Method for solving flexible job shop scheduling problem based on improved MOEA / D optimization algorithm

Through the improved MOEA/D algorithm, using integer encoding and adaptive weight adjustment, the problem of insufficient solutions in multi-objective flexible operation workshop scheduling is solved, and stronger convergence and diversity are achieved. More scheduling solutions.

CN118982185APending Publication Date: 2024-11-19HUAIYIN INSTITUTE OF TECHNOLOGY
View PDF 0 Cites 1 Cited by

Patent Information

Application Number
CN202411047743.4
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-01
Publication Date
2024-11-19

AI Technical Summary

Technical Problem

The existing multi-objective flexible operation workshop scheduling algorithm loses sensitivity when processing complex Pareto cutting edge shapes, resulting in insufficient diversity and uniformity of solutions and is unable to effectively cover the entire Pareto cutting edge.

Method used

The improved MOEA/D algorithm uses integer encoding and adaptive weight adjustment, combined with Chebishev function values ​​and neighborhood search to enhance the diversity and uniformity of understanding, ensuring that the populations are convergent and sparse in evolutionary process match Pareto frontiers. .

Benefits of technology

The solution convergence and diversity of flexible work workshop scheduling problems can be improved, and more valuable scheduling solutions can be obtained, covering a wider range of Pareto optimal solution sets.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118982185A_ABST
    Figure CN118982185A_ABST
Patent Text Reader

Abstract

The invention discloses a method for solving a flexible job shop scheduling problem based on an improved MOEA / D optimization algorithm. The method comprises the following steps: establishing a multi-target flexible job shop scheduling problem mathematical model; performing coding processing on the flexible job shop scheduling problem; setting initialization parameters; solving based on an improved MOEA / D algorithm; the improved MOEA / D algorithm not only has strong convergence, but also can adaptively adjust the weight vector according to the sparsity of the population, so that the weight vector can be matched with the Pareto leading edge, and the diversity and uniformity of understanding are increased; by using the improved MOEA / D algorithm to solve the flexible job shop scheduling problem, more valuable scheduling schemes can be obtained.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention relates to a method for solving a flexible job shop scheduling problem, and in particular to a method for solving a flexible job shop scheduling problem based on an improved MOEA / D optimization algorithm, and belongs to the technical field of flexible job shop scheduling. Background Art

[0002] Shop scheduling is an important enabling tool for intelligent manufacturing. A good shop scheduling solution can not only meet the diverse production needs of customers, but also reduce production costs and shorten delivery time. In the traditional shop scheduling model, each process of each workpiece can only be processed on one machine and only once, and the processing time is fixed. As the era of large-scale continuous production is gradually being replaced by multi-variety, small-batch discrete production that adapts to market dynamics, the production capacity and competitiveness of an enterprise depend to a large extent on whether it can reduce production costs, shorten production cycles, and produce multi-variety products with higher quality. Traditional shop scheduling is no longer suitable for the needs of today's enterprises. Flexible job shop scheduling is an extension of traditional shop scheduling. It has high "flexibility". Each process of the workpiece can be processed on multiple machines, and the processing time is not necessarily the same. Each process of the workpiece can be processed on multiple identical or different machines, which can flexibly select resources and improve processing flexibility. Flexible job shop scheduling has become a hot topic of concern in the industry in the past decade due to its excellent equipment utilization, production stability and production efficiency.

[0003] Flexible job shop scheduling can be divided into two cases: single-objective and multi-objective. Domestic and foreign scholars' research on single-objective flexible job shop scheduling mainly focuses on optimizing the maximum completion time. Pezzella et al. designed a genetic algorithm (GA), which uses a strategy of mixing two machine allocation rules and three process sorting rules to generate a high-performance initial population. Wang et al. proposed a dual-population distribution estimation algorithm, which uses two sub-populations to adjust machine allocation and process sorting sequences respectively, and uses the dominant population to build a probability model, which is then used to generate new individuals. Zhang et al. designed a machine GLR selection method considering machine load balancing, which includes global selection, local selection and random selection. Although the single-objective flexible job shop scheduling problem has been widely studied, the actual scheduling production often considers multiple objectives. Therefore, it is very necessary to study multi-objective flexible job shop scheduling. There are roughly two methods to solve the multi-objective flexible job shop scheduling problem: a priori method and a posteriori method. In the a priori method, multiple objectives are often formed into a single objective through a linear weighted method, but this method can only obtain one or a few solutions and cannot achieve a compromise between the various objectives. In addition, this method often makes it difficult to accurately obtain the weight of each objective. Xia and Wu discussed a hybrid algorithm that used PSO and simulated annealing (SA) to deal with machine allocation and process sequencing problems respectively. Gao et al. proposed a GA combined with a bottleneck drift local search method. Zhang et al. proposed a method that effectively mixed PSO and taboo search (TS). The difference between the a posteriori method and the priori method is that the Pareto dominance relationship is used to compare the advantages and disadvantages of the solutions, and a set of mutually non-dominated solutions can be obtained, which can reflect the trade-off relationship between different objectives. In recent years, domestic and foreign research on multi-objective flexible job shop scheduling has gradually tended to the a posteriori method. Wang et al. proposed a multi-objective genetic algorithm (MOGA), which uses immune entropy to maintain the diversity of the population. Frutos et al. combined the second-generation non-dominated sorting genetic algorithm (NSGA-II) with local search technology to solve this type of problem. Chiang and Lin proposed a simple and effective MOEA, which only requires two parameters: population size and evolutionary generations.

[0004] In 2007, Zhang et al. proposed a multi-objective evolutionary algorithm based on decomposition (MOEA / D), which has now developed into one of the mainstream frameworks for solving multi-objective optimization problems. In the MOEA / D algorithm, the idea of ​​decomposition is used to decompose the multi-objective optimization problem into multiple scalar optimization sub-problems and optimize them simultaneously. When decomposing multi-objective optimization sub-problems, MOEA / D needs to rely on a set of uniformly distributed weight vectors. The commonly used decomposition methods are weighted sum method, Chebyshev method and boundary intersection method, among which Chebyshev decomposition method is the most widely used. When optimizing each sub-problem, MOEA / D only uses the relevant information of the neighboring sub-problems of the sub-problem, which greatly reduces the computational complexity of the algorithm. In 2013, Qi et al. proposed an adaptive weight-adjusted MOEA / D (MOEA / D-AWA). The algorithm proposed a new weight vector initialization method and weight vector adaptive adjustment strategy. In 2016, Yang et al. studied the impact of penalty factors on PBI and proposed two new penalty schemes: using adaptive penalty scheme and sub-problem-based penalty scheme (SPS) to enhance the ability of MOEA / D to search for Pareto optimal solutions to responsible problems. In 2016, Jiang et al. proposed an improved MOEA / D method based on a two-stage strategy (TP) and niche guidance, namely MOEA / D-TPN. In this algorithm, a two-stage strategy (TP) is used to divide the entire optimization process into two stages. Based on the congestion of the solution in the first stage, the algorithm decides whether to consume computing resources to deal with the unsolved subproblems in the second stage. In addition, a new niche scheme is introduced in the improved MOEA / D to guide the selection of mating parents and avoid duplicate solutions, which is very helpful for maintaining population diversity.

[0005] For multi-objective optimization problems with complex Pareto front shapes, since complex Pareto front shapes mainly include four forms: spikes, long tails, discontinuities, and degradation, the decomposition-based multi-objective evolutionary algorithm (MOEA / D) uses uniformly distributed weight vectors to guide the solution to converge to the Pareto optimal front to obtain the optimal solution. However, on the complex Pareto front, MOEA / D will lose its sensitivity to the Pareto front, resulting in the optimal solution set obtained by the algorithm not being able to cover the entire Pareto front and being unevenly distributed. Summary of the invention

[0006] Purpose of the invention: The purpose of the present invention is to provide a method for solving flexible job shop scheduling problems based on an improved MOEA / D optimization algorithm that can increase the diversity and uniformity of solutions.

[0007] Technical solution: The method for solving the flexible job shop scheduling problem based on the improved MOEA / D optimization algorithm described in the present invention comprises:

[0008] Step 1: Establish a mathematical model for multi-objective flexible job shop scheduling problem;

[0009] Step 2: Encode the flexible job shop scheduling problem;

[0010] Step 3: Set initialization parameters;

[0011] Step 4: Solve based on the improved MOEA / D algorithm.

[0012] Furthermore, step 1 is specifically as follows: taking the minimum maximum completion time, the minimum maximum load machine and the minimum total machine load as the objective function, a mathematical model is established as follows:

[0013]

[0014] Where the objective function f1 represents the minimum maximum completion time, f2 represents the minimum maximum load machine, and f3 represents the minimum total machine load; C j represents the completion time of each workpiece, p ijh represents the processing time of the hth process of the jth workpiece on machine i; x ijh is a conditional parameter. When it is 1, it means that the hth process of the jth workpiece selects machine i, otherwise it is 0; n represents the total number of workpieces; m represents the total number of machines; h j Represents the total number of processes for the jth job.

[0015] Furthermore, it is characterized in that the mathematical model has the following constraints:

[0016] The same machine can only process one workpiece at a time;

[0017] The same process of the same workpiece can only be processed by one machine at the same time;

[0018] Once each process of each workpiece begins, the processing cannot be interrupted;

[0019] Different artifacts have the same priority;

[0020] There is no order constraint between the processes of different workpieces, but there is order constraint between the processes of the same workpiece;

[0021] All workpieces can be processed at time zero.

[0022] Furthermore, the step 2 specifically uses integer coding MSOS to encode chromosomes, and the integer coding MSOS includes: a machine selection part and a process sorting part;

[0023] In the machine selection part, the chromosome length is the total number of all workpiece processes, each gene bit is represented by an integer, and is arranged in the order of workpieces and workpiece processes, and each integer represents the sequential number of the processing machine selected for the current process in the set of selectable machines;

[0024] The process sorting part is encoded using a process-based encoding method. The chromosome length is equal to the sum of the processes of all workpieces. Each gene is directly encoded with the workpiece number. The order in which the workpiece numbers appear represents the processing order between the workpiece processes.

[0025] Furthermore, the step 3 specifically includes:

[0026] Step 3.1: Set the population size N and initialize the population x 1 , ..., x N , evaluate the fitness value of the population;

[0027] Among them, the initialization of the machine selection part is carried out using the GLR machine selection method, and the initialization of the process sorting part is produced by random method;

[0028] Step 3.2: Initialize the weight vector and the number of neighbors T;

[0029] Each population generates a weight vector, generating N uniformly distributed weight vectors: λ 1 , ..., λ N , use Euclidean distance to calculate the T nearest neighbor vectors to each weight vector; for B(l) = {l1, ..., l T}, l = 1, .., N, where B(l) is the neighborhood subproblem of the lth subproblem, l T is the Tth neighborhood subproblem of the lth subproblem, is the distance λ l The nearest T neighbor vectors;

[0030] Step 3.3: Initialize reference point z * , the reference point is the minimum value of the objective functions f1, f2 and f3;

[0031] Step 3.4: Set the stopping criteria.

[0032] Furthermore, the GLR machine selection method described in step 3.1 includes: global selection, local selection and random selection.

[0033] Furthermore, the global selection is specifically to set an array whose length is equal to the number of machines, the order of the array corresponds to the order of the processing machines, and the value of each bit corresponds to the processing time on the machine; randomly select a workpiece from the workpiece set, starting from the first process of the current workpiece, add the processing time of the optional processing machine of the current process to the corresponding time in the array, select the shortest time as the processing machine of the current process, and update the array.

[0034] Furthermore, the local selection is specifically to set an array with a length equal to the number of machines, select the first workpiece in the workpiece set, start with the first process of the current workpiece, add the processing time of the selectable processing machines of the current process to the corresponding time in the array, select the shortest time as the processing machine for the current process, and update the array, and so on until the processing machines for all processes of the current workpiece are selected, then reset each bit of the array to zero, select the next workpiece, and until all workpieces are selected.

[0035] Furthermore, the random selection is specifically to make the initial population dispersed in the solution space, and the machine selection part of a part of the population adopts a random initialization method; the number on each gene position, that is, the sequence number of the process that can select the machine set, is randomly generated.

[0036] Furthermore, the step 4 specifically includes:

[0037] Step 4.1: Select the parent;

[0038] The parent randomly selects a neighboring subproblem from the T neighborhoods of the subproblem;

[0039] Step 4.2: Parent crossover operation;

[0040] Uniform crossover operation is used in the machine selection part, and POX crossover operation is used in the process sorting part;

[0041] Step 4.3: mutation operation;

[0042] The reverse mutation method is used in the machine selection part, and the neighborhood search-based mutation operation is used in the process sorting part;

[0043] Step 4.4: Evaluate the fitness value of the child solution and update the reference point z * ;

[0044] Step 4.5: Replace and save the solution;

[0045] The replacement of the child solution with the neighborhood solution of the parent child problem is based on the comparison of the Chebyshev function value. The formula is as follows:

[0046]

[0047] Among them, λ a is the weight vector corresponding to the a-th subproblem, M is the number of objective functions, f b (x) represents the b-th objective function value corresponding to subproblem x, is the minimum value of the objective function b in the entire population; if the Chebyshev value of the neighborhood solution of the parent subproblem is greater than or equal to the Chebyshev value of the child solution, the neighborhood solution is replaced by the child solution and the replaced neighborhood solution is liberated into the archive;

[0048] Step 4.6: Filter the solutions in the archive and mix them with the solutions in the population;

[0049] After each iteration of all populations, the solutions in the archive are screened to remove duplicate solutions and solutions of no use value, and then the solutions in the archive are mixed with the solutions in the population;

[0050] The removing of duplicate solutions and solutions without use value specifically includes, after removing duplicate solutions, making value judgments on the solutions in the archive, calculating the Chebyshev function value of each solution in the archive and the weight vector one by one, and for each weight vector, only retaining the archive solution with the smallest Chebyshev value with the weight vector;

[0051] The mixing of the solutions in the archive with the solutions in the population is specifically to determine whether the Chebyshev function value of each solution in the population and the corresponding weight vector is greater than the Chebyshev function value of the same weight vector in the archive, and if so, replace the solution in the population with the solution in the archive;

[0052] Step 4.7: Adaptively adjust the weight vector;

[0053] After each iteration of the population, the weight vector is adaptively adjusted. The adaptive adjustment strategy includes:

[0054] According to the crowding degree of the current population, several most crowded sub-problems and corresponding weight vectors are deleted. The evaluation index of crowding degree is represented by the neighbor distance, which is specifically the distance of another individual closest to the current individual.

[0055] Use the population in the archive to calculate the crowding degree of the current population, find out the sparsest individuals in the archive relative to the current population and add them to the current population;

[0056] Assign a new weight vector λ′ to the supplemented individuals, and the calculation formula is as follows:

[0057]

[0058] where z * is the reference point, M = 3 represents the number of objective functions;

[0059] Step 4.8: Recalculate the T neighbors of the weight vector;

[0060] Step 4.9: Determine whether the improved MOEA / D algorithm meets the stopping criteria. If not, go to step 4.1 to continue. If it meets the stopping criteria, output the Pareto optimal solution set and the target vector corresponding to the Pareto optimal solution set. The target vector is the maximum completion time, maximum machine load and total machine load required for the scheduling problem.

[0061] Beneficial effects: Compared with the prior art, the present invention archives the replaced neighborhood solutions, mixes the solutions in the archive with the solutions in the population according to the aggregation function value, and eliminates the solutions with larger aggregation function values ​​under the guidance of the same vector weight, thereby improving the convergence of the population evolution process; and deletes the crowded sub-problems and corresponding weight vectors according to the congestion of the solutions in the population, and then fills the sparse areas of the population with the solutions in the archive and creates a new weight vector; the improved MOEA / D algorithm of the present invention not only has strong convergence, but also can adaptively adjust the weight vector according to the sparsity of the population so that the weight vector can match the Pareto frontier, thereby increasing the diversity and uniformity of the solutions; using the improved MOEA / D algorithm to solve the flexible job shop scheduling problem can obtain more valuable scheduling solutions. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] Figure 1 is a flow chart of a method according to an embodiment of the present invention;

[0063] Figure 2 This is a result diagram of a first complex problem test function according to an embodiment of the present invention;

[0064] Figure 3 This is a result diagram of the second complex problem test function according to an embodiment of the present invention;

[0065] Figure 4 A distribution diagram of the solution of the MK06 example of an embodiment of the present invention;

[0066] Figure 5 It is a distribution diagram of the solution of the MK10 example according to an embodiment of the present invention. DETAILED DESCRIPTION

[0067] The technical solution of the present invention is further described below in conjunction with the accompanying drawings and embodiments.

[0068] As attached Figure 1 As shown, the method for solving the flexible job shop scheduling problem based on the improved MOEA / D optimization algorithm of this embodiment includes:

[0069] Step 1: Establish a mathematical model for multi-objective flexible job shop scheduling problem;

[0070] Step 2: Encode the flexible job shop scheduling problem;

[0071] Step 3: Set initialization parameters;

[0072] Step 4: Solve based on the improved MOEA / D algorithm.

[0073] Furthermore, in step 1, the flexible job shop scheduling problem (FJSP) is described as follows: n workpieces are to be processed on m machines; each workpiece contains one or more processes; the process sequence is predetermined; each process can be processed on multiple different processing machines; the processing time of the process varies with the processing machines; the scheduling goal is to select the most suitable machine for each process, determine the optimal processing sequence and start time of each process on each machine, and optimize certain performance indicators of the entire system.

[0074] This embodiment optimizes the three most common minimization objectives of the multi-objective flexible job shop scheduling problem, namely, minimizing the maximum completion time, minimizing the maximum load machine, and minimizing the total machine load.

[0075] The specific mathematical model is as follows:

[0076]

[0077] Where the objective function f1 represents the minimum maximum completion time, f2 represents the minimum maximum load machine, and f3 represents the minimum total machine load; C j represents the completion time of each workpiece, p ijh represents the processing time of the hth process of the jth workpiece on machine i; x ijh is a conditional parameter. When it is 1, it means that the hth process of the jth workpiece selects machine i, otherwise it is 0; n represents the total number of workpieces; m represents the total number of machines; h j Represents the total number of processes for the jth job.

[0078] The mathematical model has the following constraints:

[0079] The same machine can only process one workpiece at a time.

[0080] The same process of the same workpiece can only be processed by one machine at the same time.

[0081] Once each process of each workpiece begins, the processing cannot be interrupted.

[0082] Different artifacts have the same priority.

[0083] There are no sequence constraints between the processes of different workpieces, but there are sequence constraints between the processes of the same workpiece.

[0084] All workpieces can be processed at time zero.

[0085] In step 2, chromosome encoding is to express the solution of the problem in the form of chromosomes. The purpose of encoding is to achieve genetic operations similar to those in the biological world, such as crossover and mutation. FJSP needs to solve two problems, namely machine selection and process sorting. The present invention uses integer encoding MSOS, which consists of two parts: machine selection part MS and process sorting part OS.

[0086] Machine selection part: The chromosome length of the machine selection part is the total number of all workpiece processes. Each gene position is represented by an integer and arranged in the order of workpieces and workpiece processes. Each integer represents the sequential number of the processing machine selected for the current process in the set of selectable machines. This ensures that the solution generated after subsequent crossover, mutation and other operations is still a feasible solution.

[0087] Process sorting part: The process-based coding method proposed by Gen is used for coding. The chromosome length is equal to the sum of the processes of all workpieces. Each gene is directly encoded with the workpiece number. The order in which the workpiece number appears indicates the order of processing between the processes of the workpiece.

[0088] Step 3 includes:

[0089] Step 3.1: Set the population size N = 100, initialize the population, and evaluate the fitness value of the population;

[0090] Initialize the population x 1 , ..., x N . Population initialization is a key issue in evolutionary algorithms, and the quality of initialization has a great impact on the algorithm's solution speed and quality. Flexible job shop scheduling not only needs to consider the machine selection problem but also needs to solve the process sequencing problem, but machine selection is more important than process sequencing. The present invention uses the GLR machine selection method to initialize the machine selection part, and the initialization of the process sequencing part is produced by a random method. The GLR machine selection method includes: global selection, local selection and random selection. Global selection and local selection are mainly to consider the load problem of machine selection so that the workload of each selected machine is balanced as much as possible and the utilization rate of the machine is fully improved. Random selection mainly considers distributing the initialized population as dispersedly as possible in the entire solution space. The organic combination of the three improves the quality of the initial solution in the machine selection part.

[0091] Global selection: Set an array with a length equal to the number of machines. The order of the array corresponds to the order of the processing machines, and the value of each bit corresponds to the processing time on the machine. Randomly select a workpiece from the workpiece set, starting from the first process of the current workpiece, add the processing time of the optional processing machine of the current process to the corresponding time in the array, select the shortest time as the processing machine of the current process, and update the array.

[0092] Local selection: Set up an array with a length equal to the number of machines, select the first workpiece in the workpiece set, start with the first process of the current workpiece, add the processing time of the selectable processing machines of the current process to the corresponding time in the array, select the shortest time as the processing machine for the current process, and update the array, and so on until the processing machines for all processes of the current workpiece are selected, reset each bit of the array to zero, select the next workpiece, and until all workpieces are selected.

[0093] Random selection: To ensure the diversity of the initial population, the initial population should be dispersed in the solution space, and the machine selection part of a part of the population adopts a random initialization method. The number on each gene position, which means the sequence number of the process selectable machine set, is randomly generated.

[0094] Step 3.2: Initialize the weight vector and the number of neighbors T

[0095] Generate N uniformly distributed weight vectors: λ 1 , ..., λ N , and then use the Euclidean distance to calculate the T nearest neighboring vectors to each weight vector. For l = 1, .., N, B(l) = {l1, ..., l T},in is the distance λ l The most recent T weight vectors;

[0096] Step 3.3: Initialize reference point z * : The reference point is the minimum value of the above three objective functions;

[0097] Step 3.4: Set the stopping criteria.

[0098] Step 4 includes:

[0099] Step 4.1: Parent selection;

[0100] During the evolution process, each population is a subproblem and parents are randomly selected from the T neighborhoods of the subproblem.

[0101] Step 4.2: Crossover operation of the parent generation;

[0102] The purpose of the crossover operation is to generate new individuals from the parent individuals through a series of operations, and to efficiently search the solution space on the basis of minimizing the probability of the effective mode being destroyed. The present invention adopts uniform crossover operation in the machine selection part and POX crossover operation in the process sorting part.

[0103] Step 4.3: mutation operation;

[0104] The mutation operation produces new individuals by randomly changing some genes of the chromosome to make minor disturbances to the chromosome, increasing the diversity of the population and affecting the local search capability to a certain extent. The present invention adopts the reverse mutation method in the machine selection part and adopts the neighborhood search-based mutation operation in the process sorting part.

[0105] Step 4.4: Evaluate the fitness value of the child solution and update the reference point z * ;

[0106] Step 4.5: Replacement and archiving of solutions;

[0107] The replacement of the child solution with the neighborhood solution of the parent subproblem is based on the comparison of the Chebyshev function value. The Chebyshev function formula of the a-th subproblem is as follows:

[0108]

[0109] in is the weight vector corresponding to the a-th subproblem, M is the number of objective functions, and f b (x) represents the a-th objective function value corresponding to subproblem x.

[0110] If the Chebyshev value of the neighborhood solution of the parent subproblem is greater than or equal to the Chebyshev value of the child solution, the neighborhood solution is replaced by the child solution and the replaced neighborhood solution is released to the archive.

[0111] Step 4.6: Screening of solutions in the archive and mixing of the population;

[0112] After each iteration of all populations, the solutions in the archive need to be screened to remove duplicate solutions and solutions that are of no use, and then the solutions in the archive are mixed with the solutions in the population to improve the convergence of the population. After removing duplicate solutions, it is necessary to make a value judgment on the solutions in the archive. The value judgment is still based on the Chebyshev function. The Chebyshev function value is calculated for each solution in the archive and the weight vector one by one. For each weight vector, only the archive solution with the smallest Chebyshev value with the weight vector is retained. The mixing of the population is whether the Chebyshev function value of each solution in the population and the corresponding weight vector is greater than the Chebyshev function value of the same weight vector in the archive. If it is greater, the solution in the population is replaced with the solution in the archive to improve the convergence of the population.

[0113] Step 4.7: Adaptive weight vector;

[0114] In the present invention, the population will perform adaptive adjustment on the weight vector every 20 iterations. The adaptive adjustment strategy is divided into three parts, as follows:

[0115] According to the crowding degree of the current population, the five most crowded sub-problems and the corresponding weight vectors are deleted. The evaluation index of crowding degree is expressed by the neighbor distance, that is, the distance to the closest individual to the current individual;

[0116] Use the population in the archive to calculate the crowding degree of the current population, find the 5 most sparse individuals in the archive relative to the current population and add them to the current population;

[0117] Assign a new weight vector λ′ to the five supplemented individuals, and the calculation formula is as follows:

[0118]

[0119] where z * is the reference point, M represents the number of objective functions, and since the present invention studies the three minimization objectives of the flexible job shop scheduling problem, M is 3 here.

[0120] Step 4.8: Recalculate the T neighbors of the weight vector. Since the weight vector was adjusted in the previous step, the T neighbors of each weight vector need to be recalculated.

[0121] Step 4.9: Determine whether the algorithm meets the stopping criteria. If not, go to step 4.1 to continue; if it meets the stopping criteria, output the Pareto optimal solution set and the target vector corresponding to the Pareto optimal solution set.

[0122] The output Pareto optimal solution set will form a specific activity scheduling plan after chromosome decoding. The objective vector corresponding to each Pareto optimal solution is the maximum completion time, maximum machine load and total machine load required by the scheduling plan. The number of different optimal solutions in the Pareto optimal solution set can show the diversity of the algorithm and provide users with more scheduling plans. For complex multi-objective optimization problems such as flexible job shop scheduling, the improved algorithm invented in this paper can obtain more scheduling plans. Under the same population size, the number of solutions obtained is greater than that of the traditional MOEA / D algorithm and NSGA-II algorithm.

[0123] The present invention proposes an improved MOEA / D algorithm. In MOEA / D, the update of the neighborhood solution is determined by the size of the aggregation function value. If the aggregation function value of the neighborhood solution is greater than the aggregation function value of the newly generated child solution, the neighborhood solution will be replaced by the child solution, but we found that some of the replaced solutions are also valuable and will supplement the sparse area of ​​the optimal solution. The improved MOEA / D algorithm performs well in many multi-objective optimization algorithm test functions. We use two multi-objective test functions with complex Pareto frontiers to test the improved MOEA / D. The results of the first complex problem test function are as follows: Figure 2 As shown, the result of the second complex problem test function is as follows Figure 3 shown.

[0124] The present invention archives the replaced neighborhood solutions, mixes the solutions in the archive with the solutions in the population according to the aggregation function value, and removes the solutions with larger aggregation function values ​​under the guidance of the same vector weight, thereby improving the convergence of the population evolution process; at the same time, according to the crowding degree of the solutions in the population, the crowded sub-problems and the corresponding weight vectors are deleted, and the sparse areas of the population are filled with the solutions in the archive and a new weight vector is created. The improved MOEA / D algorithm of the present invention not only has a strong convergence, but also can adaptively adjust the weight vector according to the sparsity of the population so that the weight vector can match the Pareto frontier to increase the diversity and uniformity of the solutions. Figure 4 and Figure 5 ,Using the improved MOEA / D algorithm to solve MK06 and MK10 in the flexible job shop scheduling problem example given by Brandimarte, more valuable scheduling solutions can be obtained; and better results have been achieved.

Claims

1. A method for solving the flexible job shop scheduling problem based on an improved MOEA / D optimization algorithm, characterized in that: include: Step 1: Establish a mathematical model for multi-objective flexible job shop scheduling problem; Step 2: Encode the flexible job shop scheduling problem; Step 3: Set initialization parameters; Step 4: Solve based on the improved MOEA / D algorithm.

2. The method for solving the flexible job shop scheduling problem based on the improved MOEA / D optimization algorithm according to claim 1 is characterized in that: Step 1 is as follows: Taking the minimum maximum completion time, the minimum maximum load machine and the minimum total machine load as the objective function, the mathematical model is established as follows: Where the objective function f1 represents the minimum maximum completion time, f2 represents the minimum maximum load machine, and f3 represents the minimum total machine load; C j represents the completion time of each workpiece, p ijh represents the processing time of the hth process of the jth workpiece on machine i; x ijh is a conditional parameter. When it is 1, it means that the hth process of the jth workpiece selects machine i, otherwise it is 0; n represents the total number of workpieces; m represents the total number of machines; h j Represents the total number of processes for the jth job.

3. The method for solving the flexible job shop scheduling problem based on the improved MOEA / D optimization algorithm according to claim 2 is characterized in that: The mathematical model has the following constraints: The same machine can only process one workpiece at a time; The same process of the same workpiece can only be processed by one machine at the same time; Once each process of each workpiece begins, the processing cannot be interrupted; Different artifacts have the same priority; There is no order constraint between the processes of different workpieces, but there is order constraint between the processes of the same workpiece; All workpieces can be processed at time zero.

4. The method for solving the flexible job shop scheduling problem based on the improved MOEA / D optimization algorithm according to claim 3 is characterized in that: The step 2 specifically uses integer coding MSOS to encode chromosomes, and the integer coding MSOS includes: a machine selection part and a process sorting part; In the machine selection part, the chromosome length is the total number of all workpiece processes, each gene position is represented by an integer, and is arranged in the order of workpieces and workpiece processes, and each integer represents the sequential number of the processing machine selected for the current process in the set of selectable machines; The process sorting part is encoded using a process-based encoding method. The chromosome length is equal to the sum of the processes of all workpieces. Each gene is directly encoded with the workpiece number. The order in which the workpiece numbers appear represents the processing order between the workpiece processes.

5. The method for solving the flexible job shop scheduling problem based on the improved MOEA / D optimization algorithm according to claim 4 is characterized in that: The step 3 specifically includes: Step 3.1: Set the population size N and initialize the population x 1 , ..., x N , evaluate the fitness value of the population; Among them, the initialization of the machine selection part is carried out using the GLR machine selection method, and the initialization of the process sorting part is produced by random method; Step 3.2: Initialize the weight vector and the number of neighbors T; Generate a weight vector for each population, and generate N uniformly distributed weight vectors: λ 1 ,...,λ N , use Euclidean distance to calculate the T nearest neighbor vectors to each weight vector; for B(l) = {l1, ..., l T }, l = 1, .., N, where B(l) is the neighborhood subproblem of the lth subproblem, l T is the Tth neighborhood subproblem of the lth subproblem, is the distance λ l The nearest T neighbor vectors; Step 3.3: Initialize reference point z * , the reference point is the minimum value of the objective functions f1, f2 and f3; Step 3.4: Set the stopping criteria.

6. The method for solving the flexible job shop scheduling problem based on the improved MOEA / D optimization algorithm according to claim 5 is characterized in that: The GLR machine selection method described in step 3.1 includes: global selection, local selection and random selection.

7. The method for solving the flexible job shop scheduling problem based on the improved MOEA / D optimization algorithm according to claim 6 is characterized in that: The global selection is specifically to set an array whose length is equal to the number of machines, the order of the array corresponds to the order of the processing machines, and the value of each bit corresponds to the processing time on the machine; randomly select a workpiece from the workpiece set, start from the first process of the current workpiece, add the processing time of the optional processing machine of the current process to the corresponding time in the array, select the shortest time as the processing machine of the current process, and update the array.

8. The method for solving the flexible job shop scheduling problem based on the improved MOEA / D optimization algorithm according to claim 6 is characterized in that: The local selection is specifically to set an array with a length equal to the number of machines, select the first workpiece in the workpiece set, start with the first process of the current workpiece, add the processing time of the selectable processing machines of the current process to the corresponding time in the array, select the shortest time as the processing machine for the current process, and update the array, and so on until the processing machines for all processes of the current workpiece are selected, then reset each bit of the array to zero, select the next workpiece, and until all workpieces are selected.

9. The method for solving the flexible job shop scheduling problem based on the improved MOEA / D optimization algorithm according to claim 6 is characterized in that: The random selection is specifically to make the initial population dispersed in the solution space, and the machine selection part of a part of the population adopts a random initialization method; the number on each gene position, that is, the sequence number of the process that can select the machine set, is randomly generated.

10. The method for solving the flexible job shop scheduling problem based on the improved MOEA / D optimization algorithm according to claim 5 is characterized in that: The step 4 specifically includes: Step 4.1: Select the parent; The parent randomly selects a neighboring subproblem from the T neighborhoods of the subproblem; Step 4.2: Parent crossover operation; Uniform crossover operation is used in the machine selection part, and POX crossover operation is used in the process sorting part; Step 4.3: mutation operation; The reverse mutation method is used in the machine selection part, and the neighborhood search-based mutation operation is used in the process sorting part; Step 4.4: Evaluate the fitness value of the child solution and update the reference point z * ; Step 4.5: Replace and save the solution; The replacement of the child solution with the neighborhood solution of the parent child problem is based on the comparison of the Chebyshev function value. The formula is as follows: Among them, λ a is the weight vector corresponding to the a-th subproblem, M is the number of objective functions, f b (x) represents the b-th objective function value corresponding to subproblem x, is the minimum value of the objective function b in the entire population; if the Chebyshev value of the neighborhood solution of the parent subproblem is greater than or equal to the Chebyshev value of the child solution, the neighborhood solution is replaced by the child solution and the replaced neighborhood solution is liberated into the archive; Step 4.6: Filter the solutions in the archive and mix them with the solutions in the population; After each iteration of all populations, the solutions in the archive are screened to remove duplicate solutions and solutions of no use value, and then the solutions in the archive are mixed with the solutions in the population; The removing of duplicate solutions and solutions without use value specifically includes, after removing duplicate solutions, making value judgments on the solutions in the archive, calculating the Chebyshev function value of each solution in the archive and the weight vector one by one, and for each weight vector, only retaining the archive solution with the smallest Chebyshev value with the weight vector; The mixing of the solutions in the archive with the solutions in the population is specifically to determine whether the Chebyshev function value of each solution in the population and the corresponding weight vector is greater than the Chebyshev function value of the same weight vector in the archive, and if so, replace the solution in the population with the solution in the archive; Step 4.7: Adaptively adjust the weight vector; After each iteration of the population, the weight vector is adaptively adjusted. The adaptive adjustment strategy includes: According to the crowding degree of the current population, several most crowded sub-problems and corresponding weight vectors are deleted. The evaluation index of crowding degree is represented by the neighbor distance, which is specifically the distance of another individual closest to the current individual. Use the population in the archive to calculate the crowding degree of the current population, find out the sparsest individuals in the archive relative to the current population and add them to the current population; Assign a new weight vector λ′ to the supplemented individuals, and the calculation formula is as follows: where z * is the reference point, M = 3 represents the number of objective functions; Step 4.8: Recalculate the T neighbors of the weight vector; Step 4.9: Determine whether the improved MOEA / D algorithm meets the stopping criteria. If not, go to step 4.1 to continue. If the stopping criteria are met, the Pareto optimal solution set and the target vector corresponding to the Pareto optimal solution set are output. The objective vector is the maximum completion time, maximum machine load and total machine load required by the scheduling problem.

Citation Information

Cited By

  • Foaming machine group production scheduling method under multi-target constraint

    CN121433149A