An edge computing offloading method based on user reputation

By introducing user reputation model and latency and energy consumption preference factors into the mobile edge computing system, combining dichotomy and particle swarm optimization algorithm, the problem of low resource utilization and difficult to guarantee user fairness is solved, and the system overhead is achieved. Improve and improve user experience.

CN115499877BActive Publication Date: 2024-11-22ZHENGZHOU UNIV +2
View PDF 5 Cites 0 Cited by

Patent Information

Application Number
CN202211172935.9
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2022-09-26
Publication Date
2024-11-22
Estimated Expiration
2042-09-26

AI Technical Summary

Technical Problem

The existing mobile edge computing (MEC) systems have low resource utilization, difficult to guarantee user fairness, and neglecting malicious competition and transmission power optimization, resulting in low computing resource allocation efficiency and large system overhead.

Method used

The credibility model is constructed based on user credibility, introduce delay and energy consumption preference factors, and the power distribution sub-model and offload decision sub-model is used to optimize transmission power and offload decisions by decomposing and computing models. The transmission power and offload decisions are optimized by using dichotomy and particle swarm optimization algorithms, and the transmission power and offload decisions are allocated reasonably Compute resources to prevent users from malicious behavior and minimize system overhead.

Benefits of technology

Through quasi-convex optimization technology and particle swarm optimization algorithm, system overhead is minimized, user experience quality is improved, system overhead is reduced, and reasonable and reliable allocation of resources is ensured and optimal transmission power and offloading strategies are ensured.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN115499877B_ABST
    Figure CN115499877B_ABST
Patent Text Reader

Abstract

The present invention proposes an edge computing unloading method based on user reputation, and the steps are as follows: first, an edge computing system based on user reputation is built, including terminal users and base stations configured with MEC servers; and a reputation model is built according to the historical interaction information between MU and MEC server, and MEC server allocates computing resources according to user reputation; secondly, based on the delay and energy consumption generated by allocating computing resources, a computing model is built by introducing user delay and energy consumption preference factors; and the computing model is decomposed into a power allocation sub-model and an unloading decision sub-model; finally, for a given unloading decision, the power allocation sub-model is optimized based on a high-efficiency power allocation algorithm of dichotomy to obtain the transmission power; and based on the transmission power, the unloading decision sub-model is optimized using an improved particle swarm optimization algorithm to obtain an unloading decision with minimized system overhead. The present invention can effectively improve the user experience quality and reduce system overhead.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of edge computing technology, and in particular to an edge computing offloading method based on user credibility. Background Art

[0002] With the continuous development of mobile Internet, many new applications have emerged, such as virtual reality, augmented reality, and interactive games. However, mobile devices with limited computing power can hardly meet the latency requirements of such applications. In recent years, the proposal of mobile edge computing (MEC) has provided a new idea for solving the above problems. Compared with traditional cloud computing, MEC has the advantages of low latency and low energy consumption, and has become a research hotspot in recent years.

[0003] At present, many scholars have conducted in-depth research on offloading strategies in MEC. For example, the literature [FENG MJ, KRUNZ M, and ZHANG WH, Joint Task Partitioning and User Association for Latency Minimization in Mobile Edge Computing Networks [J]. IEEE Transactions on Vehicular Technology, 2021, 70 (8): 8108-8182.] studied the joint optimization problem of task partitioning and user association in MEC systems, and proposed a user association scheme based on dual decomposition to minimize the average time of all user tasks. The literature [REN JK, YU GD, CAI YL, et al. Latency optimization for resource allocation in mobile-edge computation offloading [J]. IEEE Transactions on Wireless Communications, 2018, 17 (8): 5506-5519.] proposed a piecewise convex optimization method based on subgradient optimization theory, which effectively solved the problem of minimizing computing task latency. The paper [LIU X, LIU J and WUH. Energy-Efficient Task Allocation of Heterogeneous Resources in Mobile Edge Computing [J]. IEEE Access, 2021, 9: 119700-119711.] studied the task allocation problem and proposed an energy-saving offloading scheme based on a heuristic algorithm. The paper [DU JB, ZHAO LQ, FENG J, et al. Computation offloading and resource allocation in mixed fog / cloud computing systems with min-max fairness guarantee [J]. IEEE Transactions on Communications, 2017, 66 (4): 1594-1608.] proposed a low-complexity resource allocation and offloading decision based on user fairness to minimize system energy consumption under the guarantee of delay constraints.Reference [Jia Shuxia, Hao Wanming, Gao Zihan, Yang Shouyi. Resource allocation and task offloading strategy in mobile edge computing system with task migration [J]. Telecommunication Technology, 2022, 62(08): 1037-1043.] A Q-learning-based method is proposed to minimize latency and energy consumption. Reference [YU BC, ZHANG XJ, YOU I, et al. Efficient Computation Offloading in Edge Computing Enabled Smart Home [J]. IEEE Access, 2021, 9: 48631-48639.] A branch-and-bound algorithm is proposed to minimize the cost and energy consumption of mobile devices. Reference [MAO YY, ZHANG J, ang LETAIEF K B. Joint task offloading scheduling and transmit power allocation for mobile edge computing systems [C]. IEEE Wireless Communications and Networking Conference, San Francisco, USA 2017: 1-6.] An alternating iterative low-complexity optimization algorithm is proposed to reduce the latency and energy consumption required for computing tasks. The literature [WANG P, LI KL, XIAO B, et al. Multi-objective Optimization for Joint Task Offloading, Power Assignment, and Resource Allocation in Mobile Edge Computing [J]. IEEE Internet of Things Journal, 2021.] proposes a multi-objective optimization strategy that balances users' energy consumption, delay performance, and cost overhead.The literature [KAI CH, ZHOU H, YI YB, et al. Collaborative Cloud-Edge-End Task Offloading in Mobile-Edge Computing Networks With Limited Communication Capability[J]. IEEE Transactions on Cognitive Communications and Networking, 2021, 7(2): 624-634.] studied the problem of cloud-edge-end collaborative partial offloading and proposed a pipeline-based latency optimization solution.

[0004] At present, although MEC-related research has made some progress, there are still the following shortcomings. The first is low resource utilization: the impact of possible selfish behavior of users on the network is not considered; the second is that user fairness is difficult to guarantee: when users have the same priority in resource allocation, it is impossible to distinguish between good and bad users. Moreover, the existing resource allocation scheme ignores the problem of malicious competition among users, making it difficult to achieve efficient configuration of computing resources. In addition, during computing offloading, user transmission power is one of the main factors affecting offloading transmission delay and energy consumption, and power optimization is often ignored in current research. Summary of the invention

[0005] In view of the shortcomings of the above-mentioned background technology, the present invention proposes an edge computing offloading method based on user credibility, introduces user delay and energy consumption preference factors, considers the MEC system based on user credibility under multi-user competition, and prevents user malicious behavior, optimizes the overall resource configuration, and minimizes system overhead by reasonably allocating computing resources, optimizing transmission power and offloading decisions.

[0006] The technical solution of the present invention is achieved in this way:

[0007] An edge computing offloading method based on user reputation, the steps are as follows:

[0008] Step 1: Build an edge computing system based on user reputation, including end users and base stations equipped with MEC servers; randomly distribute N mobile end users (MUs) within the coverage of the base station, and the set of all MUs is defined as

[0009] Step 2: Build a reputation model based on the historical interaction information between MU and MEC server, and MEC server allocates computing resources according to user reputation;

[0010] Step 3: Based on the latency and energy consumption generated by allocating computing resources, a computing model is constructed by introducing user latency and energy consumption preference factors;

[0011] Step 4: Decompose the calculation model into a power allocation sub-model and an unloading decision sub-model;

[0012] Step 5: For a given unloading decision, the power allocation sub-model is optimized based on the efficient power allocation algorithm of the dichotomy method to obtain the transmission power p n ; Then based on the transmission power p n , the improved particle swarm optimization algorithm is used to optimize the offloading decision sub-model to obtain the offloading decision a with the minimum system overhead. n .

[0013] The reputation model is:

[0014]

[0015] in, MEC server to MU n The direct credibility, EP is the incentive penalty factor, ψ n (t) is the MU of the tth round n Credibility;

[0016] The expression of the incentive penalty factor EP is:

[0017]

[0018] Among them, λ is the reward factor for MU, θ is the penalty factor for MU, Indicates the tth round MU n The number of successful interactive perception task results to the MEC server, Indicates the tth round MU n The number of failed interactive perception task results to the MEC server; t represents the current interaction round, z represents the zth interaction round, γ is the attenuation parameter, and p z is a 0-1 variable;

[0019] The MU of the tth round n The credibility of n The expression of (t) is:

[0020]

[0021] in, is the satisfaction correction factor, e k ∈[0,1] is the evaluation value of the influencing factor k in the MU interactive perception task, w k is the evaluation weight.

[0022] The method by which the MEC server allocates computing resources according to user reputation is:

[0023]

[0024] in, Indicates that the MEC server is assigned to the MU n The computing resources of MEC server are represented by F. for The normalized value of .

[0025] The method for constructing the calculation model is:

[0026] The computing task parameter of the nth MU is expressed as Λ n ={C n ,Q n}, where C n Indicates the number of CPU cycles required to complete the computing task, Q n Indicates the size of the input computing task data;

[0027] MU n The computing power is expressed as The local calculation delay and local computing energy consumption Can be expressed as:

[0028]

[0029]

[0030] Among them, the value of κ depends on the chip structure of MU;

[0031] If multi-carrier transmission technology is used for transmission, MU n The upload rate R n (p n ) can be expressed as:

[0032]

[0033] Where B represents the channel bandwidth, p n Indicates MU n The transmission power, h n Indicates MU n Channel gain with the base station, σ 2 represents the noise power;

[0034] MU n Unloading task delay It can be expressed as:

[0035]

[0036] MEC server processing delay It can be expressed as:

[0037]

[0038] in, Indicates that the MEC server is assigned to the MU n computing resources;

[0039] Ignore the delay in returning the result, MU n Total latency of task offloading It can be expressed as:

[0040]

[0041] MEC servers are powered directly by cables, and MU n Offloaded transmission energy consumption It can be expressed as:

[0042]

[0043] MU n The cost function is defined as the weighted sum of the latency and energy consumption to complete the computing task, that is:

[0044]

[0045] in, Indicates MU n Delay, Represents the energy consumption preference factor and satisfies By reasonably optimizing resources and offloading strategies to minimize system overhead, the calculation model can be expressed as:

[0046]

[0047]

[0048]

[0049]

[0050] Among them, a is the MU offloading decision set, p is the transmit power set of the offloaded MU, and the constraint condition C 1 Indicates that the uninstall decision is a 0-1 decision; C 2 Indicates that the frequency allocated to MU cannot exceed the maximum frequency of the server; C 3 Indicates the MU transmission power range.

[0051] The power allocation sub-model is:

[0052]

[0053]

[0054] make The power distribution sub-model can be reformulated as:

[0055]

[0056]

[0057] g n (p n ) is:

[0058]

[0059] Among them, φ n =Q n *B, It can be seen that g' n (p n ) depends entirely on the numerator on the right side of the equal sign. Let:

[0060]

[0061] The efficient power allocation algorithm based on dichotomy optimizes the power allocation sub-model to obtain the transmission power p n The method is:

[0062] Input Parameters ,θ n ,φ n ,γ n ,ε,calculation

[0063] if Otherwise, let p h =0, Recalculate z n (p n );

[0064] p m =(p h +p t ) / 2;

[0065] If z n (p n )≤0,p h =p m , otherwise, p t =p m ;

[0066] When (p t -p h )≤ε, the output transmission power is optimal

[0067] The improved particle swarm optimization algorithm is:

[0068] V particles form a population S, S = {X 1 ,X 2 ,…,X V}, the search space is m-dimensional, each particle represents a feasible solution, all particles update their speed and position according to the searched group extreme value and individual extreme value, and the position vector of the kth particle is represented by X k ={x k1 ,x k2 ,…,x km}, the velocity vector of the kth particle is represented by V k = {v k1 ,v k2 ,…,v km}, the individual optimal solution of the kth particle is expressed as The global optimal solution of the entire particle swarm is expressed as Then the calculation process of the position and velocity update of the kth particle in the gth iteration is:

[0069]

[0070]

[0071] Among them, w is the inertia weight, c 1 、c 2 are acceleration factors, r 1 、r 2 is a random number between [0,1];

[0072] The updating method of inertia weight w is:

[0073] w=W max -(W max -W min )*g / G;

[0074] Among them, W max Indicates the maximum inertia weight, W min represents the minimum inertia weight, and G represents the maximum number of iterations;

[0075] The updating method of the acceleration factor is:

[0076]

[0077]

[0078] Among them, c 11 Indicates c 1 The initial value of c 12Indicates c 1 The end value, c 21 Indicates c 2 The initial value of c 22 Indicates c 2 The end value of .

[0079] Compared with the prior art, the present invention has the following beneficial effects:

[0080] 1) The present invention solves the problem of minimizing system overhead through quasi-convex optimization technology and particle swarm-based optimization offloading algorithm, achieves reasonable and reliable resource allocation, and obtains the optimal transmission power and offloading strategy.

[0081] 2) The present invention can effectively improve user experience quality and reduce system overhead. BRIEF DESCRIPTION OF THE DRAWINGS

[0082] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0083] Figure 1 This is the MEC system model based on user credibility of the present invention.

[0084] Figure 2 The relationship between user reputation and the number of uninstallation rounds.

[0085] Figure 3 This is the relationship between system overhead and the number of users.

[0086] Figure 4 This is the relationship between the system average delay and energy consumption and the delay preference factor.

[0087] Figure 5 This is the relationship between system overhead and the size of user input data. DETAILED DESCRIPTION

[0088] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0089] The embodiment of the present invention provides an edge computing offloading method based on user reputation, and the specific steps are as follows:

[0090] Step 1: Build an edge computing system based on user reputation, including end users and base stations configured with MEC servers, such as Figure 1 As shown in the figure, N mobile users (MU) are randomly distributed within the coverage area of ​​the base station, and the set of all MUs is defined as MU can offload computing tasks to nearby MEC servers to reduce its own computing burden. In return, MU needs to receive some lightweight sensing tasks assigned by MEC servers, such as temperature and humidity. We count the historical success rate of MUs in performing sensing tasks to evaluate the credibility of MUs. Each MU has a computing task to complete, and we define a n ∈{0,1} is the user's uninstall decision, a n =0 means the task is executed locally, a n =1 indicates that the task is offloaded to the edge server for execution. 1 ,a 2 ,…,a n} represents the offloading decision of all MUs.

[0091] Step 2: Build a reputation model based on the historical interaction information between MU and MEC server, and MEC server allocates computing resources according to user reputation;

[0092] The user reputation model of the present invention is constructed according to the historical interaction information between MU and MEC server, the success rate of MU in executing perception tasks, etc., and the Bayesian theory method is used to evaluate the direct reputation of MU n and MEC server.

[0093] First, assume that there are x interaction histories between MU n and the MEC server, and the results of the interactions are evaluated using binomial events in probability statistics. During each interaction, MU provides the MEC server with the results of the perception task, such as local temperature and road conditions. Assuming that the number of successful interactions is α and the number of failed interactions is β, according to the Bayesian method, the initial credibility of MU n is defined as the probability of the x+1th successful interaction. p=P(S) represents the probability that MU will successfully interact the perception results to the MEC server, where the prior probability p follows a uniform distribution, and the prior distribution is denoted as π(p). The probability formula for proposition X is "In the existing x historical interactions, the number of successful interactions of the perception results is α times" is expressed as follows:

[0094]

[0095]

[0096] The updated information of proposition X is represented by the posterior success probability density function. It is deduced that the posterior probability density function of p obeys Beta(α+1,β+1). Assuming proposition Y is "In the existing x historical interactions, the number of successful interactions of perception results is α times and the x+1th time is successful", the expected value of the probability density function of the successful interaction perception result of MU in the proposition can be used to represent the prediction of the MEC server on the future successful interaction information of MU n, that is, the MEC server's reputation ψ of MU n n It can be expressed as:

[0097]

[0098] According to the historical interaction information between MU and MEC server, in the interaction process between MU and MEC server, considering the influence of external factors, deviations may occur in some MU interaction processes. Evaluating user credibility based on only a single factor of historical interaction success rate may cause evaluation errors. Based on the above reasons, the present invention introduces a satisfaction correction factor s(t) to correct formula (3). The formula is defined as:

[0099]

[0100] Among them, e k ∈[0,1] is the evaluation value of the influencing factor k in the MU interactive perception task, w k is the evaluation weight, w k ∈[0,1]. In the construction of the dynamic long-term reputation model, it is assumed that MU does not move within a certain range, there are a large number of computationally intensive tasks to be processed, and MU computing tasks are continuously performed in different rounds t. The dynamic long-term reputation model of MU is established, and the long-term reputation evaluation method considering the interactive round is added to formula (3) as shown in (5):

[0101]

[0102] Among them, α(t) represents the number of times MU n successfully interacts with the MEC server to perceive the task results in the tth round, and β(t) represents the number of failures in the tth round.

[0103] When considering dynamic long-term reputation, a time decay factor is introduced to reflect the dynamic change of user reputation with the interaction rounds, which is defined as:

[0104]

[0105] Among them, t represents the current interaction round, z represents the zth interaction round, and γ is the decay parameter. At the same time, in order to facilitate the use of the Beta density function of credibility considerations, p is introduced z Variable, p z=1 indicates that the interaction perception result between MU and MEC server is successful, p z =0 indicates failure. When the attenuation factor γ is close to 1, it means that the calculation of credibility takes into account the success / failure of all historical rounds of unloading as much as possible; when γ is close to 0, it means that the calculation of credibility takes into account the recent unloading behavior of MU more. In this paper, γ is set to 0.99. Due to the multiplication of the attenuation parameters, the behavior of MU 10 rounds ago has little effect on the calculation of credibility in the simulation experiment and is set to be negligible. That is, in this paper, the calculation of credibility takes into account the success / failure of MU's recent 10 rounds of unloading to the MEC server.

[0106] Substituting formula (6) into formula (3) for MU n, we can get the credibility of MU n in the tth round as follows:

[0107]

[0108] According to the interaction of the historical offloading perception task results between MU and MEC server, this paper introduces a reward and punishment mechanism into the constructed reputation model to calculate the long-term reputation of MU n ψ n (t), if the number of interaction failures of the perception task result is less than the average value, a penalty factor is introduced, otherwise a reward factor is introduced, as shown in formula (8):

[0109]

[0110] Among them, λ is the reward factor for MU, and θ is the penalty factor for MU. After the introduction of the reward and punishment mechanism, the value of user reputation is still between [0,1]. According to the reward and punishment mechanism, the direct reputation of the MEC server to MU n can be expressed as:

[0111]

[0112] In practical applications, to facilitate comparison between MUs, the long-term credibility of MU n is normalized and expressed as:

[0113]

[0114] Therefore, a more fair and reasonable resource allocation method is that the MEC server allocates computing resources based on user reputation.

[0115]

[0116] in, Indicates that the MEC server is assigned to the MU n computing resources, and F represents the computing capacity of the MEC server.

[0117] Step 3: Based on the latency and energy consumption generated by allocating computing resources, a computing model is constructed by introducing user latency and energy consumption preference factors;

[0118] The calculation task parameter of the nth MU is expressed as Λ n ={C n ,Q n}, where C n Indicates the number of CPU cycles required to complete the computing task, Q n Indicates the size of the input computing task data;

[0119] MU n The computing power is expressed as The local calculation delay and local computing energy consumption Can be expressed as:

[0120]

[0121]

[0122] The value of κ depends on the chip structure of MU; κ = 10 (-26) .

[0123] In order to avoid multi-user interference and frequency selective fading, multi-carrier transmission technology is adopted, that is, the calculation data of different MUs are transmitted through different carriers, so MU n The upload rate R n (p n ) can be expressed as:

[0124]

[0125] Where B represents the channel bandwidth, p n Indicates MU n The transmission power, h n Indicates MU n Channel gain with the base station, σ 2 represents the noise power;

[0126] MU n Unloading task delay It can be expressed as:

[0127]

[0128] MEC server processing delay It can be expressed as:

[0129]

[0130] in, Indicates that the MEC server is assigned to the MU n computing resources;

[0131] Since the amount of feedback data is much smaller than the amount of input data, the delay of the result return is usually negligible. n Total latency of task offloading It can be expressed as:

[0132]

[0133] The MEC server is directly powered by the cable, so the energy consumption of MEC is not considered. n Offloaded transmission energy consumption It can be expressed as:

[0134]

[0135] MU n The cost function is defined as the weighted sum of the latency and energy consumption to complete the computing task, that is:

[0136]

[0137] in, Indicates MU n Delay, Represents the energy consumption preference factor and satisfies MU can flexibly adjust the preference factor to meet the task requirements. By reasonably optimizing resources and unloading strategies to minimize system overhead, the calculation model can be expressed as:

[0138]

[0139]

[0140]

[0141]

[0142] Among them, a is the MU offloading decision set, p is the transmit power set of the offloaded MU, and the constraint condition C 1 Indicates that the uninstall decision is a 0-1 decision; C 2 Indicates that the frequency allocated to MU cannot exceed the maximum frequency of the server; C 3 Indicates the MU transmission power range.

[0143] Step 4: Decompose the calculation model into a power allocation sub-model and an unloading decision sub-model;

[0144] In order to optimize resource allocation, improve user experience quality and reduce system overhead, a computing offloading scheme based on user reputation is proposed. This scheme divides the system overhead minimization problem into two parts: power allocation and offloading decision. First, under the initial offloading decision, the optimal configuration of computing resources is achieved based on user reputation, and the transmission power is optimized by combining the binary search method to reduce transmission delay and energy consumption; finally, the offloading problem is constructed as a multi-party NP (non-deterministic polynomially) problem, and the optimization offloading algorithm based on particle swarm is used to minimize the system overhead.

[0145] The variables that need to be optimized in the above problem (20) include the transmission power and the MU unloading decision. n is a continuous variable, and the unloading decision a n is an integer variable, so the problem can be reduced to MINLP (mixed integer nonlinear programming). To solve the above problem, it can be decomposed into two related sub-problems. First, for a given offloading decision, optimize the transmission power p n ; Then, based on this, the MU unloading decision a n Optimization, and finally multiple iterations to minimize system overhead.

[0146] Step 5: For a given unloading decision, the power allocation sub-model is optimized based on the efficient power allocation algorithm of the dichotomy method to obtain the transmission power p n ; Then based on the transmission power p n , the improved particle swarm optimization algorithm is used to optimize the offloading decision sub-model to obtain the offloading decision a with the minimum system overhead. n .

[0147] Given the unloading strategy, the optimization problem is transformed into solving the optimal transmission power. Unloading MU n, About p n The cost minimization optimization problem (i.e., the power allocation submodel) can be expressed as:

[0148]

[0149]

[0150] make It can be seen that it is only related to power, and the optimization problem can be reformulated as:

[0151]

[0152]

[0153] Due to g n(p n ) The second-order derivative in the feasible domain is not always positive, so the problem in equation (22) is a non-convex fractional optimization problem. In the literature [TRAN TX, POMPILI D. Joint task offloading and resource allocation for multi-server mobile-edge computing networks [J]. IEEE Transactions on Vehicular Technology, 2018, 68 (1): 856-868.], this optimization problem has been proved to be a quasi-convex problem.

[0154] g n (p n ) is:

[0155]

[0156] Among them, φ n =Q n *B, It can be seen that g' n (p n ) depends entirely on the numerator on the right side of the equal sign. Let:

[0157]

[0158] Because z' n (p n )>0,z n (0)<0, according to the first-order optimality condition, the only global optimal solution of the problem in equation (22) exists, and the optimal solution of the problem Obtain or satisfy the constraint boundary Therefore, an efficient power allocation algorithm based on dichotomy is proposed.

[0159] Algorithm 1: Efficient power allocation algorithm based on binary search

[0160] Input Parameters ,θ n ,φ n ,γ n ,ε,calculation

[0161] if Otherwise, let p h =0, Recalculate z n (p n );

[0162] p m =(ph +p t ) / 2;

[0163] If z n (p n )≤0,p h =p m , otherwise, p t =p m ;

[0164] When (p t -p h )≤ε, the output transmission power is optimal

[0165] After obtaining the optimal transmission power, the offloading decision problem is still an NP problem. Therefore, the particle swarm optimization (PSO) algorithm is used to solve it in order to obtain the offloading decision that minimizes the system overhead.

[0166] The PSO algorithm is a swarm intelligence optimization algorithm that imitates the foraging of bird flocks. Assume that a population S is composed of V particles, S = {X 1 ,X 2 ,…,X V}, the search space is m-dimensional, each particle represents a feasible solution, all particles update their speed and position according to the searched group extreme value and individual extreme value, and the position vector of the kth particle is represented by X k ={x k1 ,x k2 ,…,x km}, the velocity vector of the kth particle is represented by V k = {v k1 ,v k2 ,…,v km}, the individual optimal solution of the kth particle is expressed as The global optimal solution of the entire particle swarm is expressed as Then the calculation process of the position and velocity update of the kth particle in the gth iteration is:

[0167]

[0168]

[0169] Among them, w is the inertia weight, c 1 、c 2 are acceleration factors, r 1 ,r 2 is a random number between [0,1].

[0170] The PSO algorithm has achieved good results in solving many problems. However, it is easy for the PSO algorithm to fall into the local optimal solution of system overhead in the process of multi-dimensional optimization decision-making. Therefore, the algorithm is improved from two aspects: inertia weight and acceleration factor to optimize the decision-making ability and reduce system overhead.

[0171] Improved 1 inertia weight

[0172] According to formula (22), the value of w has a certain impact on the experimental optimization results. By changing the fixed inertia weight in the standard PSO to a dynamic adaptive inertia weight, it can avoid falling into the local optimum to a certain extent, and better balance the global search and local search, and improve the optimization system overhead capability. In the early stage of the algorithm, the value of w is large, and a good global search can be performed through a large step size; in the later stage of the algorithm, the value of w is dynamically adjusted to enhance the local search capability. The inertia weight is dynamically adjusted as the number of iterations changes, weighing the global optimization and local optimization to obtain better decision-making ability, as shown in formula (27).

[0173] w=W max -(W max -W min )*g / G (27)

[0174] Among them, W max =0.9 is the maximum inertia weight, W min =0.4 is the minimum value of inertia weight, and G represents the maximum number of iterations.

[0175] Improved 2 speedup factor

[0176] In PSO, the acceleration factor c 1 and c 2 Controls the convergence direction and speed of particles. The acceleration factor of standard PSO is constant, but in the multidimensional decision model, when the number of iterations changes, the dominance of global search and local search is different. At the beginning of the algorithm, the number of iterations is small, c 1 The larger value is dominant, which can effectively search for local solutions. As the number of algorithm iterations increases, the value of the acceleration factor is dynamically adjusted to obtain the global optimal solution of the particle. 1 and c 2 The calculation of is as shown in equation (28) and equation (29).

[0177]

[0178]

[0179] Among them, c 11 and c 12 c 1 The initial and final values ​​of c 21 and c22 c 2 The initial and final values ​​of 11 =2.5, c 12 =0.5, c 21 =0.5, c 22 =2.5.

[0180] Algorithm 2: Calculation and uninstallation algorithm based on user reputation

[0181] Enter u n ,Λ n , B,m,V,W min ,W max ,p max ,ε,G;

[0182] Output transmission power set p, unloading decision set a;

[0183] 1) Initialize the particle's speed and position, that is, initialize the unloading decision a 0 , system overhead i=0;

[0184] 2) while i≤100 do;

[0185] 3) Update resource allocation based on user reputation;

[0186] 4) Solve the optimal transmission power of the unloaded MU in the current unloading decision set through Algorithm 1;

[0187] 5) Calculate the fitness value of each particle;

[0188] 6) Find individual extreme values and group extreme value

[0189] 7) Update the velocity and position of the particle;

[0190] 8) Update and That is, update offloading decisions and system overhead;

[0191] 9) i = i + 1;

[0192] 10) end while;

[0193] 11) Output the optimal offloading decision and system overhead.

[0194] The proposed algorithm is simulated and analyzed through the MATLAB simulation platform, and the simulation parameters are shown in Table 1. In order to evaluate the effectiveness of the proposed offloading scheme based on user reputation calculation, it is compared with local calculation, full MEC, and random offloading schemes.

[0195] Table 1 Simulation parameters

[0196]

[0197]

[0198] First, the long-term dynamic reputation model of users is verified, and the initial reputations of three MUs are defined as 0.5, 0.6, and 0.7 respectively. Under the influence of the reward and punishment mechanism and the fading factor, the first MU is set to have ten rounds of perception task interaction results that are always successful; the second MU perception task interaction results are generated randomly, and the number of successes is greater than the number of failures; the third MU interaction always fails. The reputation of the user's long-term dynamic reputation model constructed according to the present invention in the 10th round is as follows Figure 2 shown.

[0199] from Figure 2 It can be seen that the reputation of the user whose interactive perception task results have been successful has increased the fastest, from 0.5 to 0.85. The reputation of the second MU has increased slowly, from 0.6 to 0.79, because the number of successful interactions is slightly higher than the number of failed interactions. The reputation of the third MU has dropped rapidly from 0.7 to 0.14 due to continuous failures.

[0200] Figure 3 The relationship between the system overhead of four different offloading schemes and the number of users is compared. t and λ e are all set to 0.5. It can be found that with the increase in the number of users, the system overhead of the four offloading schemes all shows an upward trend. The local computing power is weak, so the system overhead is the largest. For the other three schemes, under the constraints of the server's computing resources, the increase in the number of users will lead to an increase in the amount of computing tasks, and the user computing resource competition will be more intense, which will increase the system overhead. When the number of users is small, the difference between the full offloading scheme and the method of the present invention is not much. At this time, the computing resources of the MEC server are sufficient, and users can save overhead by offloading all the calculations to the MEC server. Finally, by comparing the performance of the four offloading schemes, it can be found that the proposed scheme has the best performance.

[0201] from Figure 4 The relationship between the average delay and energy consumption of the system under different delay preference factors is compared, and the number of MUs is set to 50. Delay preference factor λ t From 0.1 to 0.9, the energy consumption preference factor is λ e =1-λ t It can be seen that as the delay preference factor λ t With the increase of , the average delay of MU gradually decreases, while the average energy consumption of MU gradually increases. This is because the reduction of average delay comes at the cost of consuming more energy.

[0202] Figure 5 The relationship between the system overhead and the size of user input data under four different offloading schemes is compared. t and λ e are all set to 0.5, and the number of MUs is set to 50. It can be found that the system overhead increases with the increase of user input data. This is because when the number of users and server computing resources remain unchanged, the increase in the amount of tasks leads to an increase in the latency and energy consumption required for computing tasks, so the system overhead also increases. Compared with the comparison scheme, the proposed scheme always has a lower system overhead. This is because the computing offloading scheme based on user reputation achieves reasonable and reliable resource allocation while obtaining the optimal offloading strategy.

[0203] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. An edge computing offloading method based on user reputation, It is characterized in that The steps are as follows: Step 1: Build an edge computing system based on user reputation, including end users and base stations equipped with MEC servers; randomly distribute N mobile end users (MUs) within the coverage of the base station, and the set of all MUs is defined as Step 2: Build a reputation model based on the historical interaction information between MU and MEC server, and MEC server allocates computing resources according to user reputation; The reputation model is: in, MEC server to MU n The direct credibility, EP is the incentive penalty factor, ψ n (t) is the MU of the tth round n Credibility; The expression of the incentive penalty factor EP is: Among them, λ is the reward factor for MU, θ is the penalty factor for MU, Indicates the tth round MU n The number of successful interactive perception task results to the MEC server, Indicates the tth round MU n The number of failed interactive perception task results to the MEC server; t represents the current interaction round, z represents the zth interaction round, γ is the attenuation parameter, and p z is a 0-1 variable; The MU of the tth round n The credibility of n The expression of (t) is: in, is the satisfaction correction factor, e k ∈[0,1] is the evaluation value of the influencing factor k in the MU interactive perception task, w k is the evaluation weight; The method by which the MEC server allocates computing resources according to user reputation is: in, Indicates that the MEC server is assigned to the MU n The computing resources of MEC server are represented by F. for The normalized value of Step 3: Based on the latency and energy consumption generated by allocating computing resources, a computing model is constructed by introducing user latency and energy consumption preference factors; The method for constructing the calculation model is: The computational task parameter of the nth MU is expressed as Λ n ={C n ,Q n }, where C n Indicates the number of CPU cycles required to complete the computing task, Q n Indicates the size of the input computing task data; MU n The computing power is expressed as The local calculation delay and local computing energy consumption Can be expressed as: Among them, the value of κ depends on the chip structure of MU; If multi-carrier transmission technology is used for transmission, MU n The upload rate R n (p n ) can be expressed as: Where B represents the channel bandwidth, p n Indicates MU n The transmission power, h n Indicates MU n Channel gain with the base station, σ 2 represents the noise power; MU n Unloading task delay It can be expressed as: MEC server processing delay It can be expressed as: in, Indicates that the MEC server is assigned to the MU n computing resources; Ignore the delay in returning the result, MU n Total latency of task offloading It can be expressed as: MEC servers are powered directly by cables, and MU n Offloaded transmission energy consumption It can be expressed as: MU n The cost function is defined as the weighted sum of the latency and energy consumption to complete the computing task, that is: in, Indicates MU n Delay, Represents the energy consumption preference factor and satisfies By reasonably optimizing resources and offloading strategies to minimize system overhead, the calculation model can be expressed as: Among them, a is the MU offloading decision set, p is the transmit power set of the offloaded MU, and the constraint condition C 1 Indicates that the uninstall decision is a 0-1 decision; C 2 Indicates that the frequency allocated to MU cannot exceed the maximum frequency of the server; C 3 Indicates the MU transmission power range; Step 4: Decompose the calculation model into a power allocation sub-model and an unloading decision sub-model; Step 5: For a given unloading decision, the power allocation sub-model is optimized based on the efficient power allocation algorithm of the dichotomy method to obtain the transmission power p n ; Based on the transmission power p n , the improved particle swarm optimization algorithm is used to optimize the offloading decision sub-model to obtain the offloading decision a with the minimum system overhead. n ; The power allocation sub-model is: make The power distribution sub-model can be reformulated as: g n (p n ) is: Among them, φ n =Q n *B, It can be seen that g' n (p n ) depends entirely on the numerator on the right side of the equal sign. Let: The efficient power allocation algorithm based on dichotomy optimizes the power allocation sub-model to obtain the transmission power p n The method is: Input Parameters θ n ,φ n ,γ n ,ε,calculation if Otherwise, let p h =0, Recalculate z n (p n ); p m =(p h +p t ) / 2; If z n (p n )≤0,p h =p m , otherwise, p t =p m ; When (p t -p h )≤ε, the output transmission power is optimal The improved particle swarm optimization algorithm is: V particles form a population S, S = {X 1 ,X 2 ,…,X V }, the search space is m-dimensional, each particle represents a feasible solution, all particles update their speed and position according to the searched group extreme value and individual extreme value, and the position vector of the kth particle is represented by X k ={x k1 ,x k2 ,…,x km }, the velocity vector of the kth particle is represented by V k = {v k1 ,v k2 ,…,v km }, the individual optimal solution of the kth particle is expressed as The global optimal solution of the entire particle swarm is expressed as Then the calculation process of the position and velocity update of the kth particle in the gth iteration is: Among them, w is the inertia weight, c 1 、c 2 are acceleration factors, r 1 、r 2 is a random number between [0,1]; The updating method of inertia weight w is: w=W max -(W max -W min )*g / G; Among them, W max Indicates the maximum inertia weight, W min represents the minimum inertia weight, and G represents the maximum number of iterations; The updating method of the acceleration factor is: Among them, c 11 Indicates c 1 The initial value of c 12 Indicates c 1 The end value, c 21 Indicates c 2 The initial value of c 22 Indicates c 2 The end value of .

Citation Information

Patent Citations

  • Unloading scheduling method for edge cloud computing task

    CN111796880A

  • Task unloading and resource allocation method in uncertain network environment

    CN113242568A

  • Task unloading method and device for mobile edge computing system

    CN114489977A

  • Mobile edge network non-cooperative game forwarding method based on user credibility

    CN114666793A

  • Quality of service (QOS) control in mobile edge computing (MEC)

    US20210076250A1