A task offloading method and terminal for an ultra-dense edge computing network

Abstract

A task offloading method and terminal for an ultra-dense edge computing network generate corresponding initial offloading decisions and initial channel allocation schemes for each task in the task population, use the golden section algorithm to calculate the optimal upload power of the channel used by each task in the corresponding initial channel allocation scheme, and improve the upload power of the task; update the initial offloading decision corresponding to the task based on the channel state and the type of the task; update the initial channel allocation scheme corresponding to the task according to the channel transmission energy consumption and the effective interference; therefore, it is possible to update the offloading decision based on the current channel state and task type to ensure that the offloaded tasks can be transmitted and processed under good channel conditions, and finally obtain an as-optimal offloading decision as possible to reduce the total energy consumption of task offloading; calculate the optimal offloading decision and optimal channel allocation scheme for each task according to the adaptive simulated annealing genetic algorithm, thereby reducing the energy consumption of task offloading.

Description

Technical Field

[0001] The present invention relates to the technical field of edge computing, and particularly to a task offloading method and a terminal for an ultra-dense edge computing network. Background Art

[0002] According to a Cisco report, the number of global mobile devices will reach 12.3 billion by 2022. With the explosive growth of some mobile devices such as smart cars, mobile phones, and drones, the number and types of mobile applications are also constantly increasing. These mobile applications usually require powerful computing resources to be executed within the response deadline, which poses severe challenges to the computing power and battery life of mobile devices.

[0003] To address this challenge, Mobile Edge Computing (MEC), as a new computing framework, can meet the computing requirements of mobile users for low latency and high reliability. MEC servers are usually deployed close to users to expand the computing power of mobile devices. By offloading computing tasks from mobile devices to the network edge, the computing pressure on mobile devices can be effectively alleviated, and the energy consumption and latency of task offloading can be reduced.

[0004] However, the computing resources of edge servers are limited. It is difficult for MEC under the deployment of traditional cellular wireless networks to meet the access and communication quality requirements of a large number of user devices, and the available radio spectrum resources during the transmission process will also become scarcer. To better address the surging requirements for computing resources and spectrum resources in the 5G and Internet of Things eras, more advanced networking methods and wireless technologies are urgently needed to improve network capacity and increase spectrum utilization.

[0005] To address the above challenges, it is considered to introduce Ultra-Dense Networks (UDNs). UDN mainly encrypts the deployment of small cells on the basis of traditional macrocell coverage, so as to provide sufficient spectrum resources for mobile devices. In addition, by reusing the limited spectrum between cells, the network spectrum utilization and network capacity can be significantly improved. Therefore, by deploying MEC servers on the base stations of UDNs, an ultra-dense edge computing network can be formed. The dense deployment of Small Base Stations (SBSs) can cover more mobile devices, provide access services for them, break the defect of single-layer coverage of traditional macro networks, improve network capacity, and increase spectrum efficiency. The MEC servers connected to the small base stations provide rich computing resources for mobile devices.

[0006] However, there are also a series of challenges in task offloading in ultra-dense edge computing networks. First, when each SBS connected to the MEC server provides access services for multiple mobile devices simultaneously, it is necessary to consider the changes in the channel state while allocating computing resources to different service requests. Second, mobile devices need to select corresponding channels to offload tasks to the MEC server. When a large number of mobile devices select the same channel to compete for limited spectrum resources, inter-device interference will occur, affecting the communication quality. Finally, during the task transmission process, the task execution delay and energy consumption will be affected by the upload power. Summary of the Invention

[0007] The technical problem to be solved by the present invention is to provide a task offloading method and a terminal for an ultra-dense edge computing network, which can reduce the energy consumption of task offloading and increase the task upload power.

[0008] To solve the above technical problems, the technical solution adopted by the present invention is as follows:

[0009] A task offloading method for an ultra-dense edge computing network includes the steps of:

[0010] Generating a task population and generating corresponding initial offloading decisions and initial channel allocation schemes for each task in the task population;

[0011] Using the golden section algorithm to calculate the optimal upload power of the channel used by each task in the corresponding initial channel allocation scheme;

[0012] Updating the initial offloading decision corresponding to the task based on the state of the channel and the type of the task;

[0013] Updating the initial channel allocation scheme corresponding to the task according to the channel transmission energy consumption and the effective interference;

[0014] Calculating the optimal offloading decision and the optimal channel allocation scheme for each task according to the adaptive simulated annealing genetic algorithm.

[0015] To solve the above technical problems, another technical solution adopted by the present invention is as follows:

[0016] A task offloading terminal for an ultra-dense edge computing network includes a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the computer program, it implements a task offloading method for an ultra-dense edge computing network according to the above solution.

[0017] The beneficial effects of the present invention are as follows: generating a task population, generating corresponding initial offloading decisions and initial channel allocation schemes for each task in the task population, using the golden section algorithm to calculate the optimal upload power of the channel used by each task in the corresponding initial channel allocation scheme, and improving the upload power of the task; updating the initial offloading decision corresponding to the task based on the channel state and the task type; updating the initial channel allocation scheme corresponding to the task according to the channel transmission energy consumption and the effective interference; therefore, it is possible to update the offloading decision based on the current channel state and task type to ensure that the offloaded tasks can be transmitted and processed under good channel conditions, and finally obtain an offloading decision that is as optimal as possible, reducing the total energy consumption of task offloading; calculating the optimal offloading decision and optimal channel allocation scheme for each task according to the adaptive simulated annealing genetic algorithm, thereby reducing the energy consumption of task offloading. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 It is a flowchart of a task offloading method for an ultra-dense edge computing network according to an embodiment of the present invention;

[0019] Figure 2 It is a schematic diagram of a task offloading terminal for an ultra-dense edge computing network according to an embodiment of the present invention;

[0020] Figure 3 It is a schematic diagram of the basic architecture of an ultra-dense edge computing network for a task offloading method according to an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0021] To describe in detail the technical content, achieved objectives and effects of the present invention, the following is described in conjunction with the embodiments and with reference to the accompanying drawings.

[0022] Please refer to Figure 1 and Figure 3 , an embodiment of the present invention provides a task offloading method for an ultra-dense edge computing network, including the steps of:

[0023] Generating a task population, and generating corresponding initial offloading decisions and initial channel allocation schemes for each task in the task population;

[0024] Using the golden section algorithm to calculate the optimal upload power of the channel used by each task in the corresponding initial channel allocation scheme;

[0025] Updating the initial offloading decision corresponding to the task based on the channel state and the task type;

[0026] Updating the initial channel allocation scheme corresponding to the task according to the channel transmission energy consumption and the effective interference;

[0027] Calculate the optimal offloading decision and optimal channel allocation scheme for each of the tasks according to the adaptive simulated annealing genetic algorithm.

[0028] As can be seen from the above description, the beneficial effects of the present invention are as follows: generating a task population, generating corresponding initial offloading decisions and initial channel allocation schemes for each task in the task population, using the golden section algorithm to calculate the optimal upload power of the channel used by each task in the corresponding initial channel allocation scheme, and improving the upload power of the task; updating the initial offloading decision corresponding to the task based on the channel state and the type of the task; updating the initial channel allocation scheme corresponding to the task according to the channel transmission energy consumption and effective interference; therefore, it is possible to update the offloading decision based on the current channel state and task type to ensure that the offloaded tasks can be transmitted and processed under good channel conditions, and finally obtain an offloading decision that is as optimal as possible, reducing the total energy consumption of task offloading; calculating the optimal offloading decision and optimal channel allocation scheme for each task according to the adaptive simulated annealing genetic algorithm, thereby reducing the energy consumption of task offloading.

[0029] Further, the use of the golden section algorithm to calculate the optimal upload power of the channel used by each task in the corresponding initial channel allocation scheme includes:

[0030] Calculate the upload power range of the task according to the maximum tolerable delay of the task;

[0031] Use the minimum upload power and the maximum upload power when the task is transmitted through the channel used in the initial channel allocation scheme to narrow the upload power range of the task;

[0032] Judge whether the difference between the maximum upload power and the minimum upload power is greater than a threshold. If it is greater than the threshold, judge whether the transmission energy consumption of the minimum upload power is greater than the transmission energy consumption of the minimum upload power. If so, increase the left boundary value of the upload power range, otherwise, decrease the right boundary value of the upload power range until the difference between the maximum upload power and the minimum upload power is less than or equal to the threshold;

[0033] If it is less than or equal to the threshold, take the average value of the maximum upload power and the minimum upload power as the optimal upload power of the task on the channel.

[0034] As can be seen from the above description, first calculate the upload power range of the task according to the maximum tolerable delay of the task, and then narrow the upload power range according to the maximum and minimum values of the upload power during channel transmission; when the difference between the maximum and minimum values of the upload power is greater than the threshold, narrow the upload power range according to the transmission energy consumption of the maximum and minimum values of the upload power, so as to calculate the optimal upload power according to the left and right boundary values of the upload power range later.

[0035] Further, calculating the upload power range according to the maximum tolerable delay of the task includes:

[0036] The effective interference of the task needs to satisfy the following according to the maximum tolerable delay of the task:

[0037]

[0038] In the formula, EI i,k represents the effective interference when task T i is transmitted on channel k, P i,k represents the power when task T i is transmitted on channel k, S i represents the data volume size of task T i , and W represents the channel bandwidth;

[0039] The range of the upload power obtained according to the formula satisfied by the task effective interference is:

[0040]

[0041] As can be seen from the above description, the conditions that the effective interference of the task needs to satisfy are obtained according to the maximum tolerable delay, and the range of the upload power is calculated, so as to ensure the optimization of the task offloading energy consumption while meeting the task deadline.

[0042] Further, based on the state of the channel and the type of the task, updating the initial offloading decision corresponding to the task includes:

[0043] If the task is not offloaded in the initial offloading decision of the task, it is judged whether the following formula is satisfied:

[0044]

[0045] In the formula, S i represents the data volume of task T i , ω i represents the amount of computation required to complete task T i , Exp(R) represents the expected value of the channel transmission rate, and f m represents the local computing power;

[0046] If so, update the initial offloading decision to offload the task to the server;

[0047] If the task is offloaded to the server in the initial offloading decision of the task, it is judged whether the task meets the preset type. If so, update the initial offloading decision to not offload the task.

[0048] As described above, according to the initial offloading decision of the task, if the task is not offloaded, it is determined whether the task needs to be offloaded based on the channel state, and the initial offloading decision is modified accordingly; if the task is offloaded, it is determined whether the task does not need to be offloaded based on the task type, and the initial offloading decision is modified accordingly. Therefore, updating the offloading decision of the task can reduce the energy consumption of task offloading.

[0049] Furthermore, the calculation of the expected value of the channel transmission rate includes:

[0050] The channel is fitted into a packet loss model, and the state of the channel is detected within a preset time slice. If the channel state is lower than the state threshold, the channel transmission rate is R B , otherwise the channel transmission rate is R G ;

[0051] If the channel state changes from poor to good, the state transition probability is P BG , if the channel state changes from good to poor, the state transition probability is P GB ;

[0052] According to the state transition probability of the channel, calculate the expected value of the channel transmission rate:

[0053]

[0054] As described above, since the channel state is constantly changing during the task transmission process, the quality of the channel state will affect the transmission rate of the current task. Therefore, fitting the channel into a packet loss model and based on the state transition probability of the channel can accurately calculate the expected value of the channel transmission rate.

[0055] Furthermore, the updating of the initial channel allocation scheme corresponding to the task according to the channel transmission energy consumption and effective interference includes:

[0056] According to the effective interference constraint condition of the task, obtain

[0057] Calculate the channel allocation scheme according to the channel transmission energy consumption and effective interference:

[0058]

[0059] In the formula, t i_deadline represents the deadline of task Ti, EI i,k represents the effective interference when task T i is transmitted on channel k, P i,k represents the power when task T i is transmitted on channel k, S i represents task T iThe data volume size, E represents the energy consumption, γ i,k represents the channel allocated to task T i and f s,j represents the computing power of server j.

[0060] As can be seen from the above description, according to the effective interference condition of the task and the channel transmission energy consumption, the channel allocation scheme can be updated by combining the deadline and the dynamic requirements of the mobile device, thereby reducing the energy consumption of task offloading.

[0061] Furthermore, the calculation of the optimal offloading decision and the optimal channel allocation scheme for each of the tasks according to the adaptive simulated annealing genetic algorithm includes:

[0062] Calculating the task population and the fitness value of each task:

[0063]

[0064] where t i_deadline represents the deadline of task T i , E represents the energy consumption, l represents the penalty coefficient, and l = 1 * 10 -2.5 ;

[0065] Performing crossover and mutation on the tasks in the task population and calculating the fitness value fitness new of the tasks after crossover and mutation. If fitness new is less than fitness, then retain the fitness value after crossover and mutation; otherwise, decide whether to retain the fitness value before update according to the annealing probability;

[0066] Calculating the crossover probability and the mutation probability and returning the optimal offloading decision and the optimal channel allocation scheme.

[0067] As can be seen from the above description, the fitness value of the population initialized by the AGA algorithm is used as the initial solution of simulated annealing, and the fitness value after crossover and mutation is used as the new solution of simulated annealing. The offloading scheme corresponding to the new solution obtained according to the update rule of simulated annealing is used as the initial scheme for the next generation of the AGA algorithm. Therefore, combining AGA and SA can overcome each other's deficiencies, not only being superior to the traditional genetic algorithm in terms of efficiency, but also improving the global search ability compared to the SA algorithm and better finding the optimal offloading solution.

[0068] Furthermore, calculating the crossover probability P c includes:

[0069]

[0070] where f average represents the average fitness value, f′ represents the fitness value of the current individual, and f maxrepresents the maximum fitness value, P c1 , P c2 represents the crossover probability, and the value of the crossover probability is P c1 = 0.99, P c2 = 0.88.

[0071] Furthermore, calculating the mutation probability Pm includes:

[0072]

[0073] In the formula, f average represents the average fitness value, f' represents the fitness value of the current individual, f max represents the maximum fitness value, P m1 , P m2 respectively represent the mutation probability, and the values of the mutation probability are P m1 = 0.1, P m2 = 0.01.

[0074] Please refer to Figure 2 , an embodiment of the present invention provides a task offloading terminal for an ultra-dense edge computing network, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the computer program, it implements a task offloading method for an ultra-dense edge computing network of the above solution.

[0075] A task offloading method and terminal for an ultra-dense edge computing network of the present invention are applicable to minimizing the total energy consumption of task offloading within the deadline. Taking the total energy consumption of task offloading of all mobile devices in the ultra-dense edge computing network as the optimization goal, combining AGA (Adaptive Genetic Algorithm) and SA (Simulated Annealing Algorithm) can overcome each other's deficiencies. It is not only superior to the traditional genetic algorithm in terms of efficiency, but also improves the global search ability compared with the SA algorithm, and better finds the optimal offloading solution. The following is illustrated by specific embodiments:

[0076] Embodiment 1

[0077] Please refer to Figure 1 and Figure 3 , a task offloading method for an ultra-dense edge computing network, includes the steps of:

[0078] S1. Generate a task population, and generate corresponding initialization offloading decisions and initialization channel allocation schemes for each task in the task population.

[0079] Specifically, the population fitness values initialized by the AGA algorithm are used as the initial solution of simulated annealing, that is, the corresponding initialization offloading decision x is generated for each task in the population i,j and the initial channel allocation decision γ i,k .

[0080] In this embodiment, each SBS is connected to an edge server, and the edge servers are represented as S = {s1, s2, s3, …, s n}, and the SBSs are deployed in a co-frequency manner. According to the orthogonal frequency division multiplexing technology, the channels of each SBS are divided into C sub-channels and are described as K = {k1, k2, k3, …, k c}, so there is no interference between different channels

[0081] Each user equipment (UE) generates a task. It is assumed that the UE does not move during the task offloading process, and UEs in different cells may reuse the same channel, so there will be interference between different users when transmitting tasks on the same channel. The set of UEs is represented as U = {U1, U2, U3, …, U M}, and the corresponding task set generated by the UE is T = {T1, T2, T3, …, T m}. At the same time, the computing resources of the server and the UE in the present invention are represented as f = {f m , f s1 , f s2 , …, f sn}.

[0082] Since each device generates a task, T i = (ω i , s i , d i , t i_deadline ) is represented as a computing task generated by a UE. Among them, ω i represents the amount of computation required to complete task T i . s i represents the data volume size of each task T i , d i is the distance from the UE generating task T i to the SBS, and t i_deadline represents the deadline for completing task T i . In the ultra-dense network, each UE has a task to complete, so x i,j = {0, 1} is defined to represent the offloading decision of the task. x i,j = 0 means that task T i is not offloaded, while x i,j = 1 means that task T i needs to be offloaded to the server Sj In addition, γ is also defined i,k = {0, 1} represents the channel allocation decision of task T i , so γ i,k = 1 means that task T i is transmitted through channel k, while γ i,k = 0 means there is no channel for transmission.

[0083] When the user UE selects to offload tasks, the interference problem of other UEs in the uplink transmission process needs to be considered. The effective interference of task T generated by user UE i when transmitted on channel k is:

[0084]

[0085] Therefore, the signal-to-noise ratio of task T i when transmitted on channel k is:

[0086]

[0087] By calculation, the transmission rate of task T generated by user UE i when transmitted on channel k is:

[0088] R = Wlog2(1 + EI i,k P i,k );

[0089] where W represents the bandwidth of the subchannel, P i,k represents the upload power of task T i , g i,k represents the channel gain corresponding to task T i when transmitted on channel k, and σ 2 represents the noise power spectral density.

[0090] When the task selects local computing, the device side needs to provide computing resources to process the computing task. Let f m represent the computing power of the device side, then the energy consumption for task processing during local computing is:

[0091]

[0092] The computing time when the task is not offloaded is expressed as:

[0093]

[0094] When selecting offloading computing, there will be corresponding transmission energy consumption and computing energy consumption when transmitting tasks through the wireless network. The transmission energy consumption of the task is:

[0095]

[0096] The task transfer time is expressed as:

[0097]

[0098] The calculated energy consumption of the task during offloading is:

[0099]

[0100] The computing time during task offloading is:

[0101]

[0102] The total energy consumption of the users who select to offload the task to the local device side is denoted as E m , then for all UEs in the ultra-dense network, the total energy consumption of the tasks during local processing is:

[0103]

[0104] Meanwhile, the total energy consumption of the offloading users is denoted as E c , and the total energy consumption of the offloaded tasks among all UEs is:

[0105]

[0106] The total delay of task offloading is:

[0107]

[0108] Since the amount of data transmitted on the downlink is small, the energy consumption and delay of the device side when receiving the return data are ignored.

[0109] S2. Use the golden section algorithm to calculate the optimal upload power of the channel used by each of the tasks in the corresponding initialized channel allocation scheme.

[0110] In this embodiment, in order to minimize the total energy consumption of task offloading within the deadline, the total energy consumption of task offloading of all mobile devices in the ultra-dense edge computing network is used as the optimization objective, which is expressed as:

[0111]

[0112] In the formula, C1 represents the maximum tolerable delay for completing task T i ; C2 represents the task T during the transmission process iThe signal-to-noise ratio of channel k must be greater than the threshold to ensure the reliability of the transmission process; C3 represents the range of the upload power; C4 and C5 ensure that a task can only be offloaded to one server for processing, and at the same time, a task can only be transmitted through one channel; C6 means that the effective interference of the task needs to be greater than the threshold to ensure the task transmission quality; C7 is the condition for channel selection.

[0113] According to the constraint condition C1, the minimum value of the task upload power can be obtained:

[0114]

[0115] Therefore, the range of the upload power is:

[0116]

[0117] Since the upload power mainly affects the transmission energy consumption, the transmission energy consumption is regarded as a function of the upload power, and the golden section algorithm is used to solve the optimal upload power.

[0118] S3. Update the initialized offloading decision corresponding to the task based on the state of the channel and the type of the task.

[0119] S31. If the task is not offloaded in the initialized offloading decision of the task, then determine whether the following formula is satisfied:

[0120]

[0121] In the formula, S i represents the data volume of task T i , ω i represents the amount of computation required to complete task T i , Exp(R) represents the expected value of the channel transmission rate, and f m represents the local computing power;

[0122] If so, update the initialized offloading decision to offload the task to the server.

[0123] Specifically, during the task offloading process, since the channel state is not fixed, when the channel state is poor, the data transmission rate is low, so more transmission energy consumption and delay will be generated; at the same time, as the number of mobile devices accessing the small base station increases, the task offloading energy consumption and delay will also increase. Therefore, according to the current channel state and task attributes, corresponding offloading schemes are formulated. First, when the initialized task chooses not to offload, if it satisfies:

[0124]

[0125] Then the current task offloading scheme needs to be updated to offload to the server.

[0126] S32. If the task is offloaded to the server in the initial offloading decision, then determine whether the task meets a preset type. If so, update the initial offloading decision to not offload the task.

[0127] If the initialized task is selected to be offloaded to the server for processing, then when the task type meets:

[0128]

[0129] Then the current offloading plan needs to be updated, and the task is selected for local processing. This plan continuously updates the offloading decision based on the current channel state and task type to ensure that the offloaded task can be transmitted and processed under good channel conditions, and finally obtains an offloading decision that can meet T i comp +T i comm ≤t i_deadline to reduce the total energy consumption of task offloading.

[0130] Among them, the calculation of the expected value of the channel transmission rate includes:

[0131] Fit the channel into a packet loss model, detect the state of the channel within a preset time slice. If the channel state is lower than the state threshold, then the channel transmission rate is R B , otherwise the channel transmission rate is R G ;

[0132] If the channel state changes from poor to good, then the state transition probability is P BG , if the channel state changes from good to poor, then the state transition probability is P GB ;

[0133] Calculate the expected value of the channel transmission rate according to the state transition probability of the channel:

[0134]

[0135] Specifically, the mobile device transmits data to the MEC server through the wireless channel. When the task is uploaded to the SBS through the wireless channel, considering that the channel state is constantly changing during the actual task transmission process, the quality of the channel state will affect the transmission rate of the current task. To better describe the impact of the current channel state on the current task upload process, the wireless channel is fitted into the Gilbert Elliot model. Assume that the state of the wireless channel detected within the time slice t is g t , when the channel state is poor, the current channel state is described as g B , if the current channel state is very good, the current channel state is described as gG Since the channel state generally affects the transmission rate, the corresponding transmission rate when g t = g G is described as R t = R G . At the same time, the transmission rate when g t = g B is described as R t = R B . In addition, the change of the channel state is described by the state transition probability. When the channel state changes from poor to good, the state transition probability is P BG , and when the channel state changes from good to poor, the state transition probability is described as P GB . Therefore, when the channel state is relatively good, the stable probability of the channel state is P BG / (P BG + P GB ), and the stable probability when the channel state is poor is P GB / (P BG + P GB ). Finally, according to the stable probability of the channel state, the expected value of the channel transmission rate can be obtained as follows:

[0136]

[0137] S4. Update the initialized channel allocation scheme corresponding to the task according to the channel transmission energy consumption and effective interference.

[0138] Among them, according to the effective interference constraint condition of the task,

[0139] Calculate the channel allocation scheme according to the channel transmission energy consumption and effective interference:

[0140]

[0141] In the formula, t i_deadline represents the deadline of task Ti, EI i,k represents the effective interference when task T i is transmitted on channel k, P i,k represents the power when task T i is transmitted on channel k, S i represents the data volume size of task T i , E represents the energy consumption, γ i,k represents the channel allocated to task T i , f s,j represents the computing power of server j.

[0142] Specifically, since each task can only select one channel for transmission, and the energy consumption of task transmission is mainly affected by the transmission rate and channel allocation decision, in order to minimize the energy consumption during task transmission, an optimization scheme for channel allocation decision is given. First, according to the constraint condition C1, the constraint condition C6 of effective interference can be deduced, and we get:

[0143]

[0144] Finally, the following channel allocation scheme is obtained

[0145]

[0146] This scheme means that during task transmission, it is necessary to select the channel allocation decision with the lowest transmission energy consumption and satisfying the constraint condition C6.

[0147] S5. Calculate the optimal offloading decision and optimal channel allocation scheme for each of the said tasks according to the adaptive simulated annealing genetic algorithm.

[0148] Among them, calculate the fitness value of the task population and each of its tasks:

[0149]

[0150] In the formula, t i_deadline represents the deadline of task T i , E represents the energy consumption, l represents the penalty coefficient, l = 1 * 10 -2.5 ;

[0151] Perform crossover and mutation on the tasks in the task population, and calculate the fitness value fitness new of the tasks after crossover and mutation. If fitness new is less than fitness, retain the fitness value after crossover and mutation. Otherwise, decide whether to retain the fitness value before update according to the annealing probability;

[0152] Calculate the crossover probability and mutation probability, and return the optimal offloading decision and optimal channel allocation scheme.

[0153] Specifically, during the initialization process, first use the possible computing offloading schemes as the chromosomes in the genetic algorithm. Take the position corresponding to each task as the gene of the chromosome, then each chromosome has a total of M genes. The encoding method is integer encoding. Each task can choose local processing and offload to any one server, and each task can choose one of the corresponding channels for transmission. At the same time, in order to measure the adaptability of each population to the living environment, the fitness value can also be used to judge and eliminate the populations with worse adaptability. The fitness value is:

[0154]

[0155] Among them, when the task completion time exceeds the maximum deadline of the task, a penalty function is added on the basis of energy consumption, where l = 1 * 10 -2.5 is the penalty coefficient, and the fitness function value is the sum of the energy consumption and the penalty function value. When the task completion time is less than the deadline, the fitness value is set to the energy consumption value. t i_deadline represents the deadline of a single task, and we calculate the deadline in the following way:

[0156]

[0157] Among them, Deadline l = T Z represents the task completion time under the AGASA algorithm, while Deadline u = 3T Z , k1 is the deadline ratio, and its value range is from 0 to 1. In the selection process of the AGASA algorithm, in order to prevent the loss of the optimal individual in the next generation, resulting in the inability to converge to the optimal solution, a selection strategy combining binary tournament and elite selection is mainly adopted. First, the optimal individual in the current population is reserved and enters the offspring; secondly, two individuals are selected from the population with a certain selection probability, and the individual with the lower fitness value is added to the offspring until all individuals have completed the tournament. In the crossover process, two crossover points are respectively selected in the first half and the second half of the parent chromosome for two-point crossover to form the offspring chromosome. In the mutation process, a single-point mutation strategy is selected. After the crossover and mutation are completed, it is necessary to compare the fitness values of the population before and after the mutation. If the fitness value before the mutation is lower, then the individual before the mutation is accepted with a certain annealing probability, and the annealing probability is set as:

[0158]

[0159] We use the fitness value after crossover and mutation as the final fitness value fitness new . In addition, the crossover probability and the mutation probability have a great impact on the convergence of the algorithm. Although the population is more likely to generate new individuals when the crossover probability is larger, when it becomes larger, the retention rate of excellent individuals in the population also decreases; for the mutation probability, if it is too large, the algorithm is equivalent to an ordinary random algorithm, losing the meaning of the AGASA algorithm. Therefore, in order to improve the performance of the algorithm, the crossover probability P c and the mutation probability P m are obtained in the following way:

[0160]

[0161] In the formula, f averagedenotes the average fitness value, f′ denotes the fitness value of the current individual, and f max denotes the maximum fitness value, and P c1 、P c2 denotes the crossover probability, and the value of the crossover probability is P c1 = 0.99, and P c2 = 0.88; P m1 、P m2 respectively denote the mutation probability, and the value of the mutation probability is P m1 = 0.1, and P m2 = 0.01.

[0162] Example Two

[0163] The main difference between this example and Example One is that it further defines how to calculate the optimal upload power of a task in a channel, specifically:

[0164] Calculate the upload power range of the task according to the maximum tolerable delay of the task;

[0165] Narrow the upload power range of the task by using the minimum upload power and the maximum upload power when the task is transmitted through the channels used in the initialization channel allocation scheme;

[0166] Judge whether the difference between the maximum upload power and the minimum upload power is greater than the threshold. If it is greater than the threshold, then judge whether the transmission energy consumption of the minimum upload power is greater than the transmission energy consumption of the minimum upload power. If so, increase the left boundary value of the upload power range, otherwise, decrease the right boundary value of the upload power range until the difference between the maximum upload power and the minimum upload power is less than or equal to the threshold;

[0167] If it is less than or equal to the threshold, then take the average value of the maximum upload power and the minimum upload power as the optimal upload power of the task on the channel.

[0168] In this example, the upload power is also one of the main factors affecting the transmission energy consumption. The higher the upload power, the farther the task can be transmitted. However, once the upload power is too high, more energy consumption will be generated during the transmission process; therefore, to solve the problem of upload power control, a golden section algorithm is adopted to determine the optimal upload power:

[0169] 1. According to the value range of the upload power it can be determined as the initial left boundary of the upload power, and at the same time b i = P max as the initial right boundary of the upload power.

[0170] To further narrow down the search range of the upload power, it is necessary to first calculate the minimum upload power when the task is transmitted through channel k and the maximum upload power

[0171] 2. By judging whether the boundary of the upload power is narrowed down to a certain range. If this condition is met, the upload power is directly calculated according to If this condition is not met, it is necessary to further compare whether it meets the requirement by calculating the transmission energy consumption If this condition is met, the left boundary of the upload power needs to be further narrowed. If this condition is not met, the right boundary of the power needs to be further narrowed until the condition is met

[0172] 3. According to the optimal upload power is output. Where τ = 0.618, this algorithm can continuously narrow down the search range and ensure that the optimal upload power of each task is obtained within a certain power range.

[0173] Embodiment III

[0174] Please refer to Figure 2 , a task offloading terminal for a hyper-dense edge computing network, including a memory, a processor, and a computer program stored on the memory and operable on the processor. When the processor executes the computer program, each step of a task offloading method for a hyper-dense edge computing network in Embodiment I or II is implemented.

[0175] In summary, a task offloading method and terminal for a hyper-dense edge computing network provided by the present invention generate a task population, generate corresponding initial offloading decisions and initial channel allocation schemes for each task in the task population, use the golden section algorithm to calculate the optimal upload power of the channel used by each task in the corresponding initial channel allocation scheme, and improve the upload power of the task; based on the channel state and task type, update the initial offloading decisions corresponding to the tasks; update the initial channel allocation schemes corresponding to the tasks according to the channel transmission energy consumption and effective interference; combining the task type and the computing power of the server, and considering the channel state change, the dynamic requirements of the mobile device, and the interference constraint, a task offloading method based on adaptive simulated annealing genetic is proposed, which optimizes the task offloading energy consumption while meeting the task deadline. In addition, in order to obtain a better upload power, the present invention uses the golden section algorithm to solve the power control problem. Experimental results show that the proposed task offloading strategy has better performance than other classical methods.

[0176] The above are only embodiments of the present invention, and do not thereby limit the patent scope of the present invention. Any equivalent transformation made by using the content of the specification and drawings of the present invention, or directly or indirectly applied in the relevant technical fields, shall equally be included in the patent protection scope of the present invention.

Claims

1. A task offloading method for an ultra-dense edge computing network, characterized in that, Including the steps: Generate a task population, and generate a corresponding initial offloading decision and an initial channel allocation scheme for each task in the task population; Use the golden section algorithm to calculate the optimal upload power of the channel used by each task in the corresponding initial channel allocation scheme; Update the initial offloading decision corresponding to the task based on the state of the channel and the type of the task; Update the initial channel allocation scheme corresponding to the task according to the channel transmission energy consumption and the effective interference; Calculate the optimal offloading decision and the optimal channel allocation scheme for each task according to the adaptive simulated annealing genetic algorithm.

2. The task offloading method of an ultra-dense edge computing network according to claim 1, characterized in that The use of the golden section algorithm to calculate the optimal upload power of the channel used by each task in the corresponding initial channel allocation scheme includes: Calculate the upload power range of the task according to the maximum tolerable delay of the task; Use the minimum upload power and the maximum upload power when the task is transmitted through the channel used in the initial channel allocation scheme to narrow the upload power range of the task; Judge whether the difference between the maximum upload power and the minimum upload power is greater than the threshold. If it is greater than the threshold, judge whether the transmission energy consumption of the minimum upload power is greater than the transmission energy consumption of the minimum upload power. If so, increase the left boundary value of the upload power range, otherwise, decrease the right boundary value of the upload power range until the difference between the maximum upload power and the minimum upload power is less than or equal to the threshold; If it is less than or equal to the threshold, take the average value of the maximum upload power and the minimum upload power as the optimal upload power of the task on the channel.

3. The task offloading method of an ultra-dense edge computing network according to claim 2, characterized in that Calculating the upload power range of the task according to the maximum tolerable delay of the task includes: Obtain that the effective interference of the task needs to satisfy according to the maximum tolerable delay of the task: Wherein, EI i,k represents the effective interference when task T i is transmitted on channel k, and P i,k represents the power when task T i is transmitted on channel k, S i represents the data volume size of task T i , and W represents the channel bandwidth; Obtain the upload power range according to the formula satisfied by the task effective interference:

4. The task offloading method of an ultra-dense edge computing network according to claim 1, characterized in that, Updating the initial offloading decision corresponding to the task based on the state of the channel and the type of the task includes: If the task is not offloaded in the initial offloading decision of the task, judge whether the following formula is satisfied: Where S i represents the data volume of task T i , ω i represents the amount of computation required to complete task T i , Exp(R) represents the expected value of the channel transmission rate, and f m represents the local computing power; If so, update the initial offloading decision to offload the task to the server; If the task is offloaded to the server in the initial offloading decision of the task, judge whether the task meets the preset type. If so, update the initial offloading decision to not offload the task.

5. The task offloading method of an ultra-dense edge computing network according to claim 4, characterized in that, The calculation of the expected value of the channel transmission rate includes: Fit the channel to a packet loss model, detect the state of the channel within a preset time slice. If the channel state is lower than the state threshold, the transmission rate of the channel is R B , otherwise the transmission rate of the channel is R G ; If the channel state changes from poor to good, the state transition probability is P BG , if the channel state changes from good to poor, the state transition probability is P GB ; Calculate the expected value of the channel transmission rate according to the state transition probability of the channel; 6. The task offloading method of an ultra-dense edge computing network according to claim 1, characterized in that The update of the initial channel allocation scheme corresponding to the task according to the channel transmission energy consumption and the effective interference includes: Based on the effective interference constraint conditions of the task, obtain Calculate the channel allocation scheme according to the channel transmission energy consumption and the effective interference; where t i_deadline represents the deadline of task Ti, EI i,k represents the effective interference when task T i is transmitted on channel k, P i,k represents the power when task T i is transmitted on channel k, S i represents the data volume of task T i , E represents the energy consumption, γ i,k represents the channel allocated to task T i , and f s,j represents the computing power of server j.

7. A task offloading method for an ultra-dense edge computing network according to claim 1, characterized in that The calculation of the optimal offloading decision and the optimal channel allocation scheme for each task according to the adaptive simulated annealing genetic algorithm includes: Calculate the fitness value of the task population and each task in it; where t i_deadline represents the deadline of task T i , E represents the energy consumption, l represents the penalty coefficient, and l = 1*10 -2.5 ; Perform crossover and mutation on the tasks in the task population, and calculate the fitness value fitness after crossover and mutation new , if fitness new is less than fitness, retain the fitness value after crossover and mutation; otherwise, decide whether to retain the fitness value before update according to the annealing probability; Calculate the crossover probability and the mutation probability, and return the optimal offloading decision and the optimal channel allocation scheme.

8. The task offloading method of an ultra-dense edge computing network according to claim 7, characterized in that, Calculate the crossover probability P c including: In the formula, f average represents the average fitness value, f' represents the fitness value of the current individual, and f max represents the maximum fitness value. P c1 and P c2 represent the crossover probability. The value of the crossover probability is P c1 = 0.99, and P c2 = 0.

88.

9. The task offloading method of an ultra-dense edge computing network according to claim 7, characterized in that Calculate the mutation probability P m Including: In the formula, f average represents the average fitness value, f’ represents the fitness value of the current individual, and f max represents the maximum fitness value. P m1 and P m2 represent the mutation probabilities respectively. The values of the mutation probabilities are P m1 = 0.1 and P m2 = 0.

01.

10. A task offloading terminal for an ultra-dense edge computing network, characterized in that, It includes a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the computer program, it implements the task offloading method of an ultra-dense edge computing network according to any one of claims 1-9 above.

Images