationship between multiple constraints is effectively coordinated, the feasibility and stability of the task offloading solution are guaranteed, and the problems of non-convergence and solution failure caused by crossing the boundary in the particle swarm optimization process are significantly suppressed, thereby improving the convergence speed and stability of the particle swarm in MEC task offloading allocation with multiple users and multiple access points.

[0016] Furthermore, for the task offloading scheme corresponding to the kth particle in the particle swarm population of the tth iteration, The specific calculation of the objective function composed of the corresponding energy consumption and delay penalty is expressed as follows:

[0017]

[0018] Where, represents the fitness of the k-th particle corresponding to the task offloading scheme at the t-th iteration, It represents the task offloading scheme of the k-th particle at the t-th iteration, that is, the position vector of the particle, including the user selection matrix and the computing resource allocation matrix, which is specifically expressed as follows:

[0019]

[0020] Where, represents the user selection matrix of the kth particle at the tth iteration, whose element u ij Select elements for offloading; M, N represent the total number of users and access points, respectively;

[0021] is the computational resource allocation matrix for the kth particle at the tth iteration, whose elements assigning elements to resources;

[0022] represents the total energy consumption of the system corresponding to the kth particle, and its specific calculation is expressed as:

[0023]

[0024] Where, E ij represents the energy consumption when the i-th user selects the j-th access point as the task offloading point, α i represents the computational complexity coefficient of the task of the i-th user; k0 represents the local computational energy consumption coefficient; l i represents the task length of the i-th user; represents the computation frequency of the local device of the i-th user;

[0025] β i represents the uplink overhead coefficient of the i-th user; P T,i is the transmission power of the i-th user's device; R ij represents the transmission rate between the i-th user and the j-th access point, and its specific calculation is as follows:

[0026]

[0027] Where B W,j represents the channel bandwidth allocated to the jth access point; represents the noise power; h ij is the cascade channel information between the i-th user and the j-th access point, that is, the cascade channel gain, which is specifically calculated as follows:

[0028]

[0029] Where β0 represents the path loss constant; d ij represents the distance between the i-th user and the j-th access point; h w,ijis the small-scale fast fading channel gain;

[0030] represents the delay penalty term, λ is the penalty factor; represents the maximum tolerable delay of the task of the i-th user; T i It represents the total execution delay of the i-th user task, and its specific calculation is as follows:

[0031]

[0032] Where, T ij represents the connection delay between the i-th user and the j-th access point.

[0033] Based on this, the present invention collaboratively constructs an optimization objective function by combining the total system energy consumption with the delay penalty term, and uses the user selection matrix and the computing resource allocation matrix in the particle position vector as joint optimization variables, so that particles can simultaneously optimize task scheduling and resource allocation strategies in the search space, thereby adaptively adjusting the computing frequency allocation of the edge server to different users, enabling offloaded tasks to obtain dynamic processing capabilities, thereby reducing task completion delays, and improving the system's task processing robustness and delay guarantee capabilities in complex scenarios such as multi-task heterogeneity and dynamic changes in access point loads.

[0034] Furthermore, for the task offloading scheme corresponding to the kth particle in the particle swarm population of the tth iteration, Its update rule at the t+1th iteration is expressed as follows:

[0035]

[0036] Where, Indicates whether the i-th user offloads the task to the j-th access point in the t+1-th iteration, that is, the user selection matrix of the k-th particle The corresponding element u ij ;

[0037] ε is a random variable that obeys uniform distribution in the interval [0,1]; represents the velocity vector corresponding to the kth particle in the particle swarm population at the t+1th iteration, and The element u of the matrix selected by the user ij The corresponding velocity components; is the Sigmoid function;

[0038] represents the computational resource allocation matrix of the kth particle at the t+1th iteration; The computing resource allocation matrix The corresponding velocity component.

[0039] Based on this, the present invention introduces the Sigmoid function to map the velocity component during the update process of the user selection matrix, realizes the probabilistic modeling of the user selection matrix update behavior, and combines it with the random perturbation judgment mechanism to improve the strategy diversity of the particle swarm during the search process, avoids the search instability problem caused by jitter or premature convergence, and ultimately enhances the global exploration ability and convergence stability of the particle swarm in the complex search space of multiple users and multiple access points.

[0040] Furthermore, for the velocity vector corresponding to the kth particle in the particle swarm population of the tth iteration, Its update rule at the t+1th iteration is expressed as follows:

[0041]

[0042] Where c1 represents the learning factor of individual extreme value; r1 is the random factor of individual extreme value, which obeys the uniform distribution of [0,1]. Optimal task offloading solution for individuals;

[0043] c2 represents the learning factor of the group extreme value; r2 represents the random factor of the group extreme value, which obeys the uniform distribution of [0,1]; is the global optimal task offloading solution for the tth iteration;

[0044] is the adaptive inertia weight of the current k-th particle at the t-th iteration, and its specific calculation is as follows:

[0045]

[0046] Where, ω min and ω max is the minimum and maximum value of the inertia weight; represents the maximum fitness of all particles in the tth iteration; It represents the average fitness of all particles in the tth iteration.

[0047] Based on this, the present invention introduces a task offloading scheme corresponding to individual extreme values and group extreme values as a guide in the velocity vector update process, and combines it with adaptive inertia weights to effectively balance the global exploration and local convergence capabilities in the particle search process, thereby improving the convergence speed and global search performance of the optimization process, and enhancing the stability and search efficiency of the algorithm in multi-constrained complex solution spaces.

[0048] Furthermore, the constraint item check includes the following sub-steps:

[0049] S401. Check whether the computing frequency allocated by each access point to all users complies with the computing resource allocation constraint. If so, proceed to the next sub-step S402. If not, perform normalization correction processing and proceed to the next sub-step S402. The specific correction is as follows:

[0050]

[0051] Where, represents the maximum available computing frequency resource of the jth access point;

[0052] S402: Check whether the user selection matrix in the task offloading scheme of all particles meets the task unique access constraint. If so, proceed to the next sub-step S403. If not, further determine the selection of any user i, i.e., the i-th row, in the user selection matrix of the corresponding particle k:

[0053] If user i does not select any access point in the user selection matrix of particle k, the velocity vector corresponding to user i is mapped by the Sigmoid function to obtain the connection probability vector, and the user selection matrix corresponding to the maximum probability is converted to The elements of are set to 1 and the rest are set to 0; for any particle k in the t+1th iteration, the user selection matrix Elements The corrections are as follows:

[0054]

[0055] Where, Indicates the maximum value index to obtain the item; Used to represent the velocity row component corresponding to the i-th row vector of the user selection matrix;

[0056] If a user i selects multiple access points in the particle's user selection matrix, the velocity vector corresponding to user i is mapped using the sigmoid function, and the mapping result is compared with the user selection matrix. Perform element-wise multiplication and convert the user-selected matrix corresponding to the maximum value in the product result into The elements of are set to 1, and the rest are set to 0; the element correction of the user-selected matrix for any particle k is expressed as follows:

[0057]

[0058] Then, after the correction is completed, the next sub-step S403 is executed;

[0059] S403: Check whether the user selection matrix in the task offloading scheme of all particles meets the access point capacity constraint. If so, complete the constraint check and set the current particle swarm population to the particle swarm population that meets the constraint. If not, further determine the allocation of any access point j, i.e., the jth column, in the user selection matrix of the corresponding particle k:

[0060] If the number of users assigned to access point j in the user selection matrix of particle k is greater than the maximum number of users allowed by the access point, the velocity vector corresponding to access point j is mapped using the sigmoid function, and the mapping result is element-wise multiplied with the user selection matrix to obtain the product result;

[0061] Next, among the users who select access point j, the indexes corresponding to the minimum values of the number of users exceeding the number of allowed users of the access point in the multiplication result are selected as the index set. The specific calculation expression is as follows:

[0062]

[0063] Where, (j) is the index set corresponding to access point j, the total number of which is Ex, and Ex is the number of users allowed to exceed the access point, which is specifically expressed as

[0064] Next, the elements corresponding to the index set in the user selection matrix are transferred to the local execution task to complete the access point capacity constraint correction, which is specifically expressed as: and

[0065] Based on this, the present invention sequentially checks and corrects the computing resource allocation constraints, task unique access constraints, and access point capacity constraints, and designs a normalized redistribution mechanism and a task repair strategy based on probability mapping for continuous variables and discrete variables respectively, thereby effectively solving the problems of infeasible solutions, constraint conflicts, and fitness failure caused by variable out-of-bounds during the particle update process, and improving the feasibility of the task offloading solution, constraint compatibility, and the overall convergence stability of the algorithm.

[0066] A task offloading scheduling device for multi-user multi-access point MEC, comprising a particle swarm population initialization unit, a fitness calculation unit, a particle swarm update unit, a constraint item checking unit, and an iteration condition judgment unit;

[0067] The particle swarm population initialization unit is used to initialize the number of particle swarm individuals, the velocity vector of the particles, and the task offloading scheme to obtain an initial particle swarm population; wherein the task offloading scheme includes a user selection matrix and a computing resource allocation matrix;

[0068] The fitness calculation unit is configured to calculate an objective function value consisting of corresponding energy consumption and delay penalty based on the cascade channel information and noise power between the user and the access point, in combination with the current task offloading scheme of the particle swarm population, to obtain the current particle swarm population and the corresponding fitness; wherein the objective function consisting of energy consumption and delay penalty is subject to several constraints, including computing resource limitation constraints, task unique access constraints, access point capacity constraints, and task offloading variable binary constraints;

[0069] The particle swarm updating unit is used to update the particle swarm according to the current particle swarm population and the corresponding fitness to obtain an updated particle swarm population;

[0070] The constraint item checking unit is used to perform constraint item checking on the task offloading scheme in the updated particle swarm population to obtain a particle swarm population that meets the constraints; wherein the constraint item checking is used to sequentially perform out-of-bounds checking and correction on the computing resource allocation constraint, the task unique access constraint, and the access point capacity constraint;

[0071] The iteration condition judgment unit is used to judge whether the global optimal task offloading scheme of the particle swarm population that meets the constraints meets the iteration threshold: if so, the iteration is completed, the user's computing tasks are offloaded and allocated according to the global optimal task offloading scheme, and the offloading scheduling of the user tasks is completed; if not, the fitness calculation unit is re-called.

[0072] A multi-user and multi-access point mobile edge computing system includes a base station and a task offloading scheduling device in communication with the base station;

[0073] The base station is provided with a plurality of access points in communication with the edge computing server, which are used to receive offloading requests corresponding to computing tasks from user equipment, and calculate the cascade channel information and noise power between each current user and each access point according to the computing task and its corresponding computing demand, and send the information to the task offloading scheduling device;

[0074] The task offloading scheduling device is the multi-user multi-access point MEC task offloading scheduling device described above.

[0075] For better understanding and implementation, the present invention is described in detail below with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0076] Figure 1 A simplified schematic diagram of signal transmission for a mobile edge computing system;

[0077] Figure 2 This is a simplified structural diagram of the task offloading scheduling device for multi-user multi-access point MEC according to the present invention;

[0078] Figure 3 This is a simplified flowchart of the multi-user multi-access point MEC task offloading scheduling method according to the present invention;

[0079] Figure 4 Schematic diagram of comparison curves of the present invention compared with existing strategies such as full local computing, full remote computing, and random selection. DETAILED DESCRIPTION

[0080] In order to solve the problems in the prior art of using particle swarm optimization algorithm to offload tasks from MEC, such as the difficulty in converting delay into explicit optimization constraints and the unchecked particle out-of-bounds behavior, the present invention generates an initial particle swarm population by initializing the number of particle swarm individuals, the speed connection of particles and the task offloading scheme; according to the cascade channel information and noise power between the user and the access point, and in combination with the current task offloading scheme of the particle swarm population, calculates the objective function value consisting of the corresponding energy consumption and delay penalty as the fitness of the particle, and obtains the current particle swarm population and the corresponding fitness; then, according to the current particle swarm population and the corresponding fitness The inertia weight is calculated based on the corresponding fitness, and the velocity vector and task offloading scheme of the particle are updated according to the inertia weight to obtain the updated particle swarm population; the task offloading scheme in the updated particle swarm population is checked for constraints. If the constraints are not met, the positions of the relevant particles are corrected; if the constraints are met, whether the current particle swarm population meets the iteration termination conditions is further judged: if so, the current task offloading scheme is output as the optimal solution to guide the access point allocation and computing resource allocation of the user's computing tasks; if not, the algorithm returns and continues to execute the fitness calculation and subsequent optimization steps until convergence is completed.

[0081] Based on this, the present invention jointly constructs the objective function by combining the total system energy consumption and the task execution delay penalty term, indirectly constructs the delay constraint through the penalty term, and introduces the user selection matrix (discrete variables) and the computing resource allocation matrix (continuous variables) for collaborative optimization. This not only improves the adaptability of the offloading strategy to delay-sensitive tasks, but also further enhances the scheduling robustness of the system in complex network environments.

[0082] In addition, the present invention effectively suppresses the out-of-bounds behavior of particles that may occur during the iteration process by introducing a multi-dimensional constraint item checking and correction mechanism, and significantly improves the convergence stability and solution feasibility of the particle swarm algorithm in the MEC scenario.

[0083] Based on the above design, the present invention proposes a task offloading scheduling method for multi-user multi-access point MEC, and based on this method, proposes a task offloading scheduling device for multi-user multi-access point MEC.

[0084] See also Figure 1 , Figure 1 A simplified schematic diagram of signal transmission in a mobile edge computing system.

[0085] A multi-user multi-access point mobile edge computing system includes a user equipment 100, a base station 101, and a multi-user multi-access point MEC task offloading scheduling device 102 according to the present invention, which is communicatively connected to the base station.

[0086] The user equipment 100 is used to generate computing tasks, such as data processing, video streaming, IoT sensor data, etc., and send the computing tasks and corresponding computing requirements to the base station 101.

[0087] The base station 101 is provided with several access points 1011 (AccessPoint, AP) that are communicatively connected to the edge computing server, and is used to calculate the cascade channel information and noise power between each current user and each access point according to the computing task and the corresponding computing requirements, and send it to the multi-user multi-access point MEC task offloading scheduling device 102 of the present invention.

[0088] Please also see Figure 2 and Figure 3 , Figure 2 This is a simplified structural diagram of the task offloading scheduling device for multi-user multi-access point MEC according to the present invention. Figure 3 This is a simplified flowchart of the multi-user multi-access point MEC task offloading scheduling method according to the present invention.

[0089] The multi-user multi-access point MEC task offloading scheduling device 102 includes a particle swarm population initialization unit 1, a fitness calculation unit 2, a particle swarm update unit 3, a constraint item checking unit 4 and an iteration condition judgment unit 5.

[0090] The particle swarm population initialization unit 1 is used to execute step S1: initializing the number of particle swarm individuals, the velocity vector of the particles and the task offloading scheme to obtain an initial particle swarm population.

[0091] Specifically, the particle swarm population includes a task offloading scheme and a velocity vector of the particles, which is specifically expressed as follows:

[0092]

[0093] Where N PSO Indicates the number of particle swarm individuals, which is used to determine the coverage of the search space. Its default value is 50 and can be adjusted according to actual needs. The present invention does not specifically limit it here;

[0094] It represents the task offloading scheme of the kth particle at the 0th iteration, that is, the position vector of the particle, including the user selection matrix and the computing resource allocation matrix, which is specifically expressed as follows:

[0095]

[0096] Where, It represents the user selection matrix of the kth particle at the 0th iteration, which is a set of binary variables, and its element u ij is the offloading selection element, which is used to indicate whether the i-th user chooses to offload the task to the j-th access point; M and N represent the total number of users and access points respectively;

[0097] The computational resource allocation matrix for the kth particle at iteration 0, whose elements is a resource allocation element, which is used to represent the task calculation frequency allocated by the edge server of the j-th access point to the i-th user, that is, the allocated computing resources.

[0098] Among them, the user selection matrix and computing resource allocation matrix The initial value of can be generated by a random variable that obeys a certain distribution, such as uniform distribution, normal distribution, etc., or can be set in combination with historical statistics or system prior information. In order to ensure that the initialization process has sufficient diversity and adaptability, thereby improving the coverage effect of the search space, different initialization methods can be used, and the present invention does not make specific limitations on this.

[0099] represents the velocity vector of the kth particle at iteration 0, and is used to control the movement trend of the particle in the multidimensional search space. Its dimension is consistent with the task offloading scheme. For discrete variables, that is, the user selection matrix, the components of the velocity vector can be expressed as the probability of state switching. For continuous variables, that is, the computing resource allocation matrix, the components of the velocity vector can be expressed as the change amplitude of the particle's current position.

[0100] Among them, the initial value of the velocity vector can be generated by a random variable that obeys a specific distribution (such as uniform distribution, normal distribution, etc.), or can be assigned based on heuristic information. The specific selection can be made according to the actual deployment scenario. The present invention does not make specific restrictions on the initialization method of the velocity.

[0101] The fitness calculation unit 2 is used to execute step S2: according to the cascade channel information and noise power between the user and the access point, and in combination with the task offloading scheme of the current particle swarm population, calculate the objective function value consisting of the corresponding energy consumption and delay penalty, and obtain the current particle swarm population and the corresponding fitness.

[0102] Specifically, for the task offloading scheme of the kth particle at the tth iteration The specific calculation of the objective function composed of the corresponding energy consumption and delay penalty is expressed as follows:

[0103]

[0104] Where, represents the fitness of the k-th particle corresponding to the task offloading scheme at the t-th iteration; represents the total energy consumption of the system corresponding to the kth particle, and its specific calculation is expressed as:

[0105]

[0106] Where, E ij represents the energy consumption when the i-th user selects the j-th access point as the task offloading point, α i represents the computational complexity coefficient of the task of the i-th user, which can be predefined by the task type or obtained through static analysis of the task; k0 represents the local computational energy consumption coefficient, and its default value is l i The length of the task of the i-th user is directly recorded when the task is generated or calculated from the input data size, and the unit is the total bit amount; Indicates the computing frequency of the local device of the i-th user, which can be obtained based on hardware performance parameters or read through the device information interface;

[0107] β i It represents the uplink overhead coefficient of the i-th user, which is used to compensate for the overhead of non-ideal factors such as control signaling and retransmission in the actual link. It is usually set to β i ≥1; P T,i is the transmission power of the i-th user's device; R ij represents the transmission rate between the i-th user and the j-th access point, and its specific calculation is as follows:

[0108]

[0109] Where B W,j represents the channel bandwidth allocated to the jth access point; represents the noise power; h ij is the cascade channel information between the i-th user and the j-th access point, that is, the cascade channel gain, which is specifically calculated as follows:

[0110]

[0111] Where β0 represents the path loss constant, which defaults to 0.1; d ijh represents the distance between the i-th user and the j-th access point, which is calculated based on the geometric position relationship between the user equipment and the access point, or can be inferred from the signal strength. The present invention does not specifically limit the calculation method of the distance; w,ij is the small-scale fast fading channel gain, which obeys a complex Gaussian distribution with mean 0 and variance 1, that is, It is used to simulate the random amplitude fluctuations caused by the superposition of signals after they pass through different propagation paths in a multipath environment.

[0112] In the case of edge offloading, since the access point server is usually connected to a fixed power source, the computing resources allocated to it are Not included in system energy consumption, that is, not included in E ij Calculation.

[0113] represents the delay penalty term, λ is the penalty factor, the default value is λ>0, and the default unit is joule / second. It is used to quantify the delay cost caused by exceeding the maximum tolerable delay of the task into an equivalent term of energy consumption; represents the maximum tolerable delay of the task of the i-th user, Adaptive values can be set based on specific task types or according to the application's Quality of Service (QoS) requirements;

[0114] T i It represents the total execution delay of the i-th user task, and its specific calculation is as follows:

[0115]

[0116] Where, T ij represents the connection delay between the i-th user and the j-th access point.

[0117] The objective function composed of energy consumption and delay penalty is subject to several constraints to ensure the feasibility of the task offloading scheme and the reasonable scheduling of system resources. The constraints include computing resource limitation constraint, task unique access constraint, access point capacity constraint and task offloading variable binary constraint.

[0118] The computing resource constraint is used to ensure that the total computing resources allocated to all users by access point j does not exceed its maximum computing capacity to prevent resource overload. Its expression is as follows:

[0119]

[0120] Where, It represents the maximum available computing frequency resource of the jth access point, which can be preset by system hardware parameters or set by the edge server operator; Indicates that the node numbered 0 is excluded from the access point number set, that is, the computing resource consumption of the local device is not considered, and only the actually deployed edge servers are constrained;

[0121] The task unique access constraint is used to ensure that each user's task must select one and only one computing node (local or access point) to perform task offloading, avoiding abnormal situations of multiple connections or unassigned tasks. Its expression is as follows:

[0122]

[0123] The access point capacity constraint is used to limit the maximum number of users that can be accessed simultaneously at each access point at the same time, reflecting the connection capacity and wireless resource scheduling capability of the edge device. Its expression is as follows:

[0124]

[0125] Where μ j Indicates the maximum number of users allowed to access the jth access point, which can be preset by system hardware parameters or set by the edge server operator;

[0126] The binary constraint of the task offloading variable is used to ensure that the task offloading selection variable conforms to Boolean logic, and its expression is as follows:

[0127]

[0128] The particle swarm updating unit 3 is used to execute step S3: updating the particle swarm according to the current particle swarm population and the corresponding fitness to obtain an updated particle swarm population.

[0129] Specifically, for the current task offloading scheme of the kth particle in the tth iteration Its update rule at the t+1th iteration is expressed as follows:

[0130]

[0131] Where, Indicates whether the i-th user offloads the task to the j-th access point in the t+1-th iteration, that is, the user selection matrix of the k-th particle The corresponding element u ij ; The element u of the matrix selected by the user ij The corresponding velocity component is used to represent the element u ij The switching probability; ε is a random variable that obeys the uniform distribution in the interval [0,1]; is a Sigmoid function used to map speed values to switching probabilities;

[0132] represents the computational resource allocation matrix of the kth particle at the t+1th iteration; The computing resource allocation matrix The corresponding velocity components, each of which is used to control the computing resource allocation matrix Update direction and magnitude.

[0133] Among them, the velocity vector of the kth particle at the tth iteration is Its update rule at the t+1th iteration is expressed as follows:

[0134]

[0135] Where, is the individual cognitive term, which is used to represent the thinking of the particle itself; c1 is the learning factor of the individual extreme value, which is used to control the step size of the particle's approach to its own historical optimal position vector. Its initial value defaults to 1.5; r1 is the random factor of the individual extreme value, which is used to introduce search disturbance to prevent falling into the local optimum. It obeys the uniform distribution of [0,1]. It is the individual optimal task offloading scheme, which is used to represent the task offloading scheme when the kth particle in the particle swarm of the current iteration obtains the minimum fitness in the historical iteration times. It will be updated once after each fitness calculation;

[0136] is the group social term, which is used to represent the information sharing and mutual cooperation between particles; c2 is the learning factor of the group extreme value, which is used to control the step size of the particle approaching the optimal position vector among all particles in history; r2 is the random factor of the group extreme value, which obeys the uniform distribution of [0,1]; is the global optimal task offloading scheme for the tth iteration, which is used to represent the task offloading scheme with the minimum fitness among all particles in the particle swarm of the current iteration. It will be updated once after each fitness calculation;

[0137] is the inertia term, is the adaptive inertia weight of the current k-th particle at the t-th iteration, which is used to control the search breadth of the particle in the search space. The value of the inertia weight is adaptively adjusted by the fitness function value of the current particle. Its specific calculation is expressed as follows:

[0138]

[0139] Where, ω min and ω max The minimum and maximum values of the inertia weight are used to control the range of the solution space search. The default values are 0.4 and 0.8; Indicates the maximum value of the fitness of all particles in the tth iteration, which is used to indicate the fitness of the worst particle; It represents the average fitness of all particles in the tth iteration.

[0140] Accordingly, when the particle fitness is greater, the adaptive inertia weight will expand the search range, thereby encouraging the particle to jump to the optimal solution; when the particle fitness is smaller, the adaptive inertia weight will narrow the search range, thereby guiding the particle to local convergence.

[0141] The constraint item checking unit 4 is used to execute step S4: performing constraint item checking on the task offloading scheme in the updated particle swarm population to obtain a particle swarm population that meets the constraints.

[0142] Specifically, the constraint item check is used to sequentially check and correct the computing resource allocation constraint, the task unique access constraint, and the access point capacity constraint for out-of-bounds, and specifically includes the following sub-steps:

[0143] S401. Check whether the computing frequency allocated by each access point to all users complies with the computing resource allocation constraint. If so, proceed to the next sub-step S402. If not, perform normalization correction processing and proceed to the next sub-step S402. The specific correction is as follows:

[0144]

[0145] S402: Check whether the user selection matrix in the task offloading scheme of all particles meets the task unique access constraint. If so, proceed to the next sub-step S403. If not, further determine the selection of any user i in the user selection matrix of the corresponding particle k, i.e., the selection of the i-th row:

[0146] If user i does not select any access point in the user selection matrix of particle k (i.e., the sum of the elements in this row is 0), the velocity vector corresponding to user i is mapped by the Sigmoid function to obtain the connection probability vector, and the user selection matrix corresponding to the maximum probability is converted to The elements of are set to 1 and the rest are set to 0; for any particle k in the t+1th iteration, the user selection matrix Elements The corrections are as follows:

[0147]

[0148] Where, Represents the maximum index acquisition item, which is used to extract the column index j corresponding to the element with the largest value in the velocity row component after mapping by the sigmoid function; Used to represent the velocity row component corresponding to the i-th row vector of the user selection matrix;

[0149] If a user i selects multiple access points in the particle's user selection matrix (the sum of the elements in this row is greater than 1), the velocity vector corresponding to user i is mapped using the sigmoid function, and the mapping result is compared with the user selection matrix. Perform element-wise multiplication and convert the user-selected matrix corresponding to the maximum value in the product result into The elements of are set to 1, and the rest are set to 0; the element correction of the user-selected matrix for any particle k is expressed as follows:

[0150]

[0151] Then, after the correction is completed, the next sub-step S403 is executed.

[0152] S403: Check whether the user selection matrix in the task offloading scheme of all particles meets the access point capacity constraint. If so, complete the constraint check and set the current particle swarm population to the particle swarm population that meets the constraint. If not, further determine the allocation of any access point j, i.e., the jth column, in the user selection matrix of the corresponding particle k:

[0153] If the number of users assigned to access point j in the user selection matrix of particle k is greater than the maximum number of users allowed by the access point (the sum of the elements in this column exceeds μ j ), the velocity vector corresponding to access point j is mapped by the sigmoid function, and the mapping result is element-wise (Hadamard) multiplied with the user selection matrix to obtain the product result;

[0154] Next, among the users who select access point j, the indexes corresponding to the minimum values of the number of users exceeding the number of allowed users of the access point in the multiplication result are selected as the index set. The specific calculation expression is as follows:

[0155]

[0156] Where, (j) is the index set corresponding to access point j, which is used to represent several row indices of the jth column in the user selection matrix. The total number is Ex, and Ex is the number of users allowed beyond the access point. It is specifically expressed as Represents the minimum index retrieval item, which is used to extract the row index i corresponding to the element with the smallest value in the product result. Since the elements of the velocity vector are usually not strictly zero, and even if they are close to zero, they can maintain a reasonable probability distribution after the Sigmoid mapping, the user's selection of zero elements in the matrix will not affect subsequent constraint modifications.

[0157] Next, the elements corresponding to the index set in the user selection matrix are transferred to the local execution task to complete the access point capacity constraint correction. The specific calculation is as follows:

[0158] and

[0159] Therefore, when modifying the access point capacity constraint, the present invention only applies to users that have been assigned to the current access point (i.e. ), the excess users are screened according to the connection probability after Sigmoid mapping to ensure the rationality of the correction logic and the physical consistency of the allocation action.

[0160] The iteration condition judgment unit 5 is used to execute step S5: determine whether the global optimal task offloading scheme of the particle swarm population that meets the constraints meets the iteration threshold: if so, complete the iteration, and offload the user's computing tasks according to the global optimal task offloading scheme to complete the offloading scheduling of the user's tasks; if not, re-call the fitness calculation unit 2.

[0161] Specifically, the judgment expression of the iteration threshold is as follows:

[0162]

[0163] Where, It represents the absolute value difference between the fitness value of the global optimal task offloading solution of the current iteration and the fitness value of the global optimal task offloading solution of the previous iteration, which is used to judge the change amplitude of the global optimal solution and then judge whether it has converged. ξ is the iteration threshold, which is used to judge whether the global fitness is stable between consecutive iterations. The smaller its value, the stricter the convergence judgment. The default value of the present invention is 0.01, which can also be customized according to the task accuracy requirements. The present invention does not specifically limit it.

[0164] It should be noted that in order to prevent the particle swarm from failing to converge but continuing to run in extreme cases, a maximum number of iterations can be introduced as an additional termination condition on the basis of the above, thereby improving the execution efficiency and stability of the algorithm. The present invention does not specifically limit the selection of the iterative termination condition introduced.

[0165] Compared with the existing technology, the present invention constructs the total system energy consumption and task execution delay as a joint optimization target, and uses the task offloading selection matrix (discrete variables) and the computing resource allocation matrix (continuous variables) as collaborative optimization variables, so that the particle swarm algorithm can simultaneously optimize in the two dimensions of task scheduling and computing resource allocation, thereby generating a high-quality task offloading strategy that can simultaneously meet energy efficiency and delay constraints.

[0166] At the same time, in order to solve the out-of-bounds problem caused by the expansion of the continuous variable range in the collaborative optimization process, and the task allocation overlapping conflict problem caused by the dual constraints of rows and columns in the two-dimensional scheduling matrix for multi-user and multi-access point offloading decisions, the present invention performs out-of-bounds checks and corrections on three types of constraints: computing resource limitations, task unique access, and access point capacity. This significantly improves the feasibility of particle update results and the overall convergence stability of the algorithm in a multi-user and multi-access point MEC environment.

[0167] Please also see Figure 4 , Figure 4 This is a schematic diagram of an example comparison curve of the present invention compared with existing strategies such as full local computing, full remote computing, and random selection. The unloading scheduling method proposed in the present invention always maintains a low level of total system energy consumption when the number of users gradually increases.

[0168] Based on the same inventive concept, the present application also provides an electronic device, which can be a terminal device such as a server, a desktop computing device, or a mobile computing device (e.g., a laptop computing device, a handheld computing device, a tablet computer, a netbook, etc.). The device includes one or more processors and a memory, wherein the processor is used to execute a program to implement the task offloading scheduling method for multi-user multi-access point MEC according to an embodiment of the present invention; and the memory is used to store a computer program executable by the processor.

[0169] Based on the same inventive concept, the present application also provides a computer-readable storage medium, corresponding to the embodiment of the aforementioned multi-user multi-access point MEC task offloading scheduling method, wherein the computer-readable storage medium stores a computer program thereon, and when the program is executed by the processor, the steps of the multi-user multi-access point MEC task offloading scheduling method recorded in any of the above embodiments are implemented.

[0170] The present application may take the form of a computer program product implemented on one or more storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing program code. Computer-usable storage media include permanent and non-permanent, removable and non-removable media, and information storage can be achieved by any method or technology. The information can be computer-readable instructions, data structures, modules of a program, or other data. Examples of computer storage media include but are not limited to: phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, read-only compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, tape disk storage or other magnetic storage devices or any other non-transmission medium that can be used to store information that can be accessed by a computing device.

[0171] The above-described embodiments merely represent several implementations of the present invention. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that a person skilled in the art would be able to make numerous modifications and improvements without departing from the spirit of the present invention, and the present invention is intended to encompass such modifications and variations.

Claims

1. A task offloading scheduling method for multi-user multi-access point MEC, characterized in that: The following steps are involved: S1. Initialize the number of particle swarm individuals, the particle velocity vector, and the task offloading scheme to obtain an initial particle swarm population; wherein the task offloading scheme includes a user selection matrix and a computing resource allocation matrix; S2. Calculate the objective function value consisting of energy consumption and delay penalty based on the cascade channel information and noise power between the user and the access point, and in combination with the current task offloading scheme of the particle swarm population, to obtain the current particle swarm population and the corresponding fitness; wherein the objective function consisting of energy consumption and delay penalty is subject to several constraints, including computing resource limitation constraint, task unique access constraint, access point capacity constraint, and task offloading variable binary constraint; S3, update the particle swarm according to the current particle swarm population and the corresponding fitness to obtain the updated particle swarm population; S4. Performing a constraint check on the task offloading scheme in the updated particle swarm population to obtain a particle swarm population that meets the constraints; wherein the constraint check is used to sequentially perform out-of-bounds checks and corrections on the computing resource allocation constraint, the task unique access constraint, and the access point capacity constraint; S5. Determine whether the global optimal task offloading scheme of the particle swarm population that meets the constraints meets the iteration threshold: if so, complete the iteration, offload the user's computing tasks according to the global optimal task offloading scheme, and complete the offloading scheduling of the user's tasks; if not, re-execute step S2.

2. The task offloading scheduling method for multi-user multi-access point MEC according to claim 1, characterized in that: For the task offloading scheme corresponding to the kth particle in the particle swarm population of the tth iteration The specific calculation of the objective function composed of the corresponding energy consumption and delay penalty is expressed as follows: Where, represents the fitness of the k-th particle corresponding to the task offloading scheme at the t-th iteration, It represents the task offloading scheme of the k-th particle at the t-th iteration, that is, the position vector of the particle, including the user selection matrix and the computing resource allocation matrix, which is specifically expressed as follows: Where, represents the user selection matrix of the kth particle at the tth iteration, whose element u ij Select elements for offloading; M, N represent the total number of users and access points, respectively; is the computational resource allocation matrix for the kth particle at the tth iteration, whose elements assigning elements to resources; represents the total energy consumption of the system corresponding to the kth particle, and its specific calculation is expressed as: Where, E ij represents the energy consumption when the i-th user selects the j-th access point as the task offloading point, α i represents the computational complexity coefficient of the task of the i-th user; k0 represents the local computational energy consumption coefficient; l i represents the task length of the i-th user; f i loc represents the computation frequency of the local device of the i-th user; β i represents the uplink overhead coefficient of the i-th user; P T,i is the transmission power of the i-th user's device; R ij represents the transmission rate between the i-th user and the j-th access point, and its specific calculation is as follows: Where B W,j represents the channel bandwidth allocated to the jth access point; represents the noise power; h ij is the cascade channel information between the i-th user and the j-th access point, that is, the cascade channel gain, which is specifically calculated as follows: Where β0 represents the path loss constant; d ij represents the distance between the i-th user and the j-th access point; h w,ij is the small-scale fast fading channel gain; represents the delay penalty term, λ is the penalty factor; represents the maximum tolerable delay of the task of the i-th user; T i It represents the total execution delay of the i-th user task, and its specific calculation is as follows: Where, T ij represents the connection delay between the i-th user and the j-th access point.

3. The task offloading scheduling method for multi-user multi-access point MEC according to claim 2, characterized in that: For the task offloading scheme corresponding to the kth particle in the particle swarm population of the tth iteration Its update rule at the t+1th iteration is expressed as follows: Where, Indicates whether the i-th user offloads the task to the j-th access point in the t+1-th iteration, that is, the user selection matrix of the k-th particle The corresponding element u ij ; ε is a random variable that obeys uniform distribution in the interval [0,1]; represents the velocity vector corresponding to the kth particle in the particle swarm population at the t+1th iteration, and The element u of the matrix selected by the user ij The corresponding velocity components; is the Sigmoid function; represents the computational resource allocation matrix of the kth particle at the t+1th iteration; The computing resource allocation matrix The corresponding velocity component.

4. The task offloading scheduling method for multi-user multi-access point MEC according to claim 3, characterized in that: For the t-th iteration, the velocity vector corresponding to the k-th particle in the particle swarm population is Its update rule at the t+1th iteration is expressed as follows: Where c1 represents the learning factor of individual extreme value; r1 is the random factor of individual extreme value, which obeys the uniform distribution of [0,1]. Optimal task offloading solution for individuals; c2 represents the learning factor of the group extreme value; r2 represents the random factor of the group extreme value, which obeys the uniform distribution of [0,1]; is the global optimal task offloading solution for the tth iteration; is the adaptive inertia weight of the current k-th particle at the t-th iteration, and its specific calculation is as follows: Where, ω min and ω max is the minimum and maximum value of the inertia weight; represents the maximum fitness of all particles in the tth iteration; It represents the average fitness of all particles in the tth iteration.

5. The task offloading scheduling method for multi-user multi-access point MEC according to claim 4, characterized in that: The constraint item check includes the following sub-steps: S401. Check whether the computing frequency allocated by each access point to all users complies with the computing resource allocation constraint. If so, proceed to the next sub-step S402. If not, perform normalization correction processing and proceed to the next sub-step S402. The specific correction is as follows: Where, represents the maximum available computing frequency resource of the jth access point; S402: Check whether the user selection matrix in the task offloading scheme of all particles meets the task unique access constraint. If so, proceed to the next sub-step S403. If not, further determine the selection of any user i, i.e., the i-th row, in the user selection matrix of the corresponding particle k: If user i does not select any access point in the user selection matrix of particle k, the velocity vector corresponding to user i is mapped by the Sigmoid function to obtain the connection probability vector, and the user selection matrix corresponding to the maximum probability is converted to The elements of are set to 1 and the rest are set to 0; for any particle k in the t+1th iteration, the user selection matrix Elements The corrections are as follows: Where, Indicates the maximum value index to obtain the item; Used to represent the velocity row component corresponding to the i-th row vector of the user selection matrix; If a user i selects multiple access points in the particle's user selection matrix, the velocity vector corresponding to user i is mapped using the sigmoid function, and the mapping result is compared with the user selection matrix. Perform element-wise multiplication and convert the user-selected matrix corresponding to the maximum value in the product result into The elements of are set to 1, and the rest are set to 0; the element correction of the user-selected matrix for any particle k is expressed as follows: Then, after the correction is completed, the next sub-step S403 is executed; S403: Check whether the user selection matrix in the task offloading scheme of all particles meets the access point capacity constraint. If so, complete the constraint check and set the current particle swarm population to the particle swarm population that meets the constraint. If not, further determine the allocation of any access point j, i.e., the jth column, in the user selection matrix of the corresponding particle k: If the number of users assigned to access point j in the user selection matrix of particle k is greater than the maximum number of users allowed by the access point, the velocity vector corresponding to access point j is mapped using the sigmoid function, and the mapping result is element-wise multiplied with the user selection matrix to obtain the product result; Next, among the users who select access point j, the indexes corresponding to the minimum values of the number of users exceeding the number of allowed users of the access point in the multiplication result are selected as the index set. The specific calculation expression is as follows: Where, (j) is the index set corresponding to access point j, the total number of which is Ex, and Ex is the number of users allowed to exceed the access point, which is specifically expressed as Next, the elements corresponding to the index set in the user selection matrix are transferred to the local execution task to complete the access point capacity constraint correction, which is specifically expressed as: and 6. A task offloading scheduling device for multi-user multi-access point MEC, characterized in that: It includes a particle swarm population initialization unit, a fitness calculation unit, a particle swarm update unit, a constraint item checking unit and an iteration condition judgment unit; The particle swarm population initialization unit is used to initialize the number of particle swarm individuals, the velocity vector of the particles, and the task offloading scheme to obtain an initial particle swarm population; wherein the task offloading scheme includes a user selection matrix and a computing resource allocation matrix; The fitness calculation unit is configured to calculate an objective function value consisting of corresponding energy consumption and delay penalty based on the cascade channel information and noise power between the user and the access point, in combination with the current task offloading scheme of the particle swarm population, to obtain the current particle swarm population and the corresponding fitness; wherein the objective function consisting of energy consumption and delay penalty is subject to several constraints, including computing resource limitation constraints, task unique access constraints, access point capacity constraints, and task offloading variable binary constraints; The particle swarm updating unit is used to update the particle swarm according to the current particle swarm population and the corresponding fitness to obtain an updated particle swarm population; The constraint item checking unit is used to perform constraint item checking on the task offloading scheme in the updated particle swarm population to obtain a particle swarm population that meets the constraints; wherein the constraint item checking is used to sequentially perform out-of-bounds checking and correction on the computing resource allocation constraint, the task unique access constraint, and the access point capacity constraint; The iteration condition judgment unit is used to judge whether the global optimal task offloading scheme of the particle swarm population that meets the constraints meets the iteration threshold: if so, the iteration is completed, the user's computing tasks are offloaded and allocated according to the global optimal task offloading scheme, and the offloading scheduling of the user tasks is completed; if not, the fitness calculation unit is re-called.

7. The task offloading scheduling device for multi-user multi-access point MEC according to claim 6, characterized in that: For the task offloading scheme corresponding to the kth particle in the particle swarm population of the tth iteration The specific calculation of the objective function composed of the corresponding energy consumption and delay penalty is expressed as follows: Where, represents the fitness of the k-th particle corresponding to the task offloading scheme at the t-th iteration, It represents the task offloading scheme of the k-th particle at the t-th iteration, that is, the position vector of the particle, including the user selection matrix and the computing resource allocation matrix, which is specifically expressed as follows: Where, represents the user selection matrix of the kth particle at the tth iteration, whose element u ij Select elements for offloading; M, N represent the total number of users and access points, respectively; is the computational resource allocation matrix for the kth particle at the tth iteration, whose elements assigning elements to resources; represents the total energy consumption of the system corresponding to the kth particle, and its specific calculation is expressed as: Where, E ij represents the energy consumption when the i-th user selects the j-th access point as the task offloading point, α i represents the computational complexity coefficient of the task of the i-th user; k0 represents the local computational energy consumption coefficient; l i represents the task length of the i-th user; f i loc represents the computation frequency of the local device of the i-th user; β i represents the uplink overhead coefficient of the i-th user; P T,i is the transmission power of the i-th user's device; R ij represents the transmission rate between the i-th user and the j-th access point, and its specific calculation is as follows: Where B W,j represents the channel bandwidth allocated to the jth access point; represents the noise power; h ij is the cascade channel information between the i-th user and the j-th access point, that is, the cascade channel gain, which is specifically calculated as follows: Where β0 represents the path loss constant; d ij represents the distance between the i-th user and the j-th access point; h w,ij is the small-scale fast fading channel gain; represents the delay penalty term, λ is the penalty factor; represents the maximum tolerable delay of the task of the i-th user; T i It represents the total execution delay of the i-th user task, and its specific calculation is as follows: Where, T ij represents the connection delay between the i-th user and the j-th access point.

8. The task offloading scheduling device for multi-user multi-access point MEC according to claim 7, characterized in that: For the task offloading scheme corresponding to the kth particle in the particle swarm population of the tth iteration Its update rule at the t+1th iteration is expressed as follows: Where, Indicates whether the i-th user offloads the task to the j-th access point in the t+1-th iteration, that is, the user selection matrix of the k-th particle The corresponding element u ij ; ε is a random variable that obeys uniform distribution in the interval [0,1]; represents the velocity vector corresponding to the kth particle in the particle swarm population at the t+1th iteration, and The element u of the matrix selected by the user ij The corresponding velocity components; is the Sigmoid function; represents the computational resource allocation matrix of the kth particle at the t+1th iteration; The computing resource allocation matrix The corresponding velocity components; Among them, the velocity vector corresponding to the kth particle in the particle swarm population of the tth iteration is Its update rule at the t+1th iteration is expressed as follows: Where c1 represents the learning factor of individual extreme value; r1 is the random factor of individual extreme value, which obeys the uniform distribution of [0,1]. Optimal task offloading solution for individuals; c2 represents the learning factor of the group extreme value; r2 represents the random factor of the group extreme value, which obeys the uniform distribution of [0,1]; is the global optimal task offloading solution for the tth iteration; is the adaptive inertia weight of the current k-th particle at the t-th iteration, and its specific calculation is as follows: Where, ω min and ω max is the minimum and maximum value of the inertia weight; represents the maximum fitness of all particles in the tth iteration; It represents the average fitness of all particles in the tth iteration.

9. The task offloading scheduling device for multi-user multi-access point MEC according to claim 8, characterized in that: The constraint item check includes the following sub-steps: S401. Check whether the computing frequency allocated by each access point to all users complies with the computing resource allocation constraint. If so, proceed to the next sub-step S402. If not, perform normalization correction processing and proceed to the next sub-step S402. The specific correction is as follows: Where, represents the maximum available computing frequency resource of the jth access point; S402: Check whether the user selection matrix in the task offloading scheme of all particles meets the task unique access constraint. If so, proceed to the next sub-step S403. If not, further determine the selection of any user i, i.e., the i-th row, in the user selection matrix of the corresponding particle k: If user i does not select any access point in the user selection matrix of particle k, the velocity vector corresponding to user i is mapped by the Sigmoid function to obtain the connection probability vector, and the user selection matrix corresponding to the maximum probability is converted to The elements of are set to 1 and the rest are set to 0; for any particle k in the t+1th iteration, the user selection matrix Elements The corrections are as follows: Where, Indicates the maximum value index to obtain the item; Used to represent the velocity row component corresponding to the i-th row vector of the user selection matrix; If a user i selects multiple access points in the particle's user selection matrix, the velocity vector corresponding to user i is mapped using the sigmoid function, and the mapping result is compared with the user selection matrix. Perform element-wise multiplication and convert the user-selected matrix corresponding to the maximum value in the product result into The elements of are set to 1, and the rest are set to 0; the element correction of the user-selected matrix for any particle k is expressed as follows: Then, after the correction is completed, the next sub-step S403 is executed; S403: Check whether the user selection matrix in the task offloading scheme of all particles meets the access point capacity constraint. If so, complete the constraint check and set the current particle swarm population to the particle swarm population that meets the constraint. If not, further determine the allocation of any access point j, i.e., the jth column, in the user selection matrix of the corresponding particle k: If the number of users assigned to access point j in the user selection matrix of particle k is greater than the maximum number of users allowed by the access point, the velocity vector corresponding to access point j is mapped using the sigmoid function, and the mapping result is element-wise multiplied with the user selection matrix to obtain the product result; Next, among the users who select access point j, the indexes corresponding to the minimum values of the number of users exceeding the number of allowed users of the access point in the multiplication result are selected as the index set. The specific calculation expression is as follows: Where, (j) is the index set corresponding to access point j, the total number of which is Ex, and Ex is the number of users allowed to exceed the access point, which is specifically expressed as Next, the elements corresponding to the index set in the user selection matrix are transferred to the local execution task to complete the access point capacity constraint correction, which is specifically expressed as: and 10. A multi-user multi-access point mobile edge computing system, characterized in that: It includes a base station and a task offloading scheduling device in communication connection with the base station; The base station is provided with a plurality of access points in communication with the edge computing server, which are used to receive offloading requests corresponding to computing tasks from user equipment, and calculate the cascade channel information and noise power between each current user and each access point according to the computing task and its corresponding computing demand, and send the information to the task offloading scheduling device; The task offloading scheduling device is the task offloading scheduling device for multi-user multi-access point MEC according to any one of claims 6 to 9.

Images