An Edge Computing Offloading Method Based on Genetic Algorithm in a Multi-Access Network Slicing Scenario

Abstract

The present invention discloses an edge computing offloading method based on a genetic algorithm in a multi-access network slicing scenario, including: First, initialize the model to obtain the task volume and delay constraints of all users, the computing capabilities of all fog access points and cloud servers, and the channel gains of all channels; Subsequently, create access network slice instances, including a large-scale Internet of Things slice and a low-latency and high-reliability slice, and set offloading strategies for different slices; Initialize the population according to the user offloading strategies in different slices, and optimize and eliminate the individuals in the initialized population; Then set the iteration step size as t and start the iteration; First, perform a selection operation on the individuals by the roulette wheel selection method; Then perform a single-point crossover operation on the selected individuals; After that, perform a mutation operation on the individuals after crossover; Then determine whether the iteration step size t is reached. If not, repeat the above selection, crossover, and mutation operations; If so, output the individual with the highest fitness value as the offloading strategy.

Description

Technical Field

[0001] The present invention relates to the field of edge computing and network slicing, and particularly to an edge computing offloading method based on a genetic algorithm in a multi-access network slicing scenario. Background Art

[0002] With the advent of the era of Internet of Everything, the number of terminals has increased explosively. The fifth-generation mobile communication network (5G) also proposes three major application scenarios, namely: enhanced mobile broadband (eMBB); ultra-reliable and low-latency communication (uRLLC); massive machine type communication (mMTC). In this case, a single network can no longer meet the service quality requirements of all scenarios. At the same time, the continuous development of application programs brings a better experience to users, but also brings a heavy burden to terminal devices.

[0003] Edge computing and network slicing technologies are considered to be the key means to solve the above problems. Among them, edge computing technology can set up edge servers with caching and computing capabilities closer to users, effectively reducing transmission latency while reducing the burden on the core network and terminal devices; network slicing technology can customize personalized slice instances for different scenarios to meet the service requirements of different networks.

[0004] Therefore, combining edge computing and network slicing technologies to study the problems of multi-user computing offloading and resource allocation in the presence of multiple network slices has promising prospects. Summary of the Invention

[0005] The purpose of the present invention is to provide an edge computing offloading method based on a genetic algorithm in a multi-access network slicing scenario, aiming at minimizing the system energy consumption, and optimizing the task offloading strategy of users and the allocation of computing resources and wireless resources.

[0006] To achieve the above purpose, the present invention adopts the following technical solutions:

[0007] An edge computing offloading method based on a genetic algorithm in a multi-access network slicing scenario, comprising the steps of:

[0008] Step S1, initializing the model, obtaining the task amounts and latency constraints of all users, obtaining the computing capabilities of all edge servers and cloud servers, and obtaining the channel gains of all channels; the system consists of multiple channels, multiple users, multiple edge servers and one cloud server. The computing capacity of the edge server has an upper limit. The algorithm runs on the cloud server to obtain resource allocation and offloading scheduling decisions; users are limited by their own devices and need to offload tasks to the edge server through wireless channels. When the edge server reaches its own computing capacity limit, the excess tasks need to be forwarded to the cloud server for calculation;

[0009] Step S2: Create access network slice instances, including a massive IoT slice and a low-latency and high-reliability slice, and set offloading strategies for different slices. Among them, the users in the massive IoT slice are fixed devices that are not sensitive to latency, such as sensors and cameras. The users in the low-latency and high-reliability slice are mobile phones, vehicle-to-everything (V2X) devices, etc. Such users have high requirements for latency.

[0010] Step S3: Initialize the population according to the offloading strategies of users in different slices, and optimize and eliminate the individuals in the initialized population.

[0011] Step S4: Set the iteration step size as t and start the iteration.

[0012] Step S5: Perform a selection operation on the individuals through the roulette wheel selection method.

[0013] Step S6: Perform a single-point crossover operation on the selected individuals.

[0014] Step S7: Perform a mutation operation on the individuals after crossover.

[0015] Step S8: Determine whether the iteration step size t is reached. If not, repeat the above selection, crossover, and mutation operations. If so, output the individual with the highest fitness value as the offloading strategy.

[0016] Further, in the above-mentioned Step S1, the sets of users and edge servers are respectively represented as: and The set of all OFDM channels is represented by .

[0017] Further, the offloading strategies of the network slice instances created in the above-mentioned Step S2 are respectively: For the users in the low-latency and high-reliability slice offload the tasks to a suitable edge server through the wireless channel. Since the latency constraints of such users are strict, the tasks in this slice will only be computed within the edge server and will not be forwarded to the cloud server, so as to reduce the possible latency. For the users in the massive IoT slice also need to first offload the tasks to the edge server through the wireless channel. When the computing power of the edge server reaches the upper limit, since such users are not sensitive to latency, the tasks in this slice will be preferentially forwarded to the cloud server for computing.

[0018] Further, in the above-mentioned Step S3, the symbolic encoding of the population individuals is specifically represented as: For each gene value on the individual, it is represented as the offloading strategy of a single user (a i , b i , c i ); where represents the channel selection variable of the i-th user; Denote the edge server selection variable of the \(i\)th user; \(c\) i \(\in \{0, 1\}\) represents the task cloud forwarding variable. If it is 0, it means the task is not forwarded to the cloud server; if it is 1, it means the task is forwarded to the cloud server for calculation; the symbolic encoding of the population individuals is \(\{(a_1, b_1, c_1), (a_2, b_2, c_2), \ldots, (a\) U , b U , c U )\}.

[0019] Furthermore, the method for optimizing and eliminating the initialized population in the step S3 is specifically as follows:

[0020] Step S301: Check whether the generated individual meets the requirements; first, judge whether all user channel selections \(a\) i in the individual are repeated to prevent channel multiplexing. If so, eliminate the individual; if not, proceed to the next judgment;

[0021] Step S302: Judge whether the transmission delay of all users in the individual meets the requirements, that is, whether the transmission rate is greater than the minimum transmission rate \(R\) min . If not, eliminate the individual; if so, proceed to the next judgment;

[0022] Step S303: Judge whether all edge servers in the individual exceed the calculation limit. If so, eliminate the individual; if not, proceed to the next optimization;

[0023] Step S304: According to the offloading strategy specified in step S2, set the cloud forwarding variable \(c\) of all users in the low-latency and high-reliability slices to 0, prohibiting task offloading to the cloud server.

[0024] Furthermore, in step S302, the transmission rate of the user is calculated as follows:

[0025] The transmission rate expression of the \(i\)th user is:

[0026]

[0027] In formula (1), \(P\) i represents the transmission power of user \(i\), and \(N_0\) is the noise power; represents the channel gain when user \(i\) communicates with the \(b\) i edge server using the \(a\) i channel; \(B\) is the bandwidth of the channel.

[0028] Furthermore, in step S5, the fitness evaluation function is constructed, which is specifically expressed as:

[0029]

[0030] Among them It represents the energy consumption of the computing offloading of user i in the low-latency and high-reliability slice, specifically:

[0031]

[0032] Among them It is the energy consumption when user i uses the wireless channel for transmission, specifically:

[0033]

[0034] Among them, Ω i is the task volume of user i;

[0035] Among them It is the energy consumption of the task of user i calculated by the edge server, specifically:

[0036]

[0037] Among them is the computing constant of the b i th edge server. ω represents the machine cycles required to compute 1 bit, represents the CPU frequency of the b i th edge server;

[0038] Among them It represents the energy consumption of the computing offloading of user i in the large-scale Internet of Things slice, expressed as:

[0039]

[0040] Among them is the energy consumption during the process of forwarding the task to the cloud server, specifically:

[0041]

[0042] Among them, μ is the energy transmission constant;

[0043] Among them is the energy consumption of the task calculated by the cloud server, specifically:

[0044]

[0045] Among them, ε c and λ c represent the computing constant and CPU cycles of the cloud server.

[0046] Furthermore, in step S5, the roulette wheel selection method is used to select individuals in the population according to the fitness evaluation function, specifically:

[0047] Step S501: Calculate the cumulative probability of each individual:

[0048]

[0049] where M is the total number of individuals in the population, and F j represents the fitness value of the j-th individual;

[0050] Step S502: Generate random numbers r1 and r2 in the range of 0 to 1;

[0051] Step S503: If P x-1 ≤ r1 ≤ P x and P y-1 ≤ r2 ≤ P y , then individuals x and y are selected;

[0052] Step S504: Determine whether individuals x and y are the same individual. If so, repeat steps S502 - S504; if not, output individuals x and y.

[0053] Further, in step S6, perform a crossover operation on the selected individuals, specifically:

[0054] Step S601: Randomly generate an integer k ∈ {1, 2,..., N - 1};

[0055] Step S602: Perform crossover on individuals x and y at the k-th gene value to generate new individuals and

[0056] Step S603: Determine whether each edge server in the new individuals and exceeds the calculation limit. If so, delete the individual and repeat steps S601 - S603; if not, output the new individuals.

[0057] Further, in step S7, perform a mutation operation on the individuals after crossover, specifically:

[0058] Step S701: Set the mutation probability α ∈ [0, 1];

[0059] Step S702: Traverse each bit in the new individuals, generate a random number β ∈ [0, 1]. If β ≤ α, then mutate this bit; otherwise, keep it unchanged;

[0060] Step S703: Determine whether the channel selection of the mutated individuals is repeated, whether the delay of all users meets the requirements, and whether the computing power of all edge servers exceeds the upper limit. If any one of the judgments is yes, repeat steps S701 - S703; if all are no, output the new mutated individuals and

[0061] Step S704: Calculate and compare the original individuals x and y, and the fitness values of the new individuals and as well as the new individuals after mutation and are obtained. Select the two individuals with the highest fitness values and put them back into the population.

[0062] Compared with the prior art, the present invention combines network slicing with edge computing technology. In the scenario of multi-access network slicing, a genetic algorithm is used to offload the tasks of users in different network slices to edge servers or cloud servers for calculation, thereby minimizing the total energy consumption of the system and having considerable application prospects.

[0063] Specifically, the present invention encodes the offloading strategy as population individuals, continuously optimizes the strategy with iterations, and finally effectively reduces the total energy consumption of the system. The network model for computing offloading in the multi-access network scenario provided by the present invention innovatively combines access network slicing with edge computing technology, meeting the quality of service requirements of different users. BRIEF DESCRIPTION OF THE DRAWINGS

[0064] Figure 1 is a schematic flowchart of an edge computing offloading method based on a genetic algorithm in a multi-access network slicing scenario provided in Embodiment 1;

[0065] Figure 2 is a network model diagram of computing offloading in a multi-access network scenario provided in Embodiment 1. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0066] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Apparently, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0067] Embodiment 1

[0068] Referring to Figure 1 - Figure 2 , this embodiment provides an edge computing offloading method based on a genetic algorithm in a multi-access network slicing scenario, aiming to minimize the system energy consumption and optimize the task offloading strategy of users and the allocation of computing resources and wireless resources. The method includes the following steps:

[0069] Step 1: Initialize the model to obtain the task volume and latency constraints of all users, the computing capabilities of all edge servers and cloud servers, and the channel gains of all channels. The system consists of multiple channels, multiple users, multiple edge servers, and one cloud server. The computing capabilities of the edge servers have an upper limit, and the algorithm runs on the cloud server to obtain resource allocation and offloading scheduling decisions. Users are restricted by their own devices and need to offload tasks to the edge servers through wireless channels. When the edge servers reach their computing capacity limits, the excess tasks need to be forwarded to the cloud server for computing.

[0070] Step 2: Create access network slice instances, including a large-scale Internet of Things slice and a low-latency and high-reliability slice, and set offloading strategies for different slices. Among them, the users in the large-scale Internet of Things slice are fixed devices that are not sensitive to latency, such as sensors and cameras. The users in the low-latency and high-reliability slice are mobile phones, vehicle-to-everything devices, etc. Such users have high requirements for latency.

[0071] Step 3: Initialize the population according to the offloading strategies of the users in different slices, and optimize and eliminate the individuals in the initialized population.

[0072] Step 4: Set the iteration step size as t and start the iteration.

[0073] Step 5: Perform a selection operation on the individuals through the roulette wheel selection method.

[0074] Step 6: Perform a single-point crossover operation on the selected individuals.

[0075] Step 7: Perform a mutation operation on the individuals after crossover.

[0076] Step 8: Determine whether the iteration step size t is reached. If not, repeat the above selection, crossover, and mutation operations. If so, output the individual with the highest fitness value as the offloading strategy.

[0077] Specifically, in this embodiment, the network model of user computing offloading is as Figure 2 shown. Assume that there are U users, F edge servers, and one cloud server, and N OFDM channels in the entire network. The sets of users and edge servers are respectively represented as: and The set of all channels is represented by .

[0078] Specifically, in this embodiment, after obtaining the information of all users, edge servers, and channels, it is necessary to create network slice instances and allocate users to different slices, and set offloading strategies for them. The specific offloading strategies are as follows: For the users in the low-latency and high-reliability slice Tasks are offloaded to appropriate edge servers via wireless channels. Since the latency constraints of such users are strict, tasks within this slice will only be computed within the edge servers and will not be forwarded to the cloud servers, thereby reducing the possible latency; users within the large-scale Internet of Things slice also need to offload tasks to the edge servers via wireless channels. After the computing capacity of the edge servers reaches the upper limit, since such users are not sensitive to latency, tasks within this slice will be preferentially forwarded to the cloud servers for computing.

[0079] Specifically, in this embodiment, before creating a population, symbolic encoding needs to be performed on the population individuals, specifically expressed as: for each gene value on an individual, it represents the offloading strategy of a single user (a i , b i , c i ); where represents the channel selection variable of the i-th user; represents the edge server selection variable of the i-th user; c i ∈{0, 1} represents the task cloud forwarding variable. If it is 0, it means the task is not forwarded to the cloud server; if it is 1, it means the task is forwarded to the cloud server for computing; the symbolic encoding of the population individuals is {(a1, b1, c1), (a2, b2, c2),..., (a U , b U , c U )}.

[0080] Specifically, in this embodiment, after the population is initialized, the method for optimizing and eliminating individuals in the population is as follows:

[0081] Step 301: Check whether the generated individuals meet the requirements; first, judge whether all user channel selections a i in the individual are repeated to prevent channel multiplexing. If so, eliminate the individual; if not, proceed to the next judgment;

[0082] Step 302: Judge whether the transmission latency of all users in the individual meets the requirements, that is, whether the transmission rate is greater than the minimum transmission rate R min . If not, eliminate the individual; if so, proceed to the next judgment;

[0083] Step 303: Judge whether all edge servers in the individual exceed the computing upper limit. If so, eliminate the individual; if not, proceed to the next optimization;

[0084] Step 304: According to the offloading strategy specified in step S2, set the cloud forwarding variable c of all users in the low-latency and high-reliability slice to 0, prohibiting tasks from being offloaded to the cloud server.

[0085] Specifically, in this embodiment, when determining whether the transmission delay of all users meets the requirements, it is necessary to calculate the transmission rate of the users. The specific calculation method is as follows:

[0086] The transmission rate expression of the i-th user is:

[0087]

[0088] Where P in formula (1) i represents the transmission power of user i, and N0 is the noise power; represents that user i uses a i channel to communicate with the b i edge server, and B is the bandwidth of the channel.

[0089] Specifically, in this embodiment, it is necessary to construct an individual fitness evaluation function to judge the pros and cons of each offloading strategy. Here, the evaluation function is expressed as the reciprocal of the total energy consumed by the system by this individual, and is specifically expressed as:

[0090]

[0091] Where represents the energy consumption of the computing offloading of user i in the low-latency and high-reliability slice, specifically:

[0092]

[0093] Where is the energy consumed when user i transmits using the wireless channel, specifically:

[0094]

[0095] Where Ω i is the task volume of user i;

[0096] Where is the energy consumption of the task of user i calculated by the edge server, specifically:

[0097]

[0098] Where is the calculation constant of the b i th edge server, ω represents the number of machine cycles required to calculate 1 bit, represents the b i th edge server's CPU frequency;

[0099] Where represents the energy consumption of the computing offloading of user i in the large-scale Internet of Things slice, and is expressed as:

[0100]

[0101] Among them is the energy consumed during the process of forwarding the task to the cloud server, specifically:

[0102]

[0103] Among them, μ is the energy transmission constant;

[0104] Among them is the energy consumed by the cloud server for task calculation, specifically:

[0105]

[0106] Among them, ε c and λ c represent the computing constants and CPU cycles of the cloud server.

[0107] Specifically, in this embodiment, according to the above fitness evaluation function, the roulette wheel selection method is used to perform selection operations on the individuals in the population, specifically:

[0108] Step 501: Calculate the cumulative probability of each individual:

[0109]

[0110] Among them, M is the total number of individuals in the population;

[0111] Step 502: Generate random numbers r1 and r2 in the range of 0 to 1;

[0112] Step 503: If P x-1 ≤ r1 ≤ P x and P y-1 ≤ r2 ≤ P y , then individuals x and y are selected;

[0113] Step 504: Determine whether individuals x and y are the same individual. If so, repeat steps 502 - 504; if not, output individuals x and y.

[0114] Specifically, in this embodiment, after the individual selection is completed, a crossover operation needs to be performed on the selected individuals, and the specific operation is as follows:

[0115] Step 601: Randomly generate an integer k ∈ {1, 2,..., N - 1};

[0116] Step 602: Crossover individuals x and y at the k-th gene value to generate new individuals and

[0117] Step 603, determine the new individuals and Check whether each edge server in exceeds the computing limit. If so, delete the individual and repeat steps 601 - 603; if not, output the new individual.

[0118] Specifically, in this embodiment, mutation operations need to be performed on the individuals after crossover. The specific operations are as follows:

[0119] Step 701, set the mutation probability α ∈ [0, 1];

[0120] Step 702, traverse each bit in the new individual, generate a random number β ∈ [0, 1]. If β ≤ α, mutate this bit; otherwise, keep it unchanged;

[0121] Step 703, check whether the channel selections of the mutated individuals are repeated, whether the time delays of all users meet the requirements, and whether the computing capabilities of all edge servers exceed the upper limit. If any one of the judgments is yes, repeat steps 701 - 703; if all are no, output the new mutated individual and

[0122] Step 704, calculate and compare the original individuals x and y, the new individuals after crossover and as well as the new individuals after mutation and of their fitness values, and select the two individuals with the highest fitness values and put them back into the population.

[0123] Continuously repeat the above selection, crossover, and mutation operations. Until after reaching the iteration number t, the system selects the best individual from the final population according to the individual fitness value as the offloading strategy output.

[0124] In summary, the present invention combines network slicing with edge computing technology. In the scenario of multi - access network slicing, using a genetic algorithm, it offloads the tasks of users in different network slices to edge servers or cloud servers for computing, thereby minimizing the total energy consumption of the system and having considerable application prospects.

[0125] Details not described in the present invention are all well - known technologies in the art.

[0126] The above has described in detail the preferred specific embodiments of the present invention. It should be understood that those of ordinary skill in the art can make many modifications and variations based on the concept of the present invention without creative work. Therefore, all technical solutions that can be obtained by those skilled in the art in the technical field of the present invention through logical analysis, reasoning, or limited experiments based on the concept of the present invention on the basis of the existing technology should be within the protection scope determined by the claims.

Claims

1. A method for edge computing offloading based on genetic algorithm in a multi-access network slicing scenario, characterized in that It includes the following steps: Step S1: Initialize the system model. The system includes multiple channels, multiple users, multiple edge servers, and one cloud server; obtain the task volumes and latency constraints of all users, obtain the computing capabilities of all edge servers and the cloud server, and obtain the channel gains of all channels. Step S2: Create access network slice instances and allocate users to different slices, and set offloading strategies for different slices. Step S3: Initialize the population according to the user offloading strategies within different slices, and optimize and eliminate the individuals in the initialized population. Step S4: Set the iteration step size and start the iteration. Step S5: Perform a selection operation on the individuals by the roulette wheel selection method. Step S6: Perform a single-point crossover operation on the selected individuals. Step S7: Perform a mutation operation on the individuals after crossover. Step S8: Determine whether the iteration step size is reached. If not, repeat the above selection, crossover, and mutation operations; if so, output the individual with the highest fitness value as the offloading strategy. The access network slice instances include a massive IoT slice and a low-latency and high-reliability slice. The users in the massive IoT slice are not sensitive to latency, and the users in the low-latency and high-reliability slice have high requirements for latency. The users in the low-latency and high-reliability slice offload tasks to appropriate edge servers through wireless channels, and the tasks in this slice will only be calculated within the edge servers and will not be forwarded to the cloud server. The users in the massive IoT slice first offload tasks to the edge servers through wireless channels. When the computing capacity of the edge servers reaches the upper limit, the tasks in this slice are forwarded to the cloud server for calculation. Suppose there are U users, F edge servers and one cloud server in the system, and N OFDM channels in total; the sets of users and edge servers are respectively represented as: and The set of all channels is represented by The set of users in the low-latency and high-reliability slice is The set of users in the massive IoT slice is Before creating the population, symbolic encoding needs to be performed on the individuals of the population. Specifically, it is represented as: for each gene value on the individual, it represents the offloading strategy of a single user (a i , b i , c i ); where represents the channel selection variable of the i-th user; represents the edge server selection variable of the i-th user; c i ∈{0, 1} represents the task cloud forwarding variable. If it is 0, it means the task is not forwarded to the cloud server; if it is 1, it means the task is forwarded to the cloud server for calculation; the symbolic encoding of the population individuals is {(a1, b1, c1), (a2, b2, c2),..., (a U , b U , c U )}.

2. The edge computing offloading method based on genetic algorithm in a multi-access network slicing scenario according to claim 1, wherein The method for optimizing and eliminating the initialized population is specifically as follows: Step S301: Check whether the generated individual meets the requirements; first, determine whether all user channel selections a in the individual i are repeated to prevent channel multiplexing. If so, eliminate the individual; if not, proceed to the next judgment. Step S302: Determine whether the transmission delays of all users in the individual meet the requirements, that is, whether the transmission rate is greater than the minimum transmission rate R min If not, eliminate the individual; if so, proceed to the next judgment; Step S303: Determine whether all edge servers in the individual exceed the computing upper limit. If so, eliminate this individual; if not, proceed to the next optimization. Step S304: According to the offloading strategy specified in Step S2, set the cloud forwarding variable c of all users in the low-latency and high-reliability slice to 0, and prohibit tasks from being offloaded to the cloud server.

3. The edge computing offloading method based on genetic algorithm in a multi-access network slicing scenario according to claim 2, wherein The specific calculation of the transmission rate of the user in Step S302 is as follows: The transmission rate expression of the i-th user is: P in formula (1) i represents the transmission power of user i, and N0 is the noise power; represents that user i uses a i channel and b i channel gain when communicating with the edge server; B is the bandwidth of the channel.

4. The edge computing offloading method based on genetic algorithm in a multi-access network slicing scenario according to claim 2, wherein In Step S303, to determine whether all edge servers in the individual exceed the computing upper limit, a fitness evaluation function needs to be constructed, which is specifically expressed as: Among them It represents the energy consumption of the computing offloading of user i in the low-latency and high-reliability slice, specifically: wherein is the energy consumed when user i uses the wireless channel for transmission, specifically: where Ω i is the task volume of user i; Among them is the energy consumption of user i's task calculated on the edge server, specifically: Among them is the computing constant of the b-th i edge server, ω represents the machine cycles required to compute 1 bit, represents the CPU frequency of the b-th i edge server; Among them It represents the energy consumption of user i's computing offloading within the large-scale Internet of Things slice, which is expressed as: Among them is the energy consumed during the process of forwarding the task to the cloud server, specifically: where μ is the energy transmission constant; wherein is the energy consumed by the task during cloud server computing, specifically: where ε c and λ c represent the computing constants and CPU cycles of the cloud server.

5. The edge computing offloading method based on genetic algorithm in a multi-access network slicing scenario according to claim 4, wherein Perform a selection operation on the individuals in the population by the roulette wheel selection method according to the fitness evaluation function, specifically as follows: Step S501: Calculate the cumulative probability of each individual: where M is the total number of individuals in the population, and F j represents the fitness value of the j-th individual; Step S502: Generate random numbers r1 and r2 in the range of 0 to 1; Step S503. If P x-1 ≤ r1 ≤ P x and P y-1 ≤ r2 ≤ P y , then individuals x and y are selected; Step S504: Determine whether individual x and individual y are the same individual. If so, repeat Steps S502 - S504; if not, output individual x and individual y.

6. The edge computing offloading method based on genetic algorithm in a multi-access network slicing scenario according to claim 1, wherein In Step S6, the crossover operation on the selected individuals is specifically as follows: Step S601: Randomly generate an integer k ∈ {1, 2,..., N - 1}; Step S602: Cross individual x and individual y at the k-th gene value to generate a new individual and Step S603: Determine whether each edge server in the new individual and exceeds the computing limit. If so, delete the individual and repeat steps S601 - S603; if not, output the new individual.

7. A method for edge computing offloading based on genetic algorithm in a multi-access network slicing scenario according to claim 1, characterized in that, In Step S7, the mutation operation on the individuals after crossover is specifically as follows: Step S701: Set the mutation probability α ∈ [0, 1]; Step S702: Traverse each bit in the new individual, generate a random number β ∈ [0, 1], if β ≤ α, then mutate this bit; otherwise, keep it unchanged. Step S703: Determine whether the channel selection of the mutated individual is repeated, whether the time delays of all users meet the requirements, and whether the computing capabilities of all edge servers exceed the upper limit. If any of the judgments is yes, repeat steps S701 - S703; if all are no, output the newly mutated individual and Step S704: Calculate and compare the original individuals x and y, and the fitness values of the new individuals and after crossover, as well as the fitness values of the new individuals and after mutation. Select the two individuals with the highest fitness values and put them back into the population.

Images