Distributed storage resource scheduling method based on improved OBXGA

Abstract

The invention relates to the technical field of data processing, in particular to a distributed storage resource scheduling method based on improved OBXGA, which comprises the following steps: selectively operating genes with low population fitness by adopting an OBX crossover method, selectively mutating mutation operators according to the gene fitness and the current algorithm iteration times, and carrying out mutation on the mutation operators according to the gene fitness and the current algorithm iteration times. The gene with low fitness is enabled to be evolved more easily; proposing a main and auxiliary population breeding strategy, and breeding by selecting a population in the population according to a fitness ratio, so that the algorithm has an excellent auxiliary population as a supplement in a main population iteration process; the method has the beneficial effects that two populations, namely a main population and an auxiliary population, are iterated, the auxiliary population supplements the main population, and the main population selects an excellent population for breeding by utilizing an iteration frequency threshold ratio, so that the population is enabled to evolve towards a better direction; and the secondary population selects elite individuals as breeding supplementation of the main population by using a gene fitness ratio, so that the diversity and convergence of the population in an iterative evolution process are ensured.

Description

Technical Field

[0001] The invention relates to the technical field of data processing, and in particular to a distributed storage resource scheduling method based on improved OBXGA. Background Art

[0002] With the continuous evolution of the world, the progress of information technology and artificial intelligence has promoted the development of mission operations in the direction of informatization and intelligence. Future mission operations need to digitize and intelligently integrate various types of information to provide more accurate data support for the planning, command, and actions of mission operations. Especially at the end of the mission operation, various storage devices at the edge of the mission operation form a distributed storage system through the network, integrating and storing a large amount of data resources that explode in a short period of time on the edge devices of the mission operation. If the storage device and data are not intelligently selected and adapted, the data is stored according to the default device, which is prone to uneven distribution of storage resources and slow storage speed.

[0003] One of the typical application scenarios of current mission action intelligent integration is to build mission action end data integration. From the perspective of JADC2 data construction, current external organizations integrate and fuse various terminal data across fields by building intelligent data integration, and use intelligent algorithms and mission action clouds for processing to form effective information that is fed back to commanders in a relatively short period of time, accelerating the decision-making process in various fields. The intelligent integration method is also used in the study of terminal data application issues. By collecting terminal data, the network is used to instantly aggregate data related to collaborative tasks. Various types of data are aggregated in the task pool for standardized processing, and the relevant data is sent to the mission action end demand unit. One of the key issues in intelligent integration work is the scientific adaptation of heterogeneous storage devices and storage data for mission actions.

[0004] In the mission action scenario, facing a large amount of mission action data resources that surge in a short period of time, different types of data have different storage speeds, and reliable and adaptable storage devices are needed to quickly save data. The basis of intelligent integrated data application is to effectively store data. The use of algorithms to reasonably schedule mission action data storage resources has become a key issue in the current mission action distributed storage optimization. Therefore, it is necessary to propose a genetic algorithm based on OBX improvement to provide a solution for mission action distributed storage. Summary of the invention

[0005] The purpose of the present invention is to provide a distributed storage resource scheduling method based on improved OBXGA to solve the problems raised in the above background technology.

[0006] To achieve the above object, the present invention provides the following technical solution: a distributed storage resource scheduling method based on improved OBXGA, the method comprising the following steps:

[0007] The OBX crossover method is used to selectively operate on genes with low population fitness. The mutation operator selectively mutates according to the gene fitness and the number of iterations of the current algorithm to ensure that genes with low fitness are easier to evolve.

[0008] A main and secondary population reproduction strategy is proposed. By selecting populations within the population according to the fitness ratio for reproduction, the algorithm can have excellent secondary populations as supplements during the iteration of the main population to improve the diversity of the population. The population as a whole evolves and iterates in a better direction, and the optimal fitness update mechanism is introduced to ensure that the local optimal cycle is escaped.

[0009] The improved OBX algorithm is verified to effectively solve the problem of mission action data storage. It is compared with five different algorithms, and five types of indicators are experimentally evaluated after 100, 200 and 600 iterations respectively.

[0010] Preferably, the specific operations of using the OBX cross method include:

[0011] Initialize random encoding to generate the main population nums1 and the sub-population nums2;

[0012] Performing main population genetic iteration and sub-population genetic iteration;

[0013] After the sub-population genetic iteration, wait for the main population genetic iteration to determine whether the population fitness is greater than the set threshold ratio. If it is not greater than the set threshold ratio, continue the main population genetic iteration. If it is greater than the set threshold ratio, mate the excellent genes of the main population nums1 and the sub-population nums2.

[0014] Generate a new main population nums1′;

[0015] Update the best fitness best-gene;

[0016] After calculating the number of iterations, determine whether the number of iterations meets the set number of iterations. If so, end the output population; if not, continue the iteration process.

[0017] Preferably, the main population genetic iteration includes: calculating the fitness of the main population, OBX crossover operation, mutation operation, and generating population nums1;

[0018] Subpopulation genetic iteration includes: fitness calculation, OBX crossover operation, mutation operation, nums2 fitness sorting, and selection of genes with fitness ratio > ε as mating genes.

[0019] Preferably, the OBX crossover algorithm further comprises:

[0020]

[0021]

[0022] in, Represents the probability function of the current gene, P c max , P c min Represent the maximum and minimum values ​​of the entire staining variation probability, f c max 、f avg Respectively represent the maximum and average values ​​of population fitness, f m represents the fitness of the current mutation individual, g is the current iteration number, G max The maximum number of iterations is used to measure the current number of iterations and correct the coefficient of variation, so that fewer excellent genes mutate and the optimization speed of the genetic algorithm is improved.

[0023] Preferably, the threshold value P is set to dynamically adjust the threshold parameter according to the number of iterations, and the expression is:

[0024]

[0025]

[0026] In the formula, The threshold P of loop iteration is dynamically modified based on the ratio of the current number of iterations to the set maximum number of iterations. g , f fit It is the value after chromosome normalization. The fitness of the secondary population is calculated after OBX crossover mutation. Excellent genes are selected as mating genes according to a certain ratio ε and enter the mating pool. The fitness of the main population and the mating gene are uniformly sorted, and the genome with the lowest fitness is replaced in turn to complete the mating process.

[0027] Optimally, the main and secondary population breeding algorithm is:

[0028] Enter the amount of data ΔD that needs to be stored this time

[0029] Output Output best assigned genome best_gene

[0030] 1).nums1,nums2←randomly generate the initial main and secondary populations by ΔD

[0031] 2).for g←1to G max do

[0032] 3).if nums1 fitness <P g then

[0033] 4).nums1←OBX algorithm / / gene crossover

[0034] 5).nums1←Adaptive mutation algorithm / / Gene mutation

[0035] 6).nums1'←Generate a new main population

[0036] 7).else

[0037] 8). Output the main population nums1' into the mating pool

[0038] 9).nums2←OBX algorithm / / Gene crossover

[0039] 10).nums2←Adaptive mutation algorithm / / Gene mutation

[0040] 11).nums2'←Generate a new main population

[0041] 12). Calculate the fitness of nums2' and generate the mating set Z

[0042] 13).for i←1to pop_size do

[0043] 14).if f(Chrom)>εthen

[0044] 15). Z=Z∪Chrom

[0045] 16).end for

[0046] 17).nums1'=nums1'∪Z

[0047] 18). Calculate the fitness of nums1

[0048] 19). Record the best fitness of this iteration best_gene (g)

[0049] 20).if best_gene (g) >best_gene then

[0050] 21).best_gene=best_gene (g) / / Update the best gene

[0051] 22). Eliminate chromosomes with poor fitness in nums1'

[0052] 23).g=g+1

[0053] 24).end for.

[0054] Preferably, the improved OBX algorithm is verified to effectively solve the task action data storage problem, and when compared with five different algorithms, the comparison algorithms include the traditional GA algorithm, RWSGA, OXGA, SEGA, and PBXGA;

[0055] By comparing the best fitness, standard deviation of fitness convergence value, algorithm iteration time and Gürler HE obtained by different algorithms under the same initial parameter conditions, online evaluation index and offline evaluation index function for evaluating genetic algorithm are proposed; the standard deviation of fitness convergence value is:

[0056]

[0057] The online performance function is:

[0058]

[0059] Its online performance reflects the change of the average fitness value of the group after smoothing, describing the overall performance and evolutionary ability of the group. The smaller its value in the fitness function, the better the overall performance and the faster the evolutionary ability. For the genetic algorithm, its offline performance is:

[0060]

[0061] f(a * ,t)=max{f(a1,t),f(a2,t),...,f(a n ,t)} is the fitness of the best individual in the current group, which reflects the change of the fitness of the best individual in the group after smoothing and describes the evolutionary ability of the individual. The smaller the value, the better the individual's evolutionary ability.

[0062] Preferably, in order to verify the performance comparison of different algorithms under the mission environment conditions, the population size, number of iterations, initial crossover mutation probability, and DNA length of the experimental settings are kept consistent. Due to the different overall performance of the equipment used at the mission site, in order to ensure the feasibility of the control experiment, the comparison experiment is performed using equipment with similar CPU computing power except for storage performance. The specific parameters are as follows:

[0063] Population size POP_SIZE, set value 10;

[0064] DNA size DNA_SIZE, set value 4;

[0065] Initial gene crossover rate CROSS_RATE, set value to 0.6;

[0066] Initial gene mutation rate MUTATE_RATE, set value 0.02;

[0067] The number of iterations GENERATION, set the value to 100, 200, 600;

[0068] The technical parameters of the storage device are based on a dual-chip architecture consisting of a Tengrui D2000 / 8 processor and an X100 bridge expansion unit, with functional expansion through the PCIe bus. The CPU uses a Tengrui D2000 / 8 eight-core processor with a main frequency of 2.0GHz and a TDP power consumption of 20W. The storage capacities are 512GB, 640GB, 832GB and 1TB respectively, and the 4k random read and write speeds are 300MB / s, 350MB / s, 100MB / s and 150MB / s. The total amount of experimental data is set to 1025.4GB, and the value coefficient is λ={λ1,λ2,...,λ DS} and are set to 0.35, 0.25, 0.2, and 0.2 respectively.

[0069] Preferably, the experiment verifies the difference in the best fitness of different algorithms by iterating 100 times and 200 times under the same initial parameter conditions, and the change in the best fitness of each algorithm by 600 times, and verifies the algorithm by comparing the parameter values.

[0070] Compared with the prior art, the present invention has the following beneficial effects:

[0071] The distributed storage resource scheduling method based on improved OBXGA proposed in the present invention iterates two populations, namely the main population and the secondary population. The secondary population is a supplement to the main population. The main population uses the threshold ratio of the number of iterations to select excellent populations for reproduction to ensure that the population evolves in a better direction; the secondary population uses the gene fitness ratio to select elite individuals as the reproduction supplement of the main population to ensure the diversity and convergence of the population in the iterative evolution process. The reproduction strategy design of the main population and the secondary population enables the overall population to achieve better results in obtaining the optimal fitness, fitness variance, population and individual evolutionary capabilities, etc. Compared with other algorithms, the OBXGA algorithm has more advantages and solves the problem of distributed data storage resource allocation for mission actions to a certain extent. BRIEF DESCRIPTION OF THE DRAWINGS

[0072] Figure 1 This is the cross selection diagram of the OBX gene of the present invention;

[0073] Figure 2 It is the flow chart of the dual population genetic algorithm of the present invention;

[0074] Figure 3 The present invention provides the best fitness comparison diagram and fitness difference diagram of OBXGA and other algorithms after 100 iterations;

[0075] Figure 4 It is the best fitness curve diagram of the present invention;

[0076] Figure 5 It is the online performance variation diagram of the six types of algorithms of the present invention;

[0077] Figure 6 This is a graph showing the offline performance changes of the six types of algorithms of the present invention. DETAILED DESCRIPTION

[0078] In order to make the purpose and technical solution of the present invention clearly and completely described, and the advantages more clearly understood, the embodiments of the present invention are further described in detail with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are part of the embodiments of the present invention, rather than all of the embodiments, and are only used to explain the embodiments of the present invention, and are not used to limit the embodiments of the present invention. All other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0079] For example, see Figures 1 to 6 The present invention provides a technical solution: a distributed storage resource scheduling method based on improved OBXGA, characterized in that the method comprises the following steps:

[0080] The OBX crossover method is used to selectively operate on genes with low population fitness. The mutation operator selectively mutates according to the gene fitness and the number of iterations of the current algorithm to ensure that genes with low fitness are easier to evolve.

[0081] A main and secondary population reproduction strategy is proposed. By selecting populations within the population according to the fitness ratio for reproduction, the algorithm can have excellent secondary populations as supplements during the iteration of the main population to improve the diversity of the population. The population as a whole evolves and iterates in a better direction, and the optimal fitness update mechanism is introduced to ensure that the local optimal cycle is escaped.

[0082] The improved OBX algorithm is verified to effectively solve the problem of mission action data storage. It is compared with five different algorithms, and five types of indicators are experimentally evaluated after 100, 200 and 600 iterations respectively.

[0083] Example 2, based on Example 1, proposes a dual population genetic algorithm based on OBX, specifically including: 1. Data storage chromosome initialization

[0084] The main steps of genetic algorithms are selection, crossover, and mutation. Different improvement methods will affect iteration time, convergence speed, local optimal solution, etc. First, it is necessary to distinguish the data type and determine whether the data type is a large file. Then, the data is sliced ​​and allocated to different storage devices. The genetic algorithm is initialized for data allocation, and different gene chromosomes are randomly generated, that is, the data storage allocation plan. For example, for a data set Chromosomes The different genes on the chromosome are determined by the population size pop_size (ps). The two-dimensional array matrix nums after chromosome encoding can be expressed as:

[0085]

[0086] 2. Fitness function

[0087] The fitness function of the genetic algorithm affects the stability and convergence of the genetic algorithm. Generally, the higher the fitness function value of the genetic algorithm, the better the gene fitness. However, in the data storage problem, F time (D) is the fitness function for measuring storage time. The smaller the storage time, the better. Function F data (D) is the data value function. The higher the value, the better. In order to unify the fitness function, F time (D) and F data (D) The value of the function is normalized, then:

[0088]

[0089] Represents the first chromosome randomly generated in the initialization chromosome gene sequence The normalized value.

[0090] 3. Crossover operator

[0091] In genetic algorithms, there are many crossover operators for populations, such as partial mapping crossover (PMX), sequential crossover (OX), position-based crossover (PBX), cyclic crossover (CX), sub-line exchange crossover (SEX) and other selection methods. A high mating probability can easily destroy the existing favorable pattern, increase randomness, and easily miss the optimal individual. The limitation of the mating probability method cannot effectively update the population. OBX is an order-based crossover method (Order-Based Crossover), which is a gene selection method for improving the crossover operator in genetic algorithms. Compared with ordinary gene crossover, the OBX method is mostly used to consider the element order allocation optimization problem, and has more advantages in dealing with multi-type combination allocation optimization problems;

[0092] Among them, the literature referenced in this article:

[0093] Mapping crossover (PMX)——Li Guojun, Zheng Ziwan, Fan Yingsheng, et al. Multi-UAV patrol path optimization based on hybrid genetic algorithm[J]. Journal of Zhejiang University(Science Edition),2024,51(1).

[0094] Order Crossover (OX) —— Linke Y Y, Borisov I S. Universal nonparametric kernel-type estimators for the mean and covariance functions of a stochastic process[J]. Theory of Probability & Its Applications, 2024, 69(1): 35-58.;

[0095] Position-Based Crossover (PBX) —— Linke Y Y, Borisov I S. Universal nonparametric kernel-type estimators for the mean and covariance functions of a stochastic process[J]. Theory of Probability & Its Applications, 2024, 69(1): 35-58.;

[0096] Cycle Crossover (CX) —— Xia D, Wu X, Yan M, et al. An adaptive stochastic ranking-based tournament selection method for differential evolution[J]. The Journal of Supercomputing, 2024, 80(1): 20-49;

[0097] Sub-line Exchange Crossover (SEX) —— Wu D, Wang Y, Zhou E. Data-driven ranking and selection under input uncertainty[J]. Operations Research, 2024, 72(2): 781-795;

[0098] OBX method——Kumar R, Memoria M, Thapliyal M, et al.Analyzing ThePerformance Of Crossover Operators(OX,OBX,PBX,MPX)To Solve CombinatorialProblems[C] / / 2022International Conference on Machine Learning,Big Data,Cloudand Parallel Computing(COM-IT-CON).IEEE,2022,1:817-821.

[0099] The steps of the OBX gene selection algorithm are as follows:

[0100] Algorithm 2 OBX crossover algorithm

[0101]

[0102]

[0103] OBX gene cross selection diagram Figure 1 As shown:

[0104] 4. Mutation Operator

[0105] The mutation operator uses an adaptive method. Adaptive mutation dynamically adjusts the mutation probability according to the current state of the algorithm. As the number of iterations increases, the algorithm continues to converge. In order to reduce the damage to the existing excellent solutions, the mutation probability of genes with better fitness is reduced; in the early stage of the search or when it falls into the local optimum, the mutation rate will increase to introduce more new mutations, which helps to jump out of the current local structure. In the past, mutations generally corrected the mutation probability according to fitness, but this method still has limitations in iteration speed.

[22] , this paper accelerates the speed of gene mutation by adding the number of iterations. The formula before correction is as follows:

[0106]

[0107] in, Represents the probability function of the current gene, P c max , P c min Represent the maximum and minimum values ​​of the entire staining variation probability, f c max 、f avg Respectively represent the maximum and average values ​​of population fitness, f mrepresents the fitness of the current mutation individual, g is the current iteration number, G max The maximum number of iterations is used to measure the current number of iterations and correct the coefficient of variation, so that fewer excellent genes mutate and the optimization speed of the genetic algorithm is improved.

[0108] 5. Algorithm 2 Adaptive Mutation Algorithm

[0109]

[0110] 6. Dual breeding strategy

[0111] The update and iteration strategy of the population is the key method to maintain the optimization iteration of the genetic algorithm. The traditional dual-population update iteration is to initialize and generate two populations, perform genetic mutation operations respectively, and merge the populations after a certain number of iterations to form the final population. However, this method has strong randomness and is not suitable for fast iteration. This paper combines the differential evolution algorithm and the co-evolution strategy to propose a main and secondary population reproduction method. By generating two main and secondary populations, the main population is used as the main population for iterative update, and the secondary population is used as the supplementary population of the main population. The main population first performs OBX crossover mutation to generate a new population nums1' before mating, and calculates the fitness of the population. If it is greater than the set threshold P, it enters the mating pool for mating. The secondary population nums2 performs OBX crossover mutation to generate a new population nums2', calculates the fitness of the population, and sorts it according to the fitness. The proportion of population nums2' with a fitness greater than ε is mated with the main population to generate a new main population nums1', and the best calculated fitness f of this round of iteration is recorded. The number of iterations increases until the number of iterations meets the loop condition and stops.

[0112] Among them, the final population - Kumar R, Memoria M, Gupta A, et al. Critical analysis of genetic algorithm under crossover and mutation rate [C] / / 2021 3rdinternational conference on advances in computing, communication control and networking (ICAC3N). IEEE, 2021:976-980;

[0113] Differential evolution algorithm——Salgotra R, Gandomi A HA novel multi-hybriddifferential evolution algorithm for optimization of frame structures[J].Scientific Reports,2024,14(1):4877;

[0114] Li M, Zhao Y, Cao R, et al. A recursive framework for improving the performance of multi-objective differential evolution algorithms for genes selection[J]. Swarm and Evolutionary Computation, 2024,87:101546;

[0115] Ahmed ZH,Haron H,Al-Tameem A.Appropriate Combination of CrossoverOperator and Mutation Operator in Genetic Algorithms for the TravelingSalesman Problem[J].Computers,Materials&Continua,2024,79(2);

[0116] Co-evolution strategy - Han Meihui, Wang Peng, Li Ruixu, et al. An adaptive constrained multi-objective evolutionary algorithm based on co-evolution[J]. Computer Engineering, 2024, 50(06): 124-137. DOI: 10.19678 / j.issn.1000-3428.0067986;

[0117] Liu Z,Yang Y,Cao J,et al.A Coevolutionary algorithm using Self-organizing map approach for multimodal multi-objective optimization[J].Applied Soft Computing,2024:111954;

[0118] Liu S, Feng J, Yang S, et al.A Coevolutionary Algorithm With Detectionand Supervision Strategies for Constrained Multiobjective Optimization[J]. IEEE Transactions on Evolutionary Computation, 2024;

[0119] The threshold P is a threshold parameter that is dynamically adjusted according to the number of iterations, and its expression is:

[0120]

[0121] In the formula The threshold P of loop iteration is dynamically modified based on the ratio of the current number of iterations to the set maximum number of iterations. g , f fit is the value after chromosome normalization; the fitness of the secondary population is calculated after OBX crossover mutation, and excellent genes are selected as mating genes according to a certain ratio ε to enter the mating pool, and the fitness of the main population and mating genes are uniformly sorted, and the genome with the lowest fitness is replaced in turn to complete the mating process. The main and secondary population reproduction algorithm is as follows:

[0122] Algorithm 3: Primary and secondary population reproduction algorithm

[0123] Enter the amount of data ΔD that needs to be stored this time

[0124] Output Output best assigned genome best_gene

[0125] 1.nums1,nums2←randomly generate the initial main and secondary populations by ΔD

[0126] 2.for g←1to G max do

[0127] 3.if nums1 fitness <P g then

[0128] 4.nums1←OBX algorithm / / Gene crossover

[0129] 5.nums1←Adaptive mutation algorithm / / Gene mutation

[0130] 6.nums1'←Generate a new main population

[0131] 7. else

[0132] 8. Output the main population nums1' into the mating pool

[0133] 9.nums2←OBX algorithm / / Gene crossover

[0134] 10.nums2←Adaptive mutation algorithm / / Gene mutation

[0135] 11.nums2'←Generate a new main population

[0136] 12. Calculate nums2' fitness and generate mating set Z

[0137] 13.for i←1to pop_size do

[0138] 14.if f(Chrom)>εthen

[0139] 15.Z=Z∪Chrom

[0140] 16.end for

[0141] 17.nums1'=nums1'∪Z

[0142] 18. Calculate the fitness of nums1

[0143] 19. Record the best fitness of this generation iteration best_gene (g)

[0144] 20.if best_gene (g) >best_gene then

[0145] 21. best_gene = best_gene (g) / / Update the best gene

[0146] 22. Eliminate chromosomes with poor fitness in nums1'

[0147] 23. g = g + 1

[0148] 24.end for.

[0149] Example 3, based on Example 2, proposes an experimental setup and result analysis, specifically including:

[0150] In order to verify that the proposed improved OBX algorithm can effectively solve the task action data storage problem, it is experimentally compared with other genetic algorithms of the same type. The comparison algorithms include traditional GA algorithm, RWSGA, OXGA, SEGA, PBXGA and the improved OBX algorithm proposed in this paper.

[0151] Among them, RWSGA - Gulayeva N, Ustilov A. Analysis of Selection Methods Usedin Genetic Algorithms[J].

[0152] OXGA——Meniz B, Tiryaki F.Genetic Algorithm Optimization withSelection Operator Decider[J]. Arabian Journal for Science and Engineering, 2024:1-11;

[0153] SEGA——Sharma B,Khurana M,Singh H.Optimized hybrid image encryptionscheme using arnold transform,chaotic map and genetic algorithm for improvedsecurity and robustness[C] / / AIP Conference Proceedings.AIP Publishing,2024,3081(1);

[0154] PBXGA——Bulut M, E.Planning of prescriptive maintenance types for generator with fuzzy logic-based genetic algorithm in a hydroelectric power plant[J].Expert Systems with Applications,2024,240:122480.

[0155] 1. Performance indicators

[0156] The evaluation of genetic algorithms mainly focuses on two aspects: solution efficiency and solution quality. Different indicators are used to measure the performance of the algorithm. By comparing the best fitness, standard deviation of fitness convergence value, and iteration time of the algorithm obtained by different algorithms under the same initial parameter conditions, Gürler HE et al. proposed online evaluation indicators and offline evaluation indicator functions for evaluating genetic algorithms.

[0157] Among them, Gürler HE——Gürler HE, M, Pamucar D. Determining criteria weights with genetic algorithms for multi-criteria decision making methods: The case of logistics performance index rankings of European Union countries[J]. Socio-Economic Planning Sciences, 2024,91:101758..

[0158] The standard deviation of fitness convergence value is:

[0159]

[0160] The online performance function is:

[0161]

[0162] Its online performance reflects the change of the average fitness value of the group after smoothing, describing the overall performance and evolutionary ability of the group. The smaller its value in the fitness function, the better the overall performance and the faster the evolutionary ability. For the genetic algorithm, its offline performance is:

[0163]

[0164] f(a * ,t)=max{f(a1,t),f(a2,t),...,f(a n ,t)} is the fitness of the best individual in the current group, which reflects the change of the fitness of the best individual in the group after smoothing and describes the evolutionary ability of the individual. The smaller the value, the better the individual's evolutionary ability.

[0165] 2. Parameter settings

[0166] In order to verify the performance comparison of different algorithms under the mission environment conditions, the population size, number of iterations, initial crossover mutation probability, and DNA length of the experimental settings are kept consistent. Due to the different overall performance of the equipment used at the mission site, in order to ensure the feasibility of the control experiment, the comparison is set to use equipment with similar computing power such as CPU except for storage performance as much as possible. The specific parameters are as follows:

[0167] Table 1 Common initial parameter setting table

[0168]

[0169] The technical parameters of the storage device are based on a dual-chip architecture consisting of a Tengrui D2000 / 8 processor and an X100 bridge expansion unit, with functional expansion through the PCIe bus. The CPU uses a Tengrui D2000 / 8 eight-core processor with a main frequency of 2.0GHz and a TDP power consumption of 20W. The storage capacities are 512GB, 640GB, 832GB and 1TB respectively, and the 4k random read and write speeds are 300MB / s, 350MB / s, 100MB / s and 150MB / s. The total amount of experimental data is set to 1025.4GB, and the value coefficient is λ={λ1,λ2,...,λ DS} and are set to 0.35, 0.25, 0.2, and 0.2 respectively.

[0170] 3. Experimental results analysis

[0171] The experiment verifies the difference in the best fitness of different algorithms after 100 and 200 iterations under the same initial parameter conditions, as well as the change in the best fitness of various algorithms after 600 iterations. Finally, the algorithm is verified by comparing the values ​​of various parameters. Table 2, Table 3 and Table 4 respectively list the differences in the best convergence f, covariance σ, and algorithm iteration time T of the improved OBX algorithm and other comparison algorithms. sum , Online performance value P on and offline performance value P off Numerical values,Table 2 shows the various indicators obtained after 100 iterations of each algorithm, Table 3 shows 200 iterations, and Table 4 shows 600 iterations. The optimal results of each indicator are indicated in bold.

[0172] Figure 3 The best fitness comparison and fitness difference diagram of OBXGA and other algorithms after 100 iterations are shown in Figure 2. In this paper, the smaller the fitness function value, the better the gene fitness. For the best convergence index f, SEGA performs best. In the early stage of iteration, OBXGA has the highest priority to obtain the local best fitness compared with other algorithms. Figure 3 From the comparison of (a), (b), and (c), we can see that the local optimal fitness of OBXGA is always better than that of the traditional GA, RWSGA, and OXGA algorithms. However, as the iteration number g increases, the difference in fitness gradually decreases, indicating that the gap between the algorithms in the iterative process is constantly decreasing, and OBXGA cannot quickly increase the gap. Figure 3 From the comparison between (d) and (e), we can see that OBXGA is better than SEGA and PBXGA in the initial iteration. However, when iterating to the 40th to 60th generation, due to the small number of iterations, the subpopulation iteration cannot play its good role in supplementing excellent genes. The gene selection methods of SEGA and PBXGA are better than OBXGA in a shorter number of iterations.

[0173] From the analysis of Table 2, we can see that the best fitness of OBXGA, OXGA and PBXGA is not much different and is at the same level. In comparison, the fitness variance of PBXGA is the smallest, which means that its fitness result deviation is the smallest and the search result is the most stable. Compared with it, the fitness variance of OBXGA is larger and its stability is not as good as PBXGA. In terms of the iteration time of the algorithm, due to the complexity of various operators, the running time of the program is increased, and OBXGA takes the longest time. Online performance function P on The value SEGA performs best, which shows that within the range of 100 iterations, the population evolution ability SEGA is optimal. For the offline performance function P off The value PBXGA is the smallest, and the individual evolution ability PBXGA is the best. Due to the small number of iterations, the main and secondary population reproduction methods and the threshold ratio P g The population selection function was not well played, and OBXGA failed to achieve the optimal results.

[0174] Table 2 Algorithm evaluation index table (100 generations)

[0175]

[0176] From the analysis in Table 3, it can be seen that after 200 iterations of the algorithm, each indicator has been significantly optimized. For the best fitness f, OBXGA achieved the optimal value after iteration, but compared with SEGA and PBXGA, the fitness difference was small and no obvious advantage was achieved. For the fitness variance σ, PBXGA still achieved the best effect and the population iteration stability was the best, but the difference between OBXGA and PBXGA was optimized from 2.89 to 1.24, which was a relative improvement of 42.9%, and the population fitness stability was further improved. Due to the complexity of the operator, the algorithm takes T sum The minimum value of the index is the traditional GA algorithm. For the online performance function value and the offline performance function value, OBXGA performs best in the group evolution ability and individual evolution ability. Compared with SEGA and PBXGA, which perform best in 100 iterations, the fitness threshold P g The subpopulation ratio ε gradually takes effect, but the gap between OBXGA and other algorithms is not obvious enough and is still within the same level range. The comparison table of algorithm indicators after 200 iterations is as follows:

[0177] Table 3 Algorithm evaluation index table (200 generations)

[0178]

[0179] After the number of iterations reaches 600, the comprehensive analysis of Table 4 and Figure 4 It can be seen that with the help of the iteration of the secondary population, although the algorithm iteration of OBXGA takes a long time due to the complexity of the algorithm, all indicators except the algorithm time T sumAll other algorithms reach the optimal value. Through experimental comparison, it can be analyzed that in the early iteration process, the OBXGA solution is only better than the traditional GA, RWSGA and OXGA, but the overall population's evolutionary ability and search ability are not as good as SEGA and PBXGA, and it is also not as good as SEGA and PBXGA in solving the optimal fitness problem. However, after a long number of iterations, due to the improved main and secondary population breeding strategy adopted by OBXGA, the improvement of selecting excellent genes of the secondary population can quickly help the main population to iterate in the direction of the optimal solution of the gene. The more iterations, the better the performance of OBXGA's optimal fitness. At the same time, the variance of the population fitness has been greatly improved compared with other algorithms, reflecting that the population's solution set has better stability. Due to the existence of a mechanism to update the best gene, the OBXGA algorithm will not fall into the local optimal solution during the iteration process like ordinary genetic algorithms. The overall performance of OBXGA is better than other algorithms. The evaluation index table and the best fitness curve are shown in the figure. Figure 4 shown.

[0180] Table 4 Algorithm evaluation index table (600 generations)

[0181]

[0182] In order to more clearly reflect the group evolution ability and individual evolution ability of various algorithms, the online performance function value and offline performance function value change process of each algorithm at different iteration times are shown in Figure 5 and Figure 6 to make a display;

[0183] from Figure 5 and Figure 6 The analysis shows that for the online performance function, regardless of the number of iterations, OBXGA has a clear advantage over GA, RWSGA, and OXGA. However, compared with PBXGA and SEGA, OBXGA algorithm has no obvious advantage within 200 generations. on From the perspective of indicators, the online performance function values ​​of the three types of algorithms are P on The overall gap is small, P on The function shows a downward trend, reflecting that the population evolution ability of the OBXGA algorithm is weaker than that of other algorithms. off PBXGA benefited from the use of PBX gene crossing method in the early stage. off The overall function value is small, and the individual evolution ability is the best. The P offAlthough the function did not perform as well as other algorithms in the early stage, its individual evolution ability reached the same level as PBXGA after 200 generations. After 600 generations, except for the algorithm iteration time, all other parameter indicators of OBXGA achieved the optimal value, which was significantly better than the other five algorithms. In the iterative process, it can be analyzed that compared with the other five algorithms, the online performance value and offline performance value of OBXGA decreased much more than other algorithms. Due to the reproduction strategy of the main and secondary populations, other algorithms are prone to fall into the local optimal solution in exploring the optimal fitness problem. Although they can also jump out of the local cycle through adjustment, the solution effect is not as good as the OBXGA method. In addition, under the condition of a large number of iterations, the overall speed of OBXGA to obtain the optimal solution is lower than that of other algorithms. In terms of the growth of the optimal fitness, SEGA increased by 13.14%, PBXGA increased by 12.55%, and OBXGA increased by 19.94%. Under the conditions of task action, it is crucial to be able to solve the most suitable data storage solution in the shortest time, and the difference in algorithm iteration time consumption has little effect on the time magnitude of task action data storage.

[0184] In summary, when the number of iterations is 100-200 generations, the OBXGA algorithm fails to achieve the optimal value in terms of best fitness, fitness variance, algorithm time consumption, online and offline performance, but the gap with the optimal algorithm in various indicators is small, and the downward trend of the evolutionary ability function of the population and individuals is more obvious than that of other algorithms. As the number of iterations increases to 600 generations, except for the algorithm iteration time consumption indicator, the other indicators such as the best fitness, fitness variance, online and offline performance are verified to achieve the optimal value. The experimental results can fully prove that after a long period of algorithm iteration, whether it is fitness evaluation or evaluation of algorithm evolution ability, the overall performance of the OBXGA proposed in this paper is significantly better than the five types of comparison algorithms.

[0185] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A distributed storage resource scheduling method based on improved OBXGA, characterized by: The method comprises the following steps: The OBX crossover method is used to selectively operate on genes with low population fitness. The mutation operator selectively mutates according to the gene fitness and the number of iterations of the current algorithm to ensure that genes with low fitness are easier to evolve. A main and secondary population reproduction strategy is proposed. By selecting populations within the population according to the fitness ratio for reproduction, the algorithm can have excellent secondary populations as supplements during the iteration of the main population to improve the diversity of the population. The population as a whole evolves and iterates in a better direction, and the optimal fitness update mechanism is introduced to ensure that the local optimal cycle is escaped. The improved OBX algorithm is verified to effectively solve the problem of mission action data storage. It is compared with five different algorithms, and five types of indicators are experimentally evaluated after 100, 200 and 600 iterations respectively.

2. The distributed storage resource scheduling method based on improved OBXGA according to claim 1 is characterized in that: The specific operations using the OBX cross method include: Initialize random encoding to generate the main population nums1 and the secondary population nums2; Perform main population genetic iteration and sub-population genetic iteration; After the sub-population genetic iteration, wait for the main population genetic iteration to determine whether the population fitness is greater than the set threshold ratio. If it is not greater than the set threshold ratio, continue the main population genetic iteration. If it is greater than the set threshold ratio, mate the excellent genes of the main population nums1 and the sub-population nums2. Generate a new main population nums1′; Update the best fitness best-gene; After calculating the number of iterations, determine whether the number of iterations meets the set number of iterations. If so, end the output population; if not, continue the iteration process.

3. The distributed storage resource scheduling method based on improved OBXGA according to claim 2 is characterized in that: The main population genetic iteration includes: calculating the fitness of the main population, OBX crossover operation, mutation operation, and generating the population nums1; Subpopulation genetic iteration includes: calculating fitness, OBX crossover operation, mutation operation, nums2 fitness sorting, and selecting genes with fitness ratio > ε as mating genes.

4. The distributed storage resource scheduling method based on improved OBXGA according to claim 2 is characterized in that: The OBX crossover algorithm also includes: Input parent chromosomes Chrom1, Chrom2 Output offspring chromosomes Chrom1′, Chrom2′ 1). Chrom1′, Chrom2′ = {}← create an empty set as the offspring chromosome, initialize N = 1 2).while N≤2do 3). ←Randomly select k genes as the genes for crossover 4). ←Use gene positions as exchange sets 5).if N=1 6). 7).else 8). 9).while i <n do 10).if then 11). / / Gene inheritance of parent 1 12).else 13). / / Gene inheritance of parent 2 14).i=i+1 15).end if 16).end while 17).N=N+1 / / Repeat crossover operation 18).end while.

5. The distributed storage resource scheduling method based on improved OBXGA according to claim 2 is characterized in that: Also includes adaptive mutation algorithms: Input chromosome population code nums Output the population code nums' after mutation 1).for m←1to n do 2).f(nums)←calculate the normalized fitness of the population 3).if f m ≥f avg then 4). 5).else 6). 7). 8). 9).r'←Random value (0,1) 10).m=m+1 11).end for; in, represents the probability function of the current gene, Represent the maximum and minimum values ​​of the entire staining variation probability, respectively. f avg Respectively represent the maximum and average values ​​of population fitness, f m represents the fitness of the current mutation individual, g is the current iteration number, G max The maximum number of iterations is used to measure the current number of iterations and correct the coefficient of variation, so that fewer excellent genes mutate and the optimization speed of the genetic algorithm is improved.

6. The distributed storage resource scheduling method based on improved OBXGA according to claim 2 is characterized in that: The threshold P is set to dynamically adjust the threshold parameter according to the number of iterations. The expression is: In the formula, The threshold P of loop iteration is dynamically modified based on the ratio of the current number of iterations to the set maximum number of iterations. g , f fit is the value after chromosome normalization; The fitness of the secondary population is calculated after OBX crossover mutation, and excellent genes are selected as mating genes according to a certain ratio ε to enter the mating pool. The fitness of the main population and the mating gene are uniformly sorted, and the genome with the lowest fitness is replaced in turn to complete the mating process.

7. The distributed storage resource scheduling method based on improved OBXGA according to claim 6 is characterized in that: Main and secondary population breeding algorithm: Enter the amount of data ΔD that needs to be stored this time Output Output best assigned genome best_gene 1).nums1,nums2←randomly generate the initial main and secondary populations by ΔD 2).for g←1to G max do 3).if nums1 fitness <P g then 4).nums1←OBX algorithm / / gene crossover 5).nums1←Adaptive mutation algorithm / / Gene mutation 6).nums1'←Generate a new main population 7).else 8). Output the main population nums1' into the mating pool 9).nums2←OBX algorithm / / Gene crossover 10).nums2←Adaptive mutation algorithm / / Gene mutation 11).nums2'←Generate a new main population 12). Calculate the fitness of nums2' and generate the mating set Z 13).for i←1to pop_size do 14).if f(Chrom)>εthen 15). Z=Z∪Chrom 16).end for 17).nums1'=nums1'∪Z 18). Calculate the fitness of nums1 19). Record the best fitness of this iteration best_gene (g) 20).if best_gene (g) >best_gene then 21).best_gene=best_gene (g) / / Update the best gene 22). Eliminate chromosomes with poor fitness in nums1' 23).g=g+1 24).end for.

8. The distributed storage resource scheduling method based on improved OBXGA according to claim 1 is characterized in that: Verify that the improved OBX algorithm effectively solves the problem of data storage for mission actions, and compares it with five different algorithms, including the traditional GA algorithm, RWSGA, OXGA, SEGA, and PBXGA; By comparing the best fitness, standard deviation of fitness convergence value, algorithm iteration time and Gürler HE obtained by different algorithms under the same initial parameter conditions, online evaluation index and offline evaluation index function for evaluating genetic algorithm are proposed; the standard deviation of fitness convergence value is: The online performance function is: Its online performance reflects the change of the average fitness value of the group after smoothing, describing the overall performance and evolutionary ability of the group. The smaller its value in the fitness function, the better the overall performance and the faster the evolutionary ability. For the genetic algorithm, its offline performance is: f(a * ,t)=max{f(a1,t),f(a2,t),...,f(a n ,t)} is the fitness of the best individual in the current group, which reflects the change of the fitness of the best individual in the group after smoothing and describes the evolutionary ability of the individual. The smaller the value, the better the individual's evolutionary ability.

9. The distributed storage resource scheduling method based on improved OBXGA according to claim 8, characterized in that: In order to verify the performance comparison of different algorithms under the mission environment conditions, the population size, number of iterations, initial crossover mutation probability, and DNA length of the experimental settings are kept consistent. Due to the different overall performance of the equipment used at the mission site, in order to ensure the feasibility of the control experiment, the comparison experiment is set up using equipment with similar CPU computing power except for storage performance. The specific parameters are as follows: Population size POP_SIZE, set value 10; DNA size DNA_SIZE, set value 4; Initial gene crossover rate CROSS_RATE, set value to 0.6; Initial gene mutation rate MUTATE_RATE, set value 0.02; The number of iterations GENERATION, set the value to 100, 200, 600; The technical parameters of the storage device are based on a dual-chip architecture consisting of a Tengrui D2000 / 8 processor and an X100 bridge expansion unit, with functional expansion through the PCIe bus. The CPU uses a Tengrui D2000 / 8 eight-core processor with a main frequency of 2.0GHz and a TDP power consumption of 20W. The storage capacities are 512GB, 640GB, 832GB and 1TB respectively, and the 4k random read and write speeds are 300MB / s, 350MB / s, 100MB / s and 150MB / s. The total amount of experimental data is set to 1025.4GB, and the value coefficient is λ={λ1,λ2,...,λ DS } and are set to 0.35, 0.25, 0.2, and 0.2 respectively.

10. The distributed storage resource scheduling method based on improved OBXGA according to claim 9, characterized in that: The experiment verifies the difference in the best fitness of different algorithms under the same initial parameter conditions by iterating 100 times and 200 times, as well as the change in the best fitness of various algorithms 600 times. The algorithm is verified by comparing the parameter values.

Images