Task scheduling method and device based on non-dominated sorting genetic algorithm

Abstract

The invention provides a task scheduling method and device based on a non-dominated sorting genetic algorithm. The method comprises the following steps: setting a task scheduling moment; obtaining task information corresponding to an edge computing task generated by each industrial field device in the industrial edge computing architecture and computing power information of each edge computing server; setting a multi-objective optimization model; setting an initial population; using a non-dominated sorting genetic algorithm and the multi-objective optimization model to carry out iteration of a preset number of times on the initial population to obtain a final population; according to task scheduling requirements and the constraint conditions, screening out a chromosome from the final population as an optimal scheduling scheme; and when the task scheduling moment arrives, scheduling the edge computing task according to the optimal scheduling strategy. According to the method, task response time delay and system energy consumption can be reduced, and the load balancing degree of the server is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of industrial scheduling, and in particular to a task scheduling method and device based on a non-dominated sorting genetic algorithm. Background Art

[0002] The calculations involved in current industrial scenarios are becoming more and more complex, so edge computing servers are set up on the industrial field equipment side. However, since there are multiple industrial field devices on the industrial field equipment side and multiple edge computing servers, how to schedule the tasks generated by each industrial field device to each edge computing server is a problem that needs to be solved at present. Summary of the invention

[0003] In view of at least one of the above technical problems, an embodiment of the present invention provides a task scheduling method and device based on a non-dominated sorting genetic algorithm.

[0004] According to the first aspect, the task scheduling method based on the non-dominated sorting genetic algorithm provided by the embodiment of the present invention includes:

[0005] Set the task scheduling time;

[0006] Obtain the task information corresponding to the edge computing tasks generated by each industrial field device in the industrial edge computing architecture and the computing capacity information of each edge computing server;

[0007] Setting a multi-objective optimization model; wherein the multi-objective optimization model includes an optimization objective function and constraint conditions, wherein the optimization objective function includes a first objective function corresponding to minimizing the average energy consumption of the task, a second objective function corresponding to minimizing the weighted average task response delay, and a third objective function corresponding to minimizing the load balancing degree of the server cluster;

[0008] Setting an initial population, wherein the initial population includes a plurality of chromosomes, each chromosome represents a task scheduling scheme, and the task scheduling scheme includes a matching relationship between each industrial field device and each edge computing server;

[0009] Using a non-dominated sorting genetic algorithm and the multi-objective optimization model, the initial population is iterated for a preset number of times to obtain a final population;

[0010] According to the task scheduling requirements and the constraints, a chromosome is selected from the final population as an optimal scheduling solution;

[0011] When the task scheduling time arrives, the edge computing task is scheduled according to the optimal scheduling strategy.

[0012] According to a second aspect, a task scheduling device based on a non-dominated sorting genetic algorithm provided by an embodiment of the present invention includes:

[0013] Time setting module, used to set task scheduling time;

[0014] An information acquisition module is used to obtain task information corresponding to edge computing tasks generated by each industrial field device in the industrial edge computing architecture and computing capacity information of each edge computing server;

[0015] A model setting module, used to set a multi-objective optimization model; wherein the multi-objective optimization model includes an optimization objective function and constraints, wherein the optimization objective function includes a first objective function corresponding to minimizing the average energy consumption of the task, a second objective function corresponding to minimizing the weighted average task response delay, and a third objective function corresponding to minimizing the load balancing degree of the server cluster;

[0016] A population setting module, used to set an initial population, wherein the initial population includes a plurality of chromosomes, each chromosome represents a task scheduling scheme, and the task scheduling scheme includes a matching relationship between each industrial field device and each edge computing server;

[0017] A genetic iteration module, used to obtain a final population after performing a preset number of iterations on the initial population using a non-dominated sorting genetic algorithm and the multi-objective optimization model;

[0018] A scheme selection module, used for selecting a chromosome from the final population as an optimal scheduling scheme according to the task scheduling requirements and the constraint conditions;

[0019] The solution execution module is used to schedule the edge computing task according to the optimal scheduling strategy when the task scheduling time arrives.

[0020] According to a third aspect, an embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed in a computer, enables the computer to execute the method provided in the first aspect.

[0021] According to a fourth aspect, an embodiment of the present invention provides a computing device, including a memory and a processor, wherein the memory stores executable code, and when the processor executes the executable code, the method provided in the first aspect is implemented.

[0022] The task scheduling method, device, medium, and equipment based on the non-dominated sorting genetic algorithm provided in the embodiment of the present invention set the task scheduling time, obtain the task information corresponding to each edge computing task and the computing power information of each edge computing server, then establish a multi-objective optimization model, set the initial population, and then use the non-dominated sorting genetic algorithm and the multi-objective optimization model to iterate the initial population for a preset number of times to obtain the final population, so as to select a chromosome from the final population as the optimal scheduling scheme according to the task scheduling requirements and the constraints; when the task scheduling time arrives, the scheduling of the edge computing task is performed according to the optimal scheduling strategy. Since the present invention comprehensively considers the computing resource information that the edge computing server can provide and the computing resource information required by each edge computing device, it is very suitable for the industrial edge computing architecture in the industrial field. Moreover, the present invention comprehensively considers the optimization indicators of the three dimensions of task response delay, system energy consumption, and server load balancing, so the optimal scheduling strategy obtained can reduce the task response delay and system energy consumption, and improve the server load balancing. Since the final population has made great improvements in terms of average task energy consumption, weighted average task response delay, and load balancing of the server cluster compared to the initial population, based on the actual scheduling needs and constraints, a scheduling scheme that meets personalized needs and actual conditions is selected from the final population as the optimal scheduling scheme, thereby achieving personalized and flexible selection of scheduling schemes. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 It is a flowchart of a task scheduling method based on a non-dominated sorting genetic algorithm in one embodiment of the present invention;

[0024] Figure 2 It is a schematic diagram of a flow chart of iterating a population for a preset number of times in one embodiment of the present invention;

[0025] Figure 3 A schematic diagram of a crossover operation process in an embodiment of the present invention;

[0026] Figure 4 It is a structural block diagram of a task scheduling device based on a non-dominated sorting genetic algorithm in one embodiment of the present invention. DETAILED DESCRIPTION

[0027] In a first aspect, an embodiment of the present invention provides a task scheduling method based on a non-dominated sorting genetic algorithm, see Figure 1 The method comprises the following steps S110 to S170:

[0028] S110, setting task scheduling time;

[0029] The task scheduling time can be represented as w. For example, it is set to 8:00:00.

[0030] S120, obtaining task information corresponding to edge computing tasks generated by each industrial field device in the industrial edge computing architecture and computing capacity information of each edge computing server;

[0031] Among them, the task information corresponding to each edge computing task may include the data volume, computing volume, expected completion time of the task, power uploaded from industrial field equipment to the edge computing server, computing load factor, memory capacity requirements, hard disk capacity requirements, required computing resource types and the demand for each computing resource.

[0032] For example, for the i-th edge computing task, its task information includes the data volume D i , the amount of calculation is B i , the expected completion time is d i The power of uploading the edge computing task from the industrial field equipment to the edge computing server is P i , calculate the load factor as ε i , Required memory capacity M fi , Required hard disk capacity S fi , the required computing resource information. The required computing resource information O i It can be expressed as: i =[o i1 ,o i2 ,…,o iu ] T , u is the number of types of computing resources, o iu It represents the demand of the i-th edge computing task on the u-th type of computing resources. When the i-th edge computing task does not need a type of computing resources, the demand of this type of computing resources is 0, and T is the inverted sign.

[0033] Among them, the computing power information of each edge computing server may include: the processing speed of the edge computing server, the communication bandwidth, the additional power when processing tasks, the memory capacity, the hard disk capacity, the types of computing resources, and the amount of each computing resource provided per unit time.

[0034] For example, for the jth edge computing server, the computing capacity information includes the processing speed C j , the communication bandwidth is W j , the additional power when processing the task is p j 、The memory capacity is M Fj , with a hard disk capacity of S Fj , computing resource information. Wherein, the computing resource information can be expressed as: Lj =[l j1 ,l j2 ,…,l ju ] T ; u is the number of types of computing resources, l ju It indicates the amount of computing resources that the j-th edge computing server can provide per unit time on the u-th type of computing resources. When the j-th edge computing server does not have a type of computing resource, the amount of computing resources provided by this type of computing resource is 0.

[0035] It is understandable that each edge computing server can process multiple edge computing tasks at the same time, and the task processing speed of the same edge computing server is the same; the communication between industrial field devices and edge computing servers is based on orthogonal frequency division multiple access technology, and the edge computing server can communicate with multiple industrial field devices at the same time.

[0036] For example, the number of collected edge computing tasks is 100, the number of edge computing servers is 5, the number of types of computing resources such as GPU and NPU required by the task is 5, the number of types of computing resources such as GPU and NPU possessed by the edge computing server is 5, the data transmission power of industrial field equipment is 0.5-2.5W, the additional power of edge computing server for calculation is 50-250W, the internal noise power of the channel is -174dBm, and the distance between all industrial field equipment and edge computing servers is 100m. The value range of the task information of edge computing tasks and the computing capacity information of edge computing servers is shown in Table 1 below.

[0037] Table 1 Value ranges of edge computing tasks and edge computing server related information

[0038]

[0039]

[0040] S130, setting a multi-objective optimization model; wherein the multi-objective optimization model includes an optimization objective function and constraints, the optimization objective function includes a first objective function corresponding to minimizing the average energy consumption of the task, a second objective function corresponding to minimizing the weighted average task response delay, and a third objective function corresponding to minimizing the load balancing degree of the server cluster;

[0041] Wherein, the first objective function may be:

[0042] Wherein, the second objective function is:

[0043] Wherein, the third objective function is: f3 = min S Ω ;

[0044] Wherein, f1 is the average energy consumption of the minimized task; f2 is the minimized weighted average task response delay; f3 is the load balancing degree of the minimized server cluster; N is the number of edge computing tasks; M is the number of edge computing servers; i is the number of edge computing tasks; j is the number of edge computing servers; E ij is the uplink transmission energy consumption of the i-th edge computing task offloaded from the industrial field device to the j-th edge computing server; x ij is a decision variable. If the i-th edge computing task is assigned to the j-th edge computing server, then x ij is 1, otherwise x ij is 0; d i is the expected completion time of the i-th edge computing task; w is the task scheduling time; T ij S is the uplink transmission time of offloading the i-th edge computing task to the j-th edge computing server; Ω is the load balancing degree of the edge computing server cluster.

[0045] It can be understood that the overall optimization objective function is f=f1+f2+f3.

[0046] Furthermore, the uplink transmission energy consumption E of the i-th edge computing task offloaded from the industrial field device to the j-th edge computing server is ij The calculation formula can be:

[0047]

[0048] The uplink transmission time T of the i-th edge computing task offloaded from the industrial field device to the j-th edge computing server ij The calculation formula can be:

[0049]

[0050] Among them, D i is the data volume of the i-th edge computing task; B i is the computational effort of the ith edge computing task; C j is the processing speed of the jth edge computing server; R ij P is the data transmission rate of the i-th edge computing task uploaded from the industrial field device to the j-th edge computing server; i is the power of the i-th edge computing task; p j is the execution power of the jth edge computing server;

[0051] in, σ 2is the Gaussian noise power inside the channel, Los = d -α , d is the physical distance between the industrial field device of the i-th edge computing task and the j-th edge computing server, and α is 2.

[0052] Among them, Los is a distance-based function constructed based on the wireless interference model of the cellular wireless environment.

[0053] Furthermore, the load balancing degree S of the edge computing server cluster Ω The calculation formula may include:

[0054]

[0055]

[0056]

[0057]

[0058] Among them, Ω j is the total computing load of the jth edge computing server; is the average computing load of all edge computing servers; Φ i is the computing load of the i-th edge computing task; ε i is the computing load factor of the i-th edge computing task; B i is the computational amount of the i-th edge computing task.

[0059] It should be noted that S Ω ≥0, so S Ω The larger the value of , the more unbalanced the task distribution in the edge computing server cluster. Ideally, when the computing load value of each edge computing server is Ω, the load balancing degree of the edge computing server cluster reaches the optimal level.

[0060] In one embodiment, the constraints may include:

[0061]

[0062] Among them, K ij is the network bandwidth allocated between the i-th edge computing task and the j-th edge computing server; W j is the communication bandwidth of the jth edge computing server; M fi is the memory capacity requirement of the i-th edge computing task; M Fj is the memory capacity of the jth edge computing server; S fi is the hard disk capacity requirement of the i-th edge computing task; S Fjis the hard disk capacity of the jth edge computing server; i is the computing resource demand vector corresponding to the i-th edge computing task, which includes various computing resources required by the i-th edge computing task and the demand for each computing resource; L j is the computing resource capability vector corresponding to the j-th edge computing server, and the computing resource capability vector includes various computing resources possessed by the j-th edge computing server and the provision amount of each computing resource per unit time.

[0063] It can be seen that constraint (1) indicates that any edge computing task can only be assigned to one edge computing server. Constraint (2) indicates that the sum of the bandwidths assigned to the same edge computing server is equal to the network bandwidth of the edge computing server. Constraint (3) indicates that the sum of the memory requirements assigned to the same edge computing server cannot exceed the total memory of the edge computing server. Constraint (4) indicates that the sum of the hard disk requirements assigned to the same edge computing server cannot exceed the total hard disk capacity of the edge computing server. Constraint (5) indicates that the sum of the resource requirements assigned to the same edge computing server cannot exceed the total computing module resources of the edge computing server. Constraints (6) and (7) represent the value range of the decision variables.

[0064] S140, setting an initial population, wherein the initial population includes a plurality of chromosomes, each chromosome represents a task scheduling scheme, and the task scheduling scheme includes a matching relationship between each industrial field device and each edge computing server;

[0065] It is understandable that there are M genes on the chromosome, each gene represents an edge computing server, and each gene is named in alphabetical order, for example, A, B, etc. Each gene corresponds to the number of one or more edge computing tasks. For example, for chromosome 1: A1(1,3)B1(2)C1(4)D1(5), gene A1 corresponds to edge computing tasks 1 and 3.

[0066] In one embodiment, setting the initial population in S140 may include: setting the population size to N p ; Generate a greedy solution chromosome and (N p -1) random chromosomes, and get N p The initial population is formed by chromosomes.

[0067] It can be seen that the initial population consists of a greedy solution chromosome and N p-1 randomly generated chromosome. In the greedy chromosome solution, edge computing tasks assigned to the same edge computing server can be transmitted simultaneously, but the edge computing tasks transmitted simultaneously need to share the network bandwidth of the edge computing server.

[0068] The process of generating the greedy chromosome solution may include S11 to S14:

[0069] S11, determining the priority of each edge computing task according to the task scheduling time, the expected completion time of each edge computing task, and the computing amount of each edge computing task, and sorting each edge computing task in order from high to low priority to obtain a task sequence;

[0070] It can be seen that the higher the priority, the earlier the edge computing task is in the task sequence, and the more priority is given to the edge computing server.

[0071] S12, calculate the edge computing server with the smallest sum of load balancing degree, task response delay and task energy consumption, and use the edge computing server as the current edge computing server;

[0072] That is, the sum of the load balancing degree, task response delay, and task energy consumption of each edge computing server is calculated, and the edge computing server with the smallest sum is selected as the current edge computing server.

[0073] S13, taking out the front edge computing task in the task sequence from the task sequence, and assigning the taken out edge computing task to the current edge computing server;

[0074] That is, the frontmost edge computing task in the task sequence is assigned to the current edge computing server. Since the frontmost edge computing task in the task sequence is taken out of the task sequence, after S13 is executed, the edge computing task is no longer in the task sequence.

[0075] S14, determining whether the task sequence is empty; if it is empty, encoding the greedy solution chromosome according to the allocation relationship between each edge computing task and each edge computing server; if it is not empty, returning to S12.

[0076] It can be seen that when the task sequence is empty, it means that all edge computing tasks have been assigned, and the greedy solution chromosome is obtained by encoding according to the previous assignment relationship. If the task sequence is not empty, it returns to S12 to continue the task assignment work.

[0077] The generation process of each random chromosome may include S21 to S23:

[0078] S21, determining a set of available edge computing servers corresponding to each edge computing task according to the memory capacity, hard disk capacity, and computing resource information limitations of each edge computing server; and initializing i to 1;

[0079] S22, randomly assigning the i-th edge computing task to an edge computing server from the corresponding set of available edge computing servers; and adding 1 to i;

[0080] S23. Determine whether i is greater than the number M of edge computing tasks; if so, encode a random chromosome according to the allocation relationship between each edge computing task and each edge computing server; if it is not empty, return to S22.

[0081] It can be seen that according to the limitation of resource information, the set of edge computing servers to which each edge computing task can be assigned, that is, the set of available edge computing servers, can be determined. For the first edge computing task, an edge computing server is randomly selected from the set of available edge computing servers of the first edge computing task as the server for executing the first edge computing task. Then, for the second edge computing task, an edge computing server is randomly selected from the set of available edge computing servers of the second edge computing task as the server for executing the second edge computing task. And so on, until all edge computing tasks are assigned. Encode according to the previous assignment relationship to obtain a random chromosome.

[0082] S150, using a non-dominated sorting genetic algorithm and the multi-objective optimization model, iterating the initial population for a preset number of times to obtain a final population;

[0083] In one embodiment, the step of using the non-dominated sorting genetic algorithm and the multi-objective optimization model in S150 to iterate the initial population for a preset number of times to obtain the final population may include S31 to S36:

[0084] S31, determining the quality level of each chromosome in the initial population, and calculating the crowding degree of each chromosome according to the multi-objective optimization model; using the initial population as the current parent population, wherein the parent population has a first number of chromosomes; and setting the initial value of the number of iterations to 1;

[0085] In one embodiment, determining the quality level of each chromosome in a population in S31 may include the following steps S41 to S44:

[0086] S41, set the initial value of u to 1;

[0087] S42, determining all non-dominated solutions in each chromosome in the current population, and defining the quality level of all non-dominated solutions as the u-th level;

[0088] S43, taking each chromosome of the uth level out from the current population;

[0089] S44, determine whether the chromosome in the current population is empty; if it is empty, the superiority level is determined; if it is not empty, add 1 to u and return to S42.

[0090] Among them, the goodness-badness level is the Pareto level. The Pareto level is defined based on the dominance relationship of solutions in a multi-objective optimization problem. If a solution cannot be dominated by any other solution, it is called a non-dominated solution, and its Pareto level is defined as 1. By gradually removing non-dominated solutions, the Pareto levels of the remaining solutions increase in sequence.

[0091] It can be seen that in the first iteration, the quality level of the selected non-dominated solution is level 1. After removing the non-dominated solutions of level 1 from each chromosome, the quality level of the non-dominated solution selected again is level 2. After removing the non-dominated solutions of level 2 from the remaining chromosomes, the quality level of the non-dominated solution selected next is level 3, and so on, until the quality levels of all chromosomes are determined.

[0092] It can be seen that the above process is actually a fast non-dominated sorting method. Since the fast non-dominated sorting method is used in the genetic algorithm, it is called a non-dominated sorting genetic algorithm.

[0093] Among them, for chromosome individuals with the same Pareto level, the calculation of crowding can further and more carefully compare the advantages and disadvantages of chromosomes. Crowding reflects the closeness between the current solution and the surrounding solutions. The purpose of crowding calculation is to make the obtained solution more uniform in the target space.

[0094] In one embodiment, calculating the crowding degree of each chromosome according to the multi-objective optimization model may include: calculating the crowding degree of each chromosome using the following calculation formula:

[0095]

[0096] Among them, x d is the crowding degree of the nth chromosome in a good or bad grade; f1 is the first objective function; f2 is the second objective function; f3 is the third objective function; is the maximum value of the first objective function; is the minimum value of the first objective function; is the maximum value of the second objective function; is the minimum value of the second objective function; is the maximum value of the third objective function; is the minimum value of the third objective function; f1(n+1) is the first objective function value corresponding to the n+1th chromosome in this merit level; f1(n-1) is the first objective function value corresponding to the n-1th chromosome in this merit level; f2(n+1) is the second objective function value corresponding to the n+1th chromosome in this merit level; f2(n-1) is the second objective function value corresponding to the n-1th chromosome in this merit level; f3(n+1) is the third objective function value corresponding to the n+1th chromosome in this merit level; f3(n-1) is the third objective function value corresponding to the n-1th chromosome in this merit level.

[0097] S32, selecting a second number of chromosomes from the parent population according to the merit level and crowding degree of each chromosome;

[0098] After calculating the merit level and the degree of crowding, a second number of chromosomes are selected from the parent population according to these two indicators. Here, the second number of chromosomes are selected from the parent population with the first number, and the selection operation can adopt a tournament selection strategy, which is a common selection operation method in genetic algorithms. The tournament selection strategy is to extract a certain number of individuals from the population each time, and select the best one to enter the offspring population. It is worth noting that a binary tournament is to take out several individuals from the population at one time for comparison, and put the best individual into the population set to be retained for the next generation. In an embodiment of the present invention, a binary tournament selection strategy can be selected.

[0099] The specific steps are as follows:

[0100] Step 1: Determine the number of individuals to be selected each time, T. If it is a binary tournament, T is 2.

[0101] Step 2: Randomly select T individuals from the parent population. First, compare the Pareto ranks of these T individuals, i.e., the superiority and inferiority ranks, and give priority to individuals with lower Pareto ranks; then compare the crowding of the T individuals, and select the uncrowded individuals to enter the next generation.

[0102] Step 3: Determine whether the number of selected individuals reaches the second number; if so, end the selection strategy; otherwise, return to step 2.

[0103] S33, processing the selected second number of chromosomes by crossover and / or mutation to obtain a progeny population containing the first number of chromosomes;

[0104] It is understandable that, according to the selected population encoding method, the length of the chromosome is the number of edge computing servers, and each gene of the solution is a set of edge computing tasks that can be processed by the edge computing server. Therefore, the crossover operation is based on the edge computing task set. Specifically, a single-point crossover method can be used for crossover. Two parent chromosomes are selected, and the crossover position of the chromosome is randomly selected. The probability of the crossover operation is p c .like Figure 3 As shown, the specific steps are as follows:

[0105] Step 1: Randomly select two parent chromosomes and randomly determine the location of the chromosome single-point crossover.

[0106] Step 2: Insert the crossover fragment selected from the first parent into the position of the crossover point of the second chromosome.

[0107] Step 3: Delete the elements in the original genes of the second parent chromosome that are repeated with the crossover genes;

[0108] Step 4: Use the gene generation method to complete the remaining elements in the chromosome.

[0109] It is understandable that the mutation operation of the genetic algorithm, as a supplement to the crossover operation, can improve the local search capability of the algorithm. Common mutation operation methods include basic bit mutation operation, uniform mutation operation and other methods. By comparing and analyzing the advantages and disadvantages of different mutation methods and applicable scenarios, the embodiment of the present invention selects the basic bit mutation operation method for setting, and the probability of mutation operation is p m The specific steps are as follows:

[0110] Step 1: Randomly select a gene on the chromosome to mutate, and determine the edge computing tasks that can be processed by the edge computing server at the location of the gene;

[0111] Step 2: Randomly generate new genes with the same number of elements at the gene position, and replace the original gene with the new gene;

[0112] Step 3: Delete repetitive elements in other gene positions and mutant genes;

[0113] Step 4: Use the gene generation method to complete the remaining elements.

[0114] S34, merging the offspring population and the parent population to obtain a merged population;

[0115] S35, determining the quality level of each chromosome in the merged population, and calculating the crowding degree of each chromosome according to the multi-objective optimization model; selecting a first number of chromosomes from the merged population according to the quality level and crowding degree of each chromosome, and using the population formed by the first number of chromosomes selected from the merged population as the current parent population;

[0116] It is understandable that the way of calculating the merit level and the crowding degree in S35 refers to the way of calculating the merit level and the crowding degree in S31. The way of selecting the first number of chromosomes from the merged population can refer to the way of selecting the second number of chromosomes from the parent population in S32.

[0117] It can be seen that a population formed by the first number of chromosomes is selected from the merged population as the current parent population, thereby realizing the update of the parent population.

[0118] S36, determine whether the number of iterations is greater than the preset number; if so, take the parent population generated by the last iteration as the final population; otherwise, add 1 to the number of iterations and return to S32.

[0119] For example, if the preset number of times is 200, if the number of iterations reaches 200, the parent population generated by the last iteration will be output as the final population. If it does not reach 200, it will continue to iterate until the number of iterations reaches 200.

[0120] For example, see Figure 2 , the initialized parent population P has 200 chromosomes. After the first merit and crowding calculation, 100 chromosomes are selected. After crossover and / or mutation operations, these 100 chromosomes get 200 chromosomes to form the offspring population Q. The merged population R after the offspring population Q and the parent population P merge has 400 chromosomes. After the second merit and crowding calculation, 200 chromosomes are selected. These 200 chromosomes are the updated parent population. This is the first iteration process. In the first iteration process, two merit and crowding calculation and selection operations are performed. Each iteration from the second iteration to the last iteration process only undergoes one merit and crowding calculation and selection operation.

[0121] S160, selecting a chromosome from the final population as an optimal scheduling solution according to the task scheduling requirements and the constraint conditions;

[0122] The scheduling strategy refers to the matching relationship between edge computing tasks and edge computing servers, that is, how each edge computing task is allocated to each edge computing server.

[0123] It is understandable that the final population obtained in S150 has made great improvements in terms of average task energy consumption, weighted average task response delay, and load balancing of the server cluster compared to the initial population. Therefore, according to the actual scheduling requirements and constraints, a scheduling scheme that meets personalized requirements and actual conditions is selected from the final population as the optimal scheduling scheme.

[0124] S170. When the task scheduling time arrives, schedule the edge computing task according to the optimal scheduling strategy.

[0125] The embodiments of the present invention have the following beneficial effects:

[0126] 1. The present invention comprehensively considers the computing resource information that the edge computing server can provide and the computing resource information required by each edge computing device, and is suitable for the industrial edge computing architecture in the industrial field.

[0127] 2. The present invention comprehensively considers task response delay, system energy consumption and server load balance as optimization indicators, and can simultaneously optimize the optimization indicators of three dimensions.

[0128] It can be seen that the embodiments of the present invention can solve the multi-objective problem in the process of deploying the edge computing task architecture, reduce the task response delay and the decision time of the algorithm, and optimize the load balancing degree of the edge computing server cluster.

[0129] In a second aspect, an embodiment of the present invention provides a task scheduling device based on a non-dominated sorting genetic algorithm, see Figure 4 , the device 100 comprises:

[0130] The time setting module 110 is used to set the task scheduling time;

[0131] An information acquisition module 120 is used to obtain task information corresponding to edge computing tasks generated by each industrial field device in the industrial edge computing architecture and computing capacity information of each edge computing server;

[0132] A model setting module 130 is used to set a multi-objective optimization model; wherein the multi-objective optimization model includes an optimization objective function and constraints, wherein the optimization objective function includes a first objective function corresponding to minimizing the average energy consumption of tasks, a second objective function corresponding to minimizing the weighted average task response delay, and a third objective function corresponding to minimizing the load balancing degree of the server cluster;

[0133] A population setting module 140 is used to set an initial population, wherein the initial population includes a plurality of chromosomes, each chromosome represents a task scheduling scheme, and the task scheduling scheme includes a matching relationship between each industrial field device and each edge computing server;

[0134] A genetic iteration module 150 is used to obtain a final population by iterating the initial population for a preset number of times using a non-dominated sorting genetic algorithm and the multi-objective optimization model;

[0135] A scheme selection module 160, configured to select a chromosome from the final population as an optimal scheduling scheme according to the task scheduling requirements and the constraint conditions;

[0136] The solution execution module 170 is used to schedule the edge computing task according to the optimal scheduling strategy when the task scheduling time arrives.

[0137] In one embodiment, the task information corresponding to each edge computing task includes the data volume, computing volume, expected completion time, power uploaded from the industrial field equipment to the edge computing server, computing load factor, memory capacity requirements, hard disk capacity requirements, required types of computing resources and the demand for each computing resource of the edge computing task; and / or, the computing capacity information of each edge computing server includes: the processing speed of the edge computing server, communication bandwidth, additional power when processing tasks, memory capacity, hard disk capacity, types of computing resources and the amount of each computing resource provided per unit time.

[0138] In one embodiment, the first objective function is:

[0139] The second objective function is:

[0140] The third objective function is: f3 = min S Ω ;

[0141] Wherein, f1 is the average energy consumption of the minimized task; f2 is the minimized weighted average task response delay; f3 is the load balancing degree of the minimized server cluster; N is the number of edge computing tasks; M is the number of edge computing servers; i is the number of edge computing tasks; j is the number of edge computing servers; E ij is the uplink transmission energy consumption of the i-th edge computing task offloaded from the industrial field device to the j-th edge computing server; x ij is a decision variable. If the i-th edge computing task is assigned to the j-th edge computing server, then x ij is 1, otherwise x ijis 0; d i is the expected completion time of the i-th edge computing task; w is the task scheduling time; T ij S is the uplink transmission time of offloading the i-th edge computing task to the j-th edge computing server; Ω The load balancing degree of the edge computing server cluster;

[0142] Among them, the uplink transmission energy consumption E of the i-th edge computing task offloaded from the industrial field equipment to the j-th edge computing server is ij The calculation formula is:

[0143]

[0144] Among them, the uplink transmission time T of the i-th edge computing task offloaded from the industrial field device to the j-th edge computing server is ij The calculation formula is:

[0145]

[0146] Among them, D i is the data volume of the i-th edge computing task; B i is the computational effort of the ith edge computing task; C j is the processing speed of the jth edge computing server; R ij P is the data transmission rate of the i-th edge computing task uploaded from the industrial field device to the j-th edge computing server; i is the power of the i-th edge computing task; p j is the execution power of the jth edge computing server;

[0147] in, σ 2 is the Gaussian noise power inside the channel, Los = d -α , d is the physical distance between the industrial field device of the i-th edge computing task and the j-th edge computing server, and α is 2;

[0148] The load balancing degree S of the edge computing server cluster is Ω The calculation formula includes:

[0149]

[0150]

[0151]

[0152]

[0153] Among them, Ωj is the total computing load of the jth edge computing server; is the average computing load of all edge computing servers; Φ i is the computing load of the i-th edge computing task; ε i is the computing load factor of the i-th edge computing task; B i is the computational amount of the i-th edge computing task.

[0154] In one embodiment, the constraints include:

[0155]

[0156] Among them, K ij is the network bandwidth allocated between the i-th edge computing task and the j-th edge computing server; W j is the communication bandwidth of the jth edge computing server; M fi is the memory capacity requirement of the i-th edge computing task; M Fj is the memory capacity of the jth edge computing server; S fi is the hard disk capacity requirement of the i-th edge computing task; S Fj is the hard disk capacity of the jth edge computing server; i is the computing resource demand vector corresponding to the i-th edge computing task, which includes various computing resources required by the i-th edge computing task and the demand for each computing resource; L j is the computing resource capability vector corresponding to the j-th edge computing server, and the computing resource capability vector includes various computing resources possessed by the j-th edge computing server and the provision amount of each computing resource per unit time.

[0157] In one embodiment, the population setting module is specifically used to: set the population size to N p ; Generate a greedy solution chromosome and (N p -1) random chromosomes, and get N p The initial population formed by chromosomes;

[0158] The process of generating the greedy chromosome solution in the population setting module includes:

[0159] S11, determining the priority of each edge computing task according to the task scheduling time, the expected completion time of each edge computing task, and the computing amount of each edge computing task, and sorting each edge computing task in order from high to low priority to obtain a task sequence;

[0160] S12, calculate the edge computing server with the smallest sum of load balancing degree, task response delay and task energy consumption, and use the edge computing server as the current edge computing server;

[0161] S13, taking out the front edge computing task in the task sequence from the task sequence, and assigning the taken out edge computing task to the current edge computing server;

[0162] S14, determining whether the task sequence is empty; if it is empty, obtaining the greedy solution chromosome according to the allocation relationship between each edge computing task and each edge computing server; if it is not empty, returning to S12;

[0163] The process of generating each random chromosome in the population setting module includes:

[0164] S21, determining a set of available edge computing servers corresponding to each edge computing task according to the memory capacity, hard disk capacity, and computing resource information limitations of each edge computing server; and initializing i to 1;

[0165] S22, randomly assigning the i-th edge computing task to an edge computing server from the corresponding set of available edge computing servers; and adding 1 to i;

[0166] S23. Determine whether i is greater than the number M of edge computing tasks; if so, encode a random chromosome according to the allocation relationship between each edge computing task and each edge computing server; if it is not empty, return to S22.

[0167] In one embodiment, the genetic iteration module is specifically used to perform the following steps:

[0168] S31, determining the quality level of each chromosome in the initial population, and calculating the crowding degree of each chromosome according to the multi-objective optimization model; using the initial population as the current parent population, wherein the parent population has a first number of chromosomes; and setting the initial value of the number of iterations to 1;

[0169] S32, selecting a second number of chromosomes from the parent population according to the merit level and crowding degree of each chromosome;

[0170] S33, processing the selected second number of chromosomes by crossover and / or mutation to obtain a progeny population containing the first number of chromosomes;

[0171] S34, merging the offspring population and the parent population to obtain a merged population;

[0172] S35, determining the quality level of each chromosome in the merged population, and calculating the crowding degree of each chromosome according to the multi-objective optimization model; selecting a first number of chromosomes from the merged population according to the quality level and crowding degree of each chromosome, and using the population formed by the first number of chromosomes selected from the merged population as the current parent population;

[0173] S36, determine whether the number of iterations is greater than the preset number; if so, take the parent population generated by the last iteration as the final population; otherwise, add 1 to the number of iterations and return to S32.

[0174] In one embodiment, the process of determining the quality level of each chromosome in a population by the genetic iteration module includes:

[0175] S41, set the initial value of u to 1;

[0176] S42, determining all non-dominated solutions in each chromosome in the current population, and defining the quality level of all non-dominated solutions as the u-th level;

[0177] S43, taking each chromosome of the uth level out from the current population;

[0178] S44, judging whether the chromosome in the current population is empty; if it is empty, the superiority and inferiority level is determined; if it is not empty, u is increased by 1, and the process returns to S42;

[0179] And / or, calculating the crowding degree of each chromosome according to the multi-objective optimization model includes: calculating the crowding degree of each chromosome using the following calculation formula:

[0180]

[0181] Among them, x d is the crowding degree of the nth chromosome in a good or bad grade; f1 is the first objective function; f2 is the second objective function; f3 is the third objective function; is the maximum value of the first objective function; is the minimum value of the first objective function; is the maximum value of the second objective function; is the minimum value of the second objective function; is the maximum value of the third objective function; is the minimum value of the third objective function; f1(n+1) is the first objective function value corresponding to the n+1th chromosome in this merit level; f1(n-1) is the first objective function value corresponding to the n-1th chromosome in this merit level; f2(n+1) is the second objective function value corresponding to the n+1th chromosome in this merit level; f2(n-1) is the second objective function value corresponding to the n-1th chromosome in this merit level; f3(n+1) is the third objective function value corresponding to the n+1th chromosome in this merit level; f3(n-1) is the third objective function value corresponding to the n-1th chromosome in this merit level.

[0182] It is understandable that the explanation, specific implementation, beneficial effects, examples, etc. of the relevant contents in the device provided in the embodiment of the present invention can be found in the corresponding parts of the method provided in the first aspect, and will not be repeated here.

[0183] In a third aspect, an embodiment of the present invention provides a computer-readable medium, on which computer instructions are stored. When the computer instructions are executed by a processor, the processor executes the method provided in the first aspect.

[0184] Specifically, a system or device equipped with a storage medium can be provided, on which software program code that implements the functions of any of the above-mentioned embodiments is stored, and a computer (or CPU or MPU) of the system or device can be enabled to read and execute the program code stored in the storage medium.

[0185] In this case, the program code itself read from the storage medium can realize the function of any one of the above-mentioned embodiments, and thus the program code and the storage medium storing the program code constitute a part of the present invention.

[0186] The storage medium embodiments for providing the program code include a floppy disk, a hard disk, a magneto-optical disk, an optical disk (such as CD-ROM, CD-R, CD-RW, DVD-ROM, DVD-RAM, DVD-RW, DVD+RW), a magnetic tape, a non-volatile memory card, and a ROM. Alternatively, the program code can be downloaded from a server computer by a communication network.

[0187] In addition, it should be clear that the functions of any of the above embodiments can be implemented not only by executing the program code read by the computer, but also by enabling an operating system operating on the computer to complete part or all of the actual operations based on instructions from the program code.

[0188] In addition, it can be understood that the program code read from the storage medium is written to a memory provided in an expansion board inserted into the computer or to a memory provided in an expansion module connected to the computer, and then based on the instructions of the program code, a CPU installed on the expansion board or expansion module is enabled to perform part or all of the actual operations, thereby realizing the functions of any of the above-mentioned embodiments.

[0189] It is understandable that the explanation, specific implementation methods, beneficial effects, examples, etc. of the relevant contents in the computer-readable medium provided in the embodiment of the present invention can be found in the corresponding parts of the method provided in the first aspect, and will not be repeated here.

[0190] In a fourth aspect, an embodiment of the present specification provides a computing device, including a memory and a processor, wherein the memory stores executable code, and when the processor executes the executable code, it implements the method in any embodiment of the specification.

[0191] It is understandable that the explanation, specific implementation, beneficial effects, examples, etc. of the relevant contents in the computing device provided in the embodiment of the present invention can be found in the corresponding parts of the method provided in the first aspect, and will not be repeated here.

[0192] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the device embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0193] Those skilled in the art should be aware that in one or more of the above examples, the functions described in the present invention can be implemented by hardware, software, widgets, or any combination thereof. When implemented by software, these functions can be stored in a computer-readable medium or transmitted as one or more instructions or codes on a computer-readable medium.

[0194] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made on the basis of the technical solution of the present invention should be included in the scope of protection of the present invention.

Claims

1. A task scheduling method based on non-dominated sorting genetic algorithm, characterized in that: include: Set the task scheduling time; Obtain the task information corresponding to the edge computing tasks generated by each industrial field device in the industrial edge computing architecture and the computing capacity information of each edge computing server; Setting a multi-objective optimization model; wherein the multi-objective optimization model includes an optimization objective function and constraint conditions, wherein the optimization objective function includes a first objective function corresponding to minimizing the average energy consumption of the task, a second objective function corresponding to minimizing the weighted average task response delay, and a third objective function corresponding to minimizing the load balancing degree of the server cluster; Setting an initial population, wherein the initial population includes a plurality of chromosomes, each chromosome represents a task scheduling scheme, and the task scheduling scheme includes a matching relationship between each industrial field device and each edge computing server; Using a non-dominated sorting genetic algorithm and the multi-objective optimization model, the initial population is iterated for a preset number of times to obtain a final population; According to the task scheduling requirements and the constraints, a chromosome is selected from the final population as an optimal scheduling solution; When the task scheduling time arrives, the edge computing task is scheduled according to the optimal scheduling strategy.

2. The method according to claim 1, characterized in that The task information corresponding to each edge computing task includes the data volume, computing volume, expected completion time, power uploaded from industrial field equipment to the edge computing server, computing load factor, memory capacity requirements, hard disk capacity requirements, required computing resource types, and the required amount of each computing resource. And / or, the computing capacity information of each edge computing server includes: the processing speed of the edge computing server, the communication bandwidth, the additional power when processing tasks, the memory capacity, the hard disk capacity, the types of computing resources, and the amount of each computing resource provided per unit time.

3. The method according to claim 1, characterized in that The first objective function is: The second objective function is: The third objective function is: f3 = min S Ω ; Wherein, f1 is the average energy consumption of the minimized task; f2 is the minimized weighted average task response delay; f3 is the load balancing degree of the minimized server cluster; N is the number of edge computing tasks; M is the number of edge computing servers; i is the number of edge computing tasks; j is the number of edge computing servers; E ij is the uplink transmission energy consumption of the i-th edge computing task offloaded from the industrial field device to the j-th edge computing server; x ij is a decision variable. If the i-th edge computing task is assigned to the j-th edge computing server, then x ij is 1, otherwise x ij is 0; d i is the expected completion time of the i-th edge computing task; w is the task scheduling time; T ij S is the uplink transmission time of offloading the i-th edge computing task to the j-th edge computing server; Ω The load balancing degree of the edge computing server cluster; Among them, the uplink transmission energy consumption E of the i-th edge computing task offloaded from the industrial field equipment to the j-th edge computing server is ij The calculation formula is: Among them, the uplink transmission time T of the i-th edge computing task offloaded from the industrial field device to the j-th edge computing server is ij The calculation formula is: Among them, D i is the data volume of the i-th edge computing task; B i is the computational effort of the ith edge computing task; C j is the processing speed of the jth edge computing server; R ij P is the data transmission rate of the i-th edge computing task uploaded from the industrial field device to the j-th edge computing server; i is the power of the i-th edge computing task; p j is the execution power of the jth edge computing server; in, σ 2 is the Gaussian noise power inside the channel, Los = d -α , d is the physical distance between the industrial field device of the i-th edge computing task and the j-th edge computing server, and α is 2; The load balancing degree S of the edge computing server cluster is Ω The calculation formula includes: Among them, Ω j is the total computing load of the jth edge computing server; is the average computing load of all edge computing servers; Φ i is the computing load of the i-th edge computing task; ε i is the computing load factor of the i-th edge computing task; B i is the computational amount of the i-th edge computing task.

4. The method according to claim 3, characterized in that The constraints include: Among them, K ij is the network bandwidth allocated between the i-th edge computing task and the j-th edge computing server; W j is the communication bandwidth of the jth edge computing server; M fi is the memory capacity requirement of the i-th edge computing task; M Fj is the memory capacity of the jth edge computing server; S fi is the hard disk capacity requirement of the i-th edge computing task; S Fj is the hard disk capacity of the jth edge computing server; i is the computing resource demand vector corresponding to the i-th edge computing task, which includes various computing resources required by the i-th edge computing task and the demand for each computing resource; L j is the computing resource capability vector corresponding to the j-th edge computing server, and the computing resource capability vector includes various computing resources possessed by the j-th edge computing server and the provision amount of each computing resource per unit time.

5. The method according to claim 1, characterized in that The setting of the initial population includes: Set the population size to N p ; Generate a greedy solution chromosome and (N p -1) random chromosomes, and get N p The initial population formed by chromosomes; The process of generating the greedy chromosome solution includes: S11, determining the priority of each edge computing task according to the task scheduling time, the expected completion time of each edge computing task, and the computing amount of each edge computing task, and sorting each edge computing task in order from high to low priority to obtain a task sequence; S12, calculate the edge computing server with the smallest sum of load balancing degree, task response delay and task energy consumption, and use the edge computing server as the current edge computing server; S13, taking out the front edge computing task in the task sequence from the task sequence, and assigning the taken out edge computing task to the current edge computing server; S14, determining whether the task sequence is empty; if it is empty, obtaining the greedy solution chromosome according to the allocation relationship between each edge computing task and each edge computing server; if it is not empty, returning to S12; The generation process of each random chromosome includes: S21, determining a set of available edge computing servers corresponding to each edge computing task according to the memory capacity, hard disk capacity, and computing resource information limitations of each edge computing server; and initializing i to 1; S22, randomly assigning the i-th edge computing task to an edge computing server from the corresponding set of available edge computing servers; and adding 1 to i; S23. Determine whether i is greater than the number M of edge computing tasks; if so, encode a random chromosome according to the allocation relationship between each edge computing task and each edge computing server; if it is not empty, return to S22.

6. The method according to claim 5, characterized in that The non-dominated sorting genetic algorithm and the multi-objective optimization model are used to iterate the initial population for a preset number of times to obtain a final population, including: S31, determining the quality level of each chromosome in the initial population, and calculating the crowding degree of each chromosome according to the multi-objective optimization model; using the initial population as the current parent population, wherein the parent population has a first number of chromosomes; and setting the initial value of the number of iterations to 1; S32, selecting a second number of chromosomes from the parent population according to the merit level and crowding degree of each chromosome; S33, processing the selected second number of chromosomes by crossover and / or mutation to obtain a progeny population containing the first number of chromosomes; S34, merging the offspring population and the parent population to obtain a merged population; S35, determining the quality level of each chromosome in the merged population, and calculating the crowding degree of each chromosome according to the multi-objective optimization model; selecting a first number of chromosomes from the merged population according to the quality level and crowding degree of each chromosome, and using the population formed by the first number of chromosomes selected from the merged population as the current parent population; S36, determine whether the number of iterations is greater than the preset number; if so, take the parent population generated by the last iteration as the final population; otherwise, add 1 to the number of iterations and return to S32.

7. The method according to claim 6, characterized in that Determine the quality of each chromosome in a population, including: S41, set the initial value of u to 1; S42, determining all non-dominated solutions in each chromosome in the current population, and defining the quality level of all non-dominated solutions as the u-th level; S43, taking each chromosome of the uth level out from the current population; S44, judging whether the chromosome in the current population is empty; if it is empty, the superiority and inferiority level is determined; if it is not empty, u is increased by 1, and the process returns to S42; And / or, calculating the crowding degree of each chromosome according to the multi-objective optimization model includes: calculating the crowding degree of each chromosome using the following calculation formula: Among them, x d is the crowding degree of the nth chromosome in a good or bad grade; f1 is the first objective function; f2 is the second objective function; f3 is the third objective function; is the maximum value of the first objective function; is the minimum value of the first objective function; is the maximum value of the second objective function; is the minimum value of the second objective function; is the maximum value of the third objective function; is the minimum value of the third objective function; f1(n+1) is the first objective function value corresponding to the n+1th chromosome in this merit level; f1(n-1) is the first objective function value corresponding to the n-1th chromosome in this merit level; f2(n+1) is the second objective function value corresponding to the n+1th chromosome in this merit level; f2(n-1) is the second objective function value corresponding to the n-1th chromosome in this merit level; f3(n+1) is the third objective function value corresponding to the n+1th chromosome in this merit level; f3(n-1) is the third objective function value corresponding to the n-1th chromosome in this merit level.

8. A task scheduling device based on a non-dominated sorting genetic algorithm, characterized in that: include: Time setting module, used to set task scheduling time; An information acquisition module is used to obtain task information corresponding to edge computing tasks generated by each industrial field device in the industrial edge computing architecture and computing capacity information of each edge computing server; A model setting module, used to set a multi-objective optimization model; wherein the multi-objective optimization model includes an optimization objective function and constraints, wherein the optimization objective function includes a first objective function corresponding to minimizing the average energy consumption of the task, a second objective function corresponding to minimizing the weighted average task response delay, and a third objective function corresponding to minimizing the load balancing degree of the server cluster; A population setting module, used to set an initial population, wherein the initial population includes a plurality of chromosomes, each chromosome represents a task scheduling scheme, and the task scheduling scheme includes a matching relationship between each industrial field device and each edge computing server; A genetic iteration module, used to obtain a final population after performing a preset number of iterations on the initial population using a non-dominated sorting genetic algorithm and the multi-objective optimization model; A scheme selection module, used for selecting a chromosome from the final population as an optimal scheduling scheme according to the task scheduling requirements and the constraint conditions; The solution execution module is used to schedule the edge computing task according to the optimal scheduling strategy when the task scheduling time arrives.

9. A computer-readable storage medium, characterized in that: A computer program is stored thereon, and when the computer program is executed in a computer, the computer is caused to execute the method according to any one of claims 1 to 7.

10. A computing device, characterized in that: The method comprises a memory and a processor, wherein the memory stores executable codes, and when the processor executes the executable codes, the method according to any one of claims 1 to 7 is implemented.

Images