Graph clonal selection algorithm optimization-based computing power network scheduling method and device

Abstract

The invention relates to a computing power network scheduling method based on graph clone selection algorithm optimization, and the method comprises the steps: obtaining the node information and task information of a target computing power network, constructing a graph model according to the information, and enabling the time minimization of a maximized task to serve as an optimization target; generating a plurality of initial antibody populations based on the uniformity of the nodes and the randomness of the tasks; determining a fitness function based on the calculation time of each node and the average calculation time; based on Pareto selection, a plurality of antibodies with the highest fitness are selected in proportion through fitness function values for cloning; based on the adaptive mutation probability, mutating the cloned antibody to obtain a plurality of variation populations; and iterating the plurality of variation populations through an elitism strategy until convergence or iteration times reach a threshold value, and obtaining an optimal task allocation diagram corresponding to the optimization target. According to the method, efficient search is carried out in the graph space through the clone selection algorithm, and low cost, low time delay and high reliability in the dynamic computing power network are achieved.

Description

Technical Field

[0001] The present invention belongs to the field of network scheduling technology, and specifically relates to a computing power network scheduling method and device based on graph cloning selection algorithm optimization. Background Art

[0002] In computing network resource scheduling scenarios, the advantages of the clonal selection algorithm make it an effective solution for addressing the challenges of dynamic heterogeneous environments. With the rapid development of cloud computing, edge computing, and distributed computing, computing networks need to coordinate heterogeneous resources across regions and architectures in real time to dynamically respond to the complex demands of large-scale computing tasks. Traditional greedy strategies and gradient optimization methods are prone to local optimality in complex network topologies, making it difficult to achieve global resource balancing. While deep learning-based scheduling schemes offer predictive capabilities, they rely on massive amounts of historical data for training and their convergence lags behind dynamic node changes, making them difficult to address the triple constraints of heterogeneity, load volatility, and latency sensitivity in computing networks.

[0003] In contrast, the clonal selection algorithm takes into account multi-objective optimization capabilities in parallel search by simulating the clonal expansion, high-frequency mutation and fitness screening mechanisms of the biological immune system: on the one hand, its cloning operator expands the local search space by replicating high-affinity antibodies, and cooperates with the mutation strategy to escape the local optimal trap, and can quickly identify high-performance computing node combinations when large-scale task groups are divided; on the other hand, the antibody group adaptive recombination mechanism can dynamically adjust the solution set distribution according to the node load. When the edge node is offline or the cloud computing power fluctuates, the scheduling plan is reconstructed by real-time evaluation of resource availability, significantly reducing the probability of queue blocking due to node failure.

[0004] Faced with the lack of adaptability of traditional algorithms in time-varying networks, this algorithm enhances global exploration capabilities through an antibody diversity maintenance mechanism. It can simultaneously handle multi-dimensional optimization objectives such as bandwidth constraints, energy consumption limitations, and task deadlines without relying on offline data pre-training, enabling it to quickly generate feasible scheduling strategies in cold start scenarios. Summary of the Invention

[0005] In order to solve the problem of unreasonable task allocation in the computing power network, which makes it difficult to achieve efficient load balancing, maximize resource utilization and minimize completion time, the first aspect of the present invention provides a computing power network scheduling method based on graph cloning selection algorithm optimization, including: obtaining node information and task information of the target computing power network; constructing a graph model based on the node information and task information, and determining constraints; based on the graph model and constraints, minimizing the time of maximizing tasks is used as the optimization goal; based on the uniformity of nodes and the randomness of tasks, multiple task allocation graphs are generated; using the multiple task allocation graphs as the initial antibody population for the optimization target; determining the fitness function based on the calculation time of each node and the average calculation time of the population to which the node belongs; based on Pareto selection, arranging the initial antibody population from large to small according to the fitness function value, selecting the first S antibodies for cloning, where S is a positive integer; mutating the cloned antibodies based on the adaptive mutation probability of the task calculation amount to obtain multiple mutant populations; based on the above-mentioned cloning and mutation operations, iterating the multiple mutant populations through the elite retention strategy until the fitness function value converges or the number of iterations reaches a threshold, thereby obtaining the optimal task allocation graph corresponding to the optimization target.

[0006] In some embodiments of the present invention, determining the fitness function based on the calculation time of each node and the average calculation time of the population to which the node belongs includes: determining the fitness load difference based on the calculation time of each node, the average calculation time of the population to which the node belongs, the penalty coefficient and the maximum calculation time of the population to which the node belongs; and determining the fitness function based on the inverse of the load difference.

[0007] In some embodiments of the present invention, the method of Pareto selection, wherein the initial antibody population is arranged from large to small according to the fitness function value, and the first S antibodies are selected for cloning, where S is a positive integer, includes: normalizing the fitness function value of each antibody based on the maximum and minimum values ​​of the fitness; calculating the number of clones of each antibody according to the normalized fitness function and a preset maximum number; and based on the number of clones, arranging the initial antibody population from large to small according to the fitness function value, and selecting the first S antibodies for cloning.

[0008] Furthermore, based on the number of clones, the initial antibody population is arranged from large to small according to the fitness function value, and the top S antibodies are selected for cloning, which includes: selecting S antibodies with the highest fitness from the initial antibody population; for the S antibodies, cloning the antibodies according to the number of clones of each antibody as the number of cloning times.

[0009] In some embodiments of the present invention, the adaptive mutation probability based on the task computing amount mutates the cloned antibodies to obtain multiple mutant populations, including: determining the task requirements, maximum task computing requirements, first mutation probability, time decay coefficient, decay rate and number of iterations of each node; calculating the mutation probability of the task requirements of each node based on the task requirements, maximum task computing requirements, first mutation probability, time decay coefficient, decay rate and number of iterations; judging whether to generate a new allocation computing node based on the size of a random function with a value interval of [0, 1] and the mutation probability: when the value of the random function is less than the mutation probability, a new computing node is randomly selected.

[0010] In the above embodiment, the mutation probability is calculated by the following method: , in, The first mutation probability, Indicates a task The probability of mutation, c i represents a computing node, C max represents the maximum task computing requirement, represent the time decay coefficient, decay rate, and number of iterations respectively.

[0011] In a second aspect of the present invention, a computing power network scheduling device based on graph cloning selection algorithm optimization is provided, comprising: an acquisition module for acquiring node information and task information of a target computing power network; constructing a graph model based on the node information and task information, and determining constraints; based on the graph model and constraints, minimizing the time of maximizing tasks is used as the optimization goal; a generation module for generating multiple task allocation graphs based on the uniformity of nodes and the randomness of tasks; using the multiple task allocation graphs as the initial antibody population for the optimization goal; a cloning module for determining a fitness function based on the calculation time of each node and the average calculation time of the population in which the node is located; based on Pareto selection, the initial antibody population is arranged from large to small according to the fitness function value, and the first S antibodies are selected for cloning, where S is a positive integer; a mutation module for mutating the cloned antibodies based on the adaptive mutation probability of the task calculation amount to obtain multiple mutant populations; an iteration module for iterating the multiple mutant populations through an elite retention strategy based on the above-mentioned cloning operation and mutation operation until the fitness function value converges or the number of iterations reaches a threshold, thereby obtaining the optimal task allocation graph corresponding to the optimization goal.

[0012] The third aspect of the present invention provides an electronic device, comprising: one or more processors; a storage device for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the computing power network scheduling method based on graph cloning selection algorithm optimization provided in the first aspect of the present invention.

[0013] A fourth aspect of the present invention provides a computer-readable medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the computing power network scheduling method based on graph cloning selection algorithm optimization provided in the first aspect of the present invention is implemented.

[0014] The beneficial effects of the present invention are: The present invention provides a computing network scheduling method based on graph clone selection algorithm optimization. To address the problem of unreasonable task allocation in computing networks, which makes it difficult to achieve efficient load balancing, maximize resource utilization, and minimize completion time, the method first models the computing network as a dynamic graph structure, determines the target task guided by the computing power of network nodes and task requirements, and then introduces a dynamic perception mutation mechanism, a load balancing-oriented cloning strategy, and a mixed population management method based on the clone selection algorithm to accelerate its algorithm convergence in the graph space, thereby optimizing and improving the overall efficiency of computing network scheduling tasks. Scheduling optimization is performed for simulated computing network test scenarios. The present invention can reduce the completion time of computing network scheduling tasks, improve the algorithm convergence speed, and support rapid rescheduling of dynamic node changes. It can be widely applied to distributed computing technologies such as cloud computing and edge computing, and related fields such as computing network scheduling.

[0015] The present invention models the computing power network as a dynamic graph structure, and uses an improved clone selection algorithm to perform efficient search in the graph space to achieve the optimal matching of tasks and resources. The improved scheme reduces the execution time of scheduling tasks and improves resource utilization while ensuring the robustness of global search, thus realizing a low-cost, low-latency, and highly reliable resource scheduling adaptive closed loop in the dynamic computing power network. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 A schematic diagram of the basic flow of a computing power network scheduling method optimized based on a graph cloning selection algorithm in some embodiments of the present invention; Figure 2 A schematic diagram of a specific process of a computing power network scheduling method based on graph cloning selection algorithm optimization in some embodiments of the present invention; Figure 3 Schematic diagram of the flow of the clone selection algorithm in some embodiments of the present invention; Figure 4Schematic diagram of simulation effect of a computing power network scheduling method optimized based on a graph cloning selection algorithm in some embodiments of the present invention; Figure 5 A schematic diagram of the structure of a computing power network scheduling device optimized based on a graph cloning selection algorithm in some embodiments of the present invention; Figure 6 Schematic diagram of the structure of an electronic device in some embodiments of the present invention. DETAILED DESCRIPTION

[0017] The principles and features of the present invention are described below with reference to the accompanying drawings. The examples given are only used to explain the present invention and are not used to limit the scope of the present invention.

[0018] refer to Figure 1 and Figure 2 In a first aspect of the present invention, a computing power network scheduling method based on graph cloning selection algorithm optimization is provided, including: S100. Obtaining node information and task information of the target computing power network; constructing a graph model based on the node information and task information, and determining constraints; based on the graph model and constraints, minimizing the time of maximizing tasks is used as the optimization goal; S200. Based on the uniformity of nodes and the randomness of tasks, multiple task allocation graphs are generated; using the multiple task allocation graphs as the initial antibody population of the optimization target; S300. Determining the fitness function based on the calculation time of each node and the average calculation time of the population to which the node belongs; based on Pareto selection, arranging the initial antibody population from large to small according to the fitness function value, and selecting the first S antibodies for cloning, where S is a positive integer; S400. Mutating the cloned antibodies based on the adaptive mutation probability of the task calculation amount to obtain multiple mutant populations; S500. Based on the above-mentioned cloning operation and mutation operation, iterating the multiple mutant populations through the elite retention strategy until the fitness function value converges or the number of iterations reaches a threshold, and obtaining the optimal task allocation graph corresponding to the optimization target.

[0019] In step S100 of some embodiments of the present invention, S101. Obtain node information and task information of the target computing network; Specifically, in the computing network scheduling scenario, the computing network is abstracted as a graph ,definition is the computing node (node ​​set), and the computing node set is represented as , where each computing node Have computing power (Unit: FLOPS); define M as a task (edge ​​set), and the task set is expressed as , where each task The computational requirements are (Unit: FLOPS); node computing power and task computing requirements are respectively and express; S101. Construct a graph model based on the node information and task information, and determine constraints; based on the graph model and constraints, minimize the time of maximizing the task as the optimization goal; For each computing node , the mathematical model of computing node load is constructed as follows: , Constraints for tasks Allows you to assign Contribute to its computing needs, where ; Compute nodes The total load calculation time for processing its tasks is: , determine the optimization goal to minimize ,in Indicates the latest node time for calculation; In step S200 of some embodiments of the present invention, multiple task allocation graphs are generated based on the uniformity of nodes and the randomness of tasks; the multiple task allocation graphs are used as the initial antibody population of the optimization target; Specifically, in the antibody population generation phase, multiple task allocation schemes are randomly generated to form the initial antibody population. Tasks are evenly distributed among nodes. For each task , corresponding to randomly selecting a computing node , forming the initial population: , in Represents discrete uniform distribution, ensuring that tasks are evenly distributed to all computing nodes and avoiding the deviation of the initial population from affecting the optimization process. P represents the antibody, which is also the task graph; In step S300 of some embodiments of the present invention, determining the fitness function based on the computation time of each node and the average computation time of the population to which the node belongs includes: S301. Determine the fitness load difference based on the calculation time of each node, the average calculation time of the node population, the penalty coefficient and the maximum calculation time of the node population; Specifically, in the adaptive evaluation phase, the maximum computation time of each allocation scheme is calculated and the load difference is calculated as a fitness indicator for antibody-antigen affinity evaluation; the fitness function generally reflects the pros and cons of the scheme. for: , Define the fitness function based on load difference : ; in, represents the computation time of each node, Indicates the average computation time of the node's population.

[0020] It can be understood that affinity is a measure to evaluate the quality of antibodies. Depending on the problem and encoding method, the definition of affinity varies. In problems such as pattern recognition and path optimization, binary encoding or integer encoding is usually used. At this time, similarity distance calculation methods such as Hamming distance and Manhattan distance are usually selected as the measure of affinity. For continuous optimization problems, real number encoding is usually used, and the objective function itself or Euclidean distance is generally used for measurement. When dealing with multimodal problems, it is necessary to consider the matching degree between antibodies and antigens, as well as the diversity of antibodies. Affinity is refined into antigen-antibody affinity and antibody-antibody affinity and defined separately.

[0021] S302. Determine a fitness function based on the inverse of the load difference.

[0022] Specifically, the fitness value is inversely proportional to the sum of the weighted maximum completion time and the load difference: , in is the penalty coefficient, The average computation time of the node.

[0023] It's understandable that during clonal proliferation, the number of antibody clone progeny is proportional to the affinity for the antigen; that is, the more superior the individual, the more clones it produces. Therefore, when implementing the algorithm, the number of clone progeny follows the definition of proportional cloning. Proportional cloning increases the search within the current optimal local area, enhancing local search capabilities. If the optimization process aims to define multiple optima within a single antibody population, all antibodies in the population can be selected for cloning. In this case, proportional cloning no longer applies, and the new number of clones is redefined so that each antibody has the same number of clones, i.e., proportional cloning.

[0024] refer to Figure 3 In step S300 of some embodiments of the present invention, the Pareto selection is based on sorting the initial antibody population from large to small according to the fitness function value, and selecting the top S antibodies for cloning, where S is a positive integer including: S303. Based on the maximum and minimum values ​​of the fitness, the fitness function value of each antibody is normalized; Specifically, the specification The value range is between: , in ) represents the maximum value of the current population fitness, Indicates the minimum value, is the probability of antibody clone after normalization; S304. Calculate the number of clones for each antibody based on the normalized fitness function and the preset maximum number; Specifically, the number of clones for each antibody is calculated based on the fitness value: , in Indicates the maximum number of clones, here ; S305. Based on the number of clones, the initial antibody population is sorted from largest to smallest according to the fitness function value, and the top S antibodies are selected for cloning.

[0025] Specifically, the antibodies with the highest fitness were selected based on the number of clones. Generate the corresponding clone pool: , By selecting the fitness Antibodies, including , The high-quality antibodies were cloned multiple times according to the corresponding ratios selected for cloning; I understand. C(X p ) is a function that takes a parameter X_p and returns a value that represents X p Therefore, C(X p ) It should be understood as a function, not an index. The specific implementation of this function may depend on the antibody X p The fitness or other characteristics of a clone determine its number.

[0026] refer to Figure 3 In step S400 of some embodiments of the present invention, the adaptive mutation probability based on the task computational amount is used to mutate the cloned antibodies to obtain multiple mutant populations, including: S401. Determine the task requirements, maximum task computing requirements, first mutation probability, time decay coefficient, decay rate and number of iterations for each node; S402. Calculate the mutation probability of the task requirement of each node based on the task requirement, the maximum task calculation requirement, the first mutation probability, the time decay coefficient, the decay rate, and the number of iterations; Specifically, for Calculation requirements tasks , let the maximum task computing requirement be , the basic mutation probability is ,Task The mutation probability of: , in is the time attenuation coefficient, is the decay rate, is the current iteration number; S403. Determine whether to generate a new allocation computing node based on the random function with a value interval of [0, 1] and the size of the mutation probability: when the value of the random function is less than the mutation probability, randomly select a new computing node.

[0027] Network scheduling task allocation scheme in each clone ,in Indicates a task The assigned computing nodes first calculate the task The mutation probability , the mutation phase generates new allocation computing nodes :

[0028] In this operation, to increase the possibility of optimization, tasks with large computational requirements are more likely to mutate. Conversely, to maintain the existing optimal solution, tasks with small computational requirements have a lower probability of mutation.

[0029] It's important to note that in genetic algorithms, mutation is used in conjunction with the crossover operator. The frequency of mutation is typically low, and its primary purpose is to increase the algorithm's ability to explore beyond local optima. In contrast, in clonal selection algorithms, the mutation operator, as one of the primary operators, must balance both local and global exploration capabilities. Therefore, its design is more complex, and the mutation rate can vary for each antibody. In areas with low affinity, the mutation rate is higher, increasing global exploration capacity; in areas with high affinity, the mutation rate is lower, increasing local exploration capacity.

[0030] In step S500 of some embodiments of the present invention, based on the above-mentioned cloning operation and mutation operation, the multiple mutant populations are iterated through the elite retention strategy until the fitness function value converges or the number of iterations reaches a threshold, thereby obtaining the optimal task allocation graph corresponding to the optimization objective.

[0031] Specifically, in the population iteration phase, the new and old populations are merged and the elite retention strategy is executed to generate a new generation of population; the previous generation population is merged The population produced by clonal mutation gather for: , The antibody fitness values ​​in step 3 are calculated for the antibodies in the combined population. After all the antibody fitness values ​​are calculated, they are arranged from large to small. In order to improve the search efficiency, an elite retention strategy is introduced to retain the best antibodies of the previous generation. : , in To preserve the ratio, select the best one from the merged population Antibodies form new populations , accelerate the convergence of the algorithm; In the convergence output phase, the optimal graph task allocation solution is output after iteration reaches the termination condition. The algorithm terminates when the fitness converges or reaches the set maximum number of iterations. During this process, the algorithm calculates the current optimal graph task allocation solution in each generation: , in is the current optimal graph task allocation solution, For all antibodies in the current population, the maximum completion time and the optimal graph task allocation plan are finally output; refer to Figure 4 , which shows some simulation results of the simulation scenario (20 nodes / 200 tasks). It should be understood that the above description of the preferred embodiment is relatively detailed, and it cannot be regarded as limiting the scope of protection of the patent of the present invention. Under the guidance of the present invention, ordinary technicians in this field can also make substitutions or modifications without departing from the scope of protection of the claims of the present invention, which all fall within the scope of protection of the present invention. The scope of protection requested by the present invention shall be based on the attached claims.

[0032] Example 2 refer to Figure 5In a second aspect, the present invention provides a computing power network scheduling device 1 based on a graph cloning selection algorithm optimization, comprising: an acquisition module 11 for acquiring node information and task information of a target computing power network; constructing a graph model based on the node information and task information and determining constraints; minimizing the time of maximizing tasks based on the graph model and constraints as the optimization goal; a generation module 12 for generating multiple task allocation graphs based on the uniformity of nodes and the randomness of tasks; using the multiple task allocation graphs as initial antibody populations for the optimization goal; a cloning module 13 for determining a fitness function based on the calculation time of each node and the average calculation time of the population to which the node belongs; arranging the initial antibody population from large to small according to the fitness function value based on Pareto selection, and selecting the first S antibodies for cloning, where S is a positive integer; a mutation module 14 for mutating the cloned antibodies based on the adaptive mutation probability of the task calculation amount to obtain multiple mutant populations; an iteration module 15 for iterating the multiple mutant populations through an elite retention strategy based on the above-mentioned cloning and mutation operations until the fitness function value converges or the number of iterations reaches a threshold, thereby obtaining an optimal task allocation graph corresponding to the optimization goal.

[0033] Furthermore, the cloning module 13 includes: a first determination unit, used to determine the fitness load difference based on the calculation time of each node, the average calculation time of the population to which the node belongs, the penalty coefficient and the maximum calculation time of the population to which the node belongs; a second determination unit, used to determine the fitness function based on the inverse of the load difference.

[0034] Example 3 refer to Figure 6 According to the third aspect of the present invention, an electronic device is provided, comprising: one or more processors; a storage device for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the computing power network scheduling method based on graph clone selection algorithm optimization in the first aspect of the present invention.

[0035] The electronic device 500 may include a processing device (e.g., a central processing unit, a graphics processing unit, etc.) 501, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 502 or a program loaded from a storage device 508 into a random access memory (RAM) 503. Various programs and data required for the operation of the electronic device 500 are also stored in the RAM 503. The processing device 501, the ROM 502, and the RAM 503 are connected to each other via a bus 504. An input / output (I / O) interface 505 is also connected to the bus 504.

[0036] Typically, the following devices may be connected to the I / O interface 505: an input device 506 including, for example, a touch screen, a touchpad, a keyboard, a mouse, a camera, a microphone, an accelerometer, a gyroscope, etc.; an output device 507 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc.; a storage device 508 including, for example, a hard disk, etc.; and a communication device 509. The communication device 509 may allow the electronic device 500 to communicate with other devices wirelessly or by wire to exchange data. Figure 6 The electronic device 500 is shown with various devices, but it should be understood that it is not required to implement or possess all of the devices shown. More or fewer devices may be implemented or possessed instead. Figure 6 Each block shown in the figure may represent one device, or may represent multiple devices as needed.

[0037] In particular, according to an embodiment of the present disclosure, the process described above with reference to the flowchart can be implemented as a computer software program. For example, an embodiment of the present disclosure includes a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program includes a program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from the network through the communication device 509, or installed from the storage device 508, or installed from the ROM 502. When the computer program is executed by the processing device 501, the above-mentioned functions defined in the method of the embodiment of the present disclosure are executed. It should be noted that the computer-readable medium described in the embodiment of the present disclosure can be a computer-readable signal medium or a computer-readable storage medium or any combination of the above two. The computer-readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to, an electrical connection having one or more conductors, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In embodiments of the present disclosure, a computer-readable storage medium may be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In embodiments of the present disclosure, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, which carries computer-readable program code. Such a propagated data signal may take a variety of forms, including, but not limited to, an electromagnetic signal, an optical signal, or any suitable combination thereof. A computer-readable signal medium may also be any computer-readable medium other than a computer-readable storage medium that can transmit, propagate, or transfer a program for use by or in conjunction with an instruction execution system, apparatus, or device. Program code embodied on a computer readable medium may be transmitted using any appropriate medium, including but not limited to wire, optical cable, RF (radio frequency), etc., or any suitable combination thereof.

[0038] The computer-readable medium may be included in the electronic device, or may exist independently without being incorporated into the electronic device. The computer-readable medium carries one or more computer programs, which, when executed by the electronic device, cause the electronic device to: Computer program code for performing the operations of embodiments of the present disclosure may be written in one or more programming languages, or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, C++, Python, and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a stand-alone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., through the Internet using an Internet service provider).

[0039] The flowcharts and block diagrams in the accompanying drawings illustrate the possible implementation architecture, functions and operations of the systems, methods and computer program products according to various embodiments of the present disclosure. In this regard, each box in the flowchart or block diagram can represent a module, program segment, or a part of code, and the module, program segment, or a part of code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order than that marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should be noted that each box in the block diagram and / or flowchart, and the combination of the boxes in the block diagram and / or flowchart, can be implemented with a dedicated hardware-based system that performs the specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0040] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A computing power network scheduling method based on graph clone selection algorithm optimization, characterized in that: include: Obtain node information and task information of the target computing network; Constructing a graph model based on the node information and task information, and determining constraint conditions; Based on the graphical model and constraints, the optimization goal is to minimize the time of the maximum task; Based on the uniformity of nodes and the randomness of tasks, multiple task allocation graphs are generated; Using the multiple task allocation graphs as an initial antibody population for optimization; The fitness function is determined based on the computation time of each node and the average computation time of the node's population. Based on Pareto selection, the initial antibody population is sorted from largest to smallest according to the fitness function value, and the top S antibodies are selected for cloning, where S is a positive integer. Based on the adaptive mutation probability of the task computational load, the cloned antibodies are mutated to obtain multiple mutant populations; Based on the above cloning and mutation operations, the multiple mutant populations are iterated through an elite retention strategy until the fitness function value converges or the number of iterations reaches a threshold, thereby obtaining an optimal task allocation graph corresponding to the optimization objective.

2. The computing power network scheduling method based on graph clone selection algorithm optimization according to claim 1 is characterized in that: Determining the fitness function based on the computation time of each node and the average computation time of the population to which the node belongs includes: Determine the fitness load difference based on the computation time of each node, the average computation time of the node's population, the penalty coefficient, and the maximum computation time of the node's population; The fitness function is determined based on the inverse of the load difference.

3. The computing power network scheduling method based on graph cloning selection algorithm optimization according to claim 1 is characterized in that: The Pareto selection method arranges the initial antibody population from large to small according to the fitness function value, and selects the first S antibodies for cloning, including: Based on the maximum and minimum values ​​of fitness, the fitness function value of each antibody is normalized; Calculate the number of clones for each antibody based on the normalized fitness function and the preset maximum number; Based on the number of clones, the initial antibody population is arranged from large to small according to the fitness function value, and the top S antibodies are selected for cloning.

4. The computing power network scheduling method based on graph clone selection algorithm optimization according to claim 4 is characterized in that: The step of arranging the initial antibody population from largest to smallest according to the fitness function value based on the number of clones and selecting the top S antibodies for cloning includes: For three antibodies, the number of clones of each antibody was used as the cloning number, and the antibodies were cloned.

5. The computing power network scheduling method based on graph cloning selection algorithm optimization according to claim 1 is characterized in that: The adaptive mutation probability based on the task computation amount is used to mutate the cloned antibodies to obtain multiple mutant populations including: Determine the task requirements, maximum task computation requirements, first mutation probability, time decay coefficient, decay rate, and number of iterations for each node; Calculate the mutation probability of the task requirement of each node based on the task requirement, the maximum task calculation requirement, the first mutation probability, the time decay coefficient, the decay rate and the number of iterations; Whether to generate a new allocation computing node is determined based on the random function with a value range of [0, 1] and the size of the mutation probability: when the value of the random function is less than the mutation probability, a new computing node is randomly selected.

6. The computing power network scheduling method based on graph cloning selection algorithm optimization according to claim 5 is characterized in that: The mutation probability is calculated by the following method: ， in, The first mutation probability, Indicates a task The probability of mutation, c i represents a computing node, C max represents the maximum task computing requirement, represent the time decay coefficient, decay rate, and number of iterations respectively.

7. A computing power network scheduling device based on graph cloning selection algorithm optimization, characterized in that: include: The acquisition module is used to obtain the node information and task information of the target computing network; Constructing a graph model based on the node information and task information, and determining constraint conditions; Based on the graphical model and constraints, the optimization goal is to minimize the time of the maximum task; A generation module is used to generate multiple task allocation graphs based on the uniformity of nodes and the randomness of tasks; Using the multiple task allocation graphs as an initial antibody population for optimization; The cloning module is used to determine the fitness function based on the computation time of each node and the average computation time of the node's population; based on Pareto selection, the initial antibody population is sorted from large to small according to the fitness function value, and the first S antibodies are selected for cloning. The initial antibody population is sorted from large to small according to the fitness function value, where S is a positive integer; The mutation module is used to mutate the cloned antibodies based on the adaptive mutation probability of the task computational load to obtain multiple mutant populations; The iteration module is used to iterate the multiple mutant populations based on the above-mentioned cloning operation and mutation operation through the elite retention strategy until the fitness function value converges or the number of iterations reaches a threshold, thereby obtaining the optimal task allocation graph corresponding to the optimization target.

8. The computing power network scheduling device based on graph cloning selection algorithm optimization according to claim 7 is characterized in that: The cloning module includes: A first determining unit is configured to determine a fitness load difference based on a computation time of each node, an average computation time of a population to which the node belongs, a penalty coefficient, and a maximum computation time of a population to which the node belongs; The second determining unit is configured to determine a fitness function based on the inverse of the load difference.

9. An electronic device comprising: one or more processors; A storage device for storing one or more programs, which, when executed by the one or more processors, enables the one or more processors to implement the computing power network scheduling method based on graph cloning selection algorithm optimization as described in any one of claims 1 to 6.

10. A computer-readable medium having a computer program stored thereon, wherein: When the computer program is executed by a processor, the computing power network scheduling method based on graph cloning selection algorithm optimization as described in any one of claims 1 to 6 is implemented.

Images