Multi-processing-module wafer manufacturing scheduling method based on improved NSGA-II algorithm
Through the improved NSGA-II algorithm, combined with double-layer coding and local search strategies, the workpiece processing sequence and machine selection are optimized, and the problem of high power loss during workpiece processing and cooling in wafer manufacturing workshops is solved, and the maximum completion time is reduced and total energy consumption effect.
Patent Information
- Application Number
- CN202411291582.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-14
- Publication Date
- 2024-12-03
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
In wafer manufacturing workshop, workpieces that need to be processed on multiple machines are high in power losses due to processes such as processing, cooling and handling, and the cooling area capacity is limited, making it difficult to effectively schedule the total instantaneous power, resulting in low production efficiency and energy Increased consumption.
Using the improved NSGA-II algorithm, the workpiece processing sequence and machine selection are optimized through bilayer-encoded chromosome encoding, heuristic pheromone population initialization, uniform crossover operator crossing, random exchange variation and local search strategies, combining the shortest processing time priority and Elite retention strategy designed a multi-objective genetic algorithm to reduce maximum completion time and total energy consumption.
It effectively reduces the maximum completion time and energy consumption of production, improves the efficiency of wafer manufacturing scheduling, ensures the energy efficiency optimization of the workshop, avoids the fall of local optimal solutions, and improves the optimization speed of the algorithm and the accuracy of decoding.
Smart Images

Figure CN119067401A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to wafer production scheduling technology, in particular to a multi-processing module wafer manufacturing scheduling method based on an improved NSGA-II algorithm. Background Art
[0002] In a wafer manufacturing workshop, there are a number of independent workpieces that need to be processed on multiple machines. Each workpiece only needs to go through one process, but this process will go through multiple different stages, each with different processing time and power requirements. The processed workpieces need to be cooled in the cooling zone, and the capacity of the cooling zone is limited. The workshop is equipped with two types of manipulators: an atmospheric manipulator and two vacuum manipulators. The atmospheric manipulator is responsible for moving the workpieces to be processed from the entrance to the vacuum manipulator, and the vacuum manipulator then moves the workpieces to the machine for processing. After the processing is completed, the same vacuum manipulator will move the workpiece to the cooling zone, and the atmospheric manipulator will finally move the cooled workpiece to the exit. It should be noted that the processing of the machine and the insulation during the non-processing (i.e. idle) time will cause power loss; the capacity of the cooling zone is limited; at any time, the instantaneous total power of the workshop cannot exceed the maximum rated power; therefore, it is necessary to find a suitable scheduling scheme to meet the multi-objective solution under many constraints. Summary of the invention
[0003] In view of the above-mentioned defects, the purpose of the present invention is to propose a wafer manufacturing scheduling method, which can reduce the maximum completion time of production and reduce energy consumption.
[0004] To achieve this object, the present invention adopts the following technical solutions:
[0005] The multi-processing module wafer manufacturing scheduling method based on the improved NSGA-Ⅱ algorithm includes the following steps:
[0006] S100, based on the production scenario, production process and problem characteristics of multi-processing module wafer manufacturing scheduling problem, a mathematical model with maximum completion time and total energy consumption of the workshop as optimization objectives was constructed;
[0007] S200, encoding the chromosome using a double-layer encoding method based on a workpiece processing sequence and a machine selection sequence, wherein the lengths of the workpiece processing sequence and the machine selection sequence are the same as the number of workpieces;
[0008] S300, based on the population initialization of the heuristic pheromone, the processing time and power of each workpiece on the machine are different, and the workpiece needs to be preferentially assigned to the machine with the lowest processing energy consumption for processing; first calculate the average processing energy consumption of each workpiece, and then assign the workpiece to the machine with the lowest processing energy consumption for processing;
[0009] S400, crossover operation, introduce uniform crossover operator in the crossover stage, and optimize the uniform crossover operator with reference to the sequential crossover operator on this basis, and propose the sequential uniform crossover operator; use the sequential uniform crossover operator to operate the order processing sequence of chromosomes, and use the uniform crossover operator to operate the machine processing sequence; the parent generation of the crossover is selected by the tournament selection method, and the crossover probability P is set. c ;
[0010] S500, mutation operation. In order to avoid falling into the local optimum during the iteration process and generate a solution that jumps out of the local optimum, it is necessary to use a random exchange operator to exchange the genes at two positions on the order processing sequence / machine processing sequence layer of each parent chromosome for mutation, and set the mutation probability to P. m ;
[0011] S600, in order to speed up the optimization of the overall algorithm, a local search operator guided by problem characteristics is designed to minimize the maximum completion time and total energy consumption;
[0012] S700, decoding operation, adopts the scheduling strategy of shortest processing time priority; when the total power of the workshop exceeds the set maximum instantaneous power threshold, first calculate the excess power value, that is, the difference between the current total power of the workshop and the maximum instantaneous power; then find all the processing machine sets that can meet or exceed this excess power requirement, and then adjust the machine set; these sets may contain only a single machine or may cover multiple machines, but the number of machines in the set of multiple machines is strictly controlled within n-1, where n represents the total number of machines currently in operation;
[0013] S800, improved elite retention strategy, modifies the multi-objective genetic algorithm, sets parameters, then directly adds N×α individuals to the new population, and finally, randomly selects the remaining N×(1-α) individuals from the suboptimal plane, and outputs specific indicator values after decoding.
[0014] Preferably, in step S100:
[0015] In order to describe the problem and accurately establish the corresponding mathematical model, the following definitions are made:
[0016] At the same time, any machine can only process one workpiece;
[0017] Each workpiece can only be processed on one machine, and the machine cannot be preempted during processing;
[0018] The first processing position of each machine can be processed without preparation.
[0019] Failure of the processing machine or changes due to external factors are not considered;
[0020] In the process of solving the wafer manufacturing workshop scheduling problem, the objective function is needed to evaluate the advantages and disadvantages of various scheduling schemes; the optimization goals are to minimize the completion time, minimize the total energy consumption and maximize the finished product qualification rate;
[0021] (1) The maximum completion time is the smallest, which can be expressed as:
[0022]
[0023] (2) The total energy consumption of the workshop is the smallest. The total energy consumption of the workshop can be expressed as:
[0024]
[0025] Formula (2.1) and Formula (2.2) are the two objective functions used to solve the wafer manufacturing workshop scheduling problem.
[0026] Preferably, step S200 is specifically as follows:
[0027] The first layer of chromosomes is the order processing sequence, which can be expressed as {OS1, OS2, …, OS i ,OS i+1 ,…,OS n}, where OS i ∈{1,2,…,n}, used to represent the serial number of the i-th processed workpiece;
[0028] The second level of chromosomes is the machine-processed sequence, which can be represented as {MS1, MS2, …, MS i ,MS i+1 ,…,MS n}, where MS i ∈{1,2,…,m}, used to represent the order OS i The assigned processing machine serial number;
[0029] For the processing sequence of workpieces assigned to the same machine, first arrange the corresponding position numbers of these orders in the order processing sequence in ascending order, and then determine the processing sequence of these orders on the current machine based on their positions in the sorted list.
[0030] Preferably, in step S300:
[0031] The average operating energy consumption of the workpiece is calculated using formula (4.1);
[0032]
[0033] S311: First, the average processing energy consumption of each workpiece is calculated according to formula (4.1), and then the workpieces are sorted in the order of increasing energy consumption, so as to obtain an orderly workpiece processing queue;
[0034] S312: allocating machines to workpieces waiting for processing according to the principle of starting processing the workpieces earliest; if there are multiple machines available for processing the workpiece at the same time, the workpiece is allocated to the machine with the lowest processing energy consumption for processing;
[0035] S313: When the workpiece is machined, the machine processing start time and end time are updated in real time;
[0036] S314: Repeat the above steps until all workpieces are processed.
[0037] Preferably, step S400 is specifically as follows:
[0038] This paper uses the OUX crossover operator to operate the order processing sequence layer of the chromosome, and uses the UX crossover method to operate the machine processing sequence layer of the chromosome; using P1 OS / P1 MS and P2 OS / P2 MS Represents the two parent generations of the order processing sequence / machine processing sequence, using O1 OS / O1 MS and O2 OS / O2 MS Two children representing order processing sequence / machine processing sequence respectively;
[0039] The main steps of the two crossover operators are described as follows:
[0040] Description of OUX crossover operator:
[0041] S411: First, randomly generate a random number N in the interval [0.2*Popsize, 0.5*Popsize] c , and then randomly select N c The genes at position J constitute subset set ;
[0042] S412: From P1 OS / P2 OS Copy and J set The genes corresponding to the order in the O1OS / O2OS are copied in the same order and position.
[0043] S413: P1 OS / P2 OS Never in O2 OS / O1 OSThe genes that appeared above are placed in O2 according to their original order. OS / O1 OS Free space on the
[0044] Description of UX crossover operator:
[0045] S421: First, randomly generate a random number N in the interval [0.2*Popsize, 0.5*Popsize] c , and then randomly select genes at Nc positions to form a subset M set ;
[0046] S422: From P1 MS / P2 MS Copy and M set The genes corresponding to the order in O1 are copied to O1 in the same order and position. MS / O2 MS superior;
[0047] S423: P1 MS / P2 MS The remaining genes are copied to O2 in the original order and position. MS / O1 MS superior.
[0048] Preferably, step S500 is specifically as follows:
[0049] The mutated parent generation adopts the individuals in the population after crossover, and the mutation probability is P m ; Use POS / PMS to represent the parent of OS / MS, and use OOS / OMS to represent the child of OS / MS.
[0050] Preferably, in step S600:
[0051] For the same processing plan, since the scheduling sequence is known, we can select the order on the machine corresponding to the maximum completion time and move it to other machines for processing;
[0052] Since not every search can obtain a new solution that dominates the original scheduling solution, the following rules are set to constrain it:
[0053] Rule 1: To reduce the computational complexity of local search, set the search to a maximum of 5 times;
[0054] Rule 2: In order to avoid falling into the local optimum, set a 25% probability of accepting a solution that does not dominate the original solution in each search; that is, within 5 searches, if a new solution that dominates the objective function value of the original scheduling solution is found, then the new solution is set as the new offspring; otherwise, the new solution population after 5 searches is sorted by non-domination, congestion calculation, and population sorting, and the new solution with sequence number 1 in the new population is accepted as the new offspring with a probability of 25%;
[0055] Rule 3: In order to avoid local search taking too long, only individuals in the first frontier are set to participate in local search; if the number of individuals in the first frontier is greater than Searchsize, the first Searchsize individuals are selected for local search
[0056] One of the above technical solutions includes the following beneficial effects:. BRIEF DESCRIPTION OF THE DRAWINGS
[0057] Figure 1 is a workshop layout diagram of an embodiment of the present invention;
[0058] Figure 2 is an example diagram of chromosome encoding according to an embodiment of the present invention;
[0059] Figure 3 is an example diagram of chromosome decoding according to an embodiment of the present invention;
[0060] Figure 4 This is a Gantt chart before adjusting the machine in an example diagram of a scheduling strategy with shortest processing time priority according to an embodiment of the present invention;
[0061] Figure 5 This is a Gantt chart after adjusting the machine in an example diagram of a scheduling strategy with shortest processing time priority according to an embodiment of the present invention;
[0062] Figure 6 This is an example of an OUX crossover operator used by the chromosome OS layer and a UX crossover operator used by the MS layer according to an embodiment of the present invention;
[0063] Figure 7 is an example diagram of a mutation operation according to an embodiment of the present invention;
[0064] Figure 8 is an improved elite retention strategy of an embodiment of the present invention;
[0065] Fig. 9 is a flow chart of solving the improved NSGA-Ⅱ algorithm according to an embodiment of the present invention;
[0066] Fig.10 is a schematic diagram of the Pareto frontier found by the LNSGA-Ⅱ algorithm of an embodiment of the present invention;
[0067] Fig.11 is a convergence curve of the LNSGA-Ⅱ algorithm in an optimization process of an embodiment of the present invention, specifically a convergence curve based on Cmax as a target;
[0068] Fig.12 It is a convergence curve of the LNSGA-II algorithm in an optimization process of an embodiment of the present invention, specifically a convergence curve based on energy consumption as a target. DETAILED DESCRIPTION
[0069] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and cannot be understood as limiting the present invention.
[0070] like Figure 1-2 As shown, the multi-processing module wafer manufacturing scheduling method based on the improved NSGA-Ⅱ algorithm includes the following steps:
[0071] S100, based on the production scenario, production process and problem characteristics of multi-processing module wafer manufacturing scheduling problem, a mathematical model with maximum completion time and total energy consumption of the workshop as optimization objectives was constructed;
[0072] S200, encoding the chromosome using a double-layer encoding method based on a workpiece processing sequence and a machine selection sequence, wherein the lengths of the workpiece processing sequence and the machine selection sequence are the same as the number of workpieces;
[0073] S300, based on the population initialization of the heuristic pheromone, the processing time and power of each workpiece on the machine are different, and the workpiece needs to be preferentially assigned to the machine with the lowest processing energy consumption for processing; first calculate the average processing energy consumption of each workpiece, and then assign the workpiece to the machine with the lowest processing energy consumption for processing;
[0074] S400, crossover operation, introduce uniform crossover operator in the crossover stage, and optimize the uniform crossover operator with reference to the sequential crossover operator on this basis, and propose the sequential uniform crossover operator; use the sequential uniform crossover operator to operate the order processing sequence of chromosomes, and use the uniform crossover operator to operate the machine processing sequence; the parent generation of the crossover is selected by the tournament selection method, and the crossover probability P is set. c ;
[0075] S500, mutation operation. In order to avoid falling into the local optimum during the iteration process and generate a solution that jumps out of the local optimum, it is necessary to use a random exchange operator to exchange the genes at two positions on the order processing sequence / machine processing sequence layer of each parent chromosome for mutation, and set the mutation probability to P.m ;
[0076] S600, in order to speed up the optimization of the overall algorithm, a local search operator guided by problem characteristics is designed to minimize the maximum completion time and total energy consumption;
[0077] S700, decoding operation, adopts the scheduling strategy of shortest processing time priority; when the total power of the workshop exceeds the set maximum instantaneous power threshold, first calculate the excess power value, that is, the difference between the current total power of the workshop and the maximum instantaneous power; then find all the processing machine sets that can meet or exceed this excess power requirement, and then adjust the machine set; these sets may contain only a single machine or may cover multiple machines, but the number of machines in the set of multiple machines is strictly controlled within n-1, where n represents the total number of machines currently in operation; after decoding, output the specific indicator value;
[0078] S800, improved elite retention strategy, modifies the multi-objective genetic algorithm, sets parameters, then directly adds N×α individuals to the new population, and finally, randomly selects the remaining N×(1-α) individuals from the suboptimal plane.
[0079] In addition, in step S100, in order to describe the problem and accurately establish the corresponding mathematical model, the following definitions are made:
[0080] At the same time, any machine can only process one workpiece;
[0081] Each workpiece can only be processed on one machine, and the machine cannot be preempted during processing;
[0082] The first processing position of each machine can be processed without preparation.
[0083] Failure of the processing machine or changes due to external factors are not considered;
[0084] In the process of solving the wafer manufacturing workshop scheduling problem, the objective function is needed to evaluate the advantages and disadvantages of various scheduling schemes; the optimization goal is to minimize the completion time, minimize the total energy consumption and maximize the finished product qualification rate; the following lists two objective functions used in solving the wafer manufacturing workshop scheduling problem;
[0085] (1) The maximum completion time is the smallest, which can be expressed as:
[0086]
[0087] Formula (2.1) treats the completion time of all workpieces as a numerical set, selects the maximum value in the set, and then the goal is to minimize this maximum value - the actual meaning is that a batch of tasks is completed in the shortest possible time.
[0088] (2) The total energy consumption of the workshop is the smallest. The total energy consumption of the workshop can be expressed as:
[0089]
[0090] Formula (2.2) shows that different machines, states, and workpieces have their own unit energy consumption, and then the energy consumption is accumulated based on the unit energy consumption to get the total energy consumption.
[0091] Among them, Figure 2-3 As shown, step S200 is specifically as follows: the first layer of the chromosome is the order processing sequence, which can be expressed as {OS1, OS2, ..., OS i ,OS i+1 ,…,OS n}, where OS i ∈{1,2,…,n}, used to represent the serial number of the i-th processed workpiece;
[0092] The second level of chromosomes is the machine-processed sequence, which can be represented as {MS1, MS2, …, MS i ,MS i+1 ,…,MS n}, where MS i ∈{1,2,…,m}, used to represent the order OS i The assigned processing machine serial number;
[0093] For the processing sequence of workpieces assigned to the same machine, first arrange the corresponding position numbers of these orders in the order processing sequence in ascending order, and then determine the processing sequence of these orders on the current machine based on their positions in the sorted list.
[0094] Wherein, in step S300:
[0095] The average operating energy consumption of the workpiece is calculated using formula (4.1);
[0096]
[0097] Formula (4.1) means that the energy consumption of the machine when it is in the processing state is calculated by multiplying the processing time by the unit energy consumption, then accumulating the result and dividing it by the quantity.
[0098] S311: First, the average processing energy consumption of each workpiece is calculated according to formula (4.1), and then the workpieces are sorted in the order of increasing energy consumption, so as to obtain an orderly workpiece processing queue;
[0099] S312: allocating machines to workpieces waiting for processing according to the principle of starting processing the workpieces earliest; if there are multiple machines available for processing the workpiece at the same time, the workpiece is allocated to the machine with the lowest processing energy consumption for processing;
[0100] S313: When the workpiece is machined, the machine processing start time and end time are updated in real time;
[0101] S314: Repeat the above steps until all workpieces are processed.
[0102] Among them, Figure 6 As shown, step S400 is specifically as follows:
[0103] This paper uses the OUX crossover operator to operate the order processing sequence layer of the chromosome, and uses the UX crossover method to operate the machine processing sequence layer of the chromosome; using P1 OS / P1 MS and P2 OS / P2 MS Represents the two parent generations of the order processing sequence / machine processing sequence, using O1 OS / O1 MS and O2 OS / O2 MS Two children representing order processing sequence / machine processing sequence respectively;
[0104] The main steps of the two crossover operators are described as follows:
[0105] Description of OUX crossover operator:
[0106] S411: First, randomly generate a random number N in the interval [0.2*Popsize, 0.5*Popsize] c , and then randomly select N c The genes at position J constitute subset set ;
[0107] S412: From P1 OS / P2 OS Copy and J set The genes corresponding to the order in the O1OS / O2OS are copied in the same order and position.
[0108] S413: P1 OS / P2 OS Never in O2 OS / O1 OS The genes that appeared above are placed in O2 according to their original order. OS / O1 OS Free space on the
[0109] Description of UX crossover operator:
[0110] S421: First, randomly generate a random number N in the interval [0.2*Popsize, 0.5*Popsize] c , and then randomly select genes at Nc positions to form a subset M set ;
[0111] S422: From P1 MS / P2 MS Copy and M set The genes corresponding to the order in O1 are copied to O1 in the same order and position. MS / O2 MS superior;
[0112] S423: P1 MS / P2 MS The remaining genes are copied to O2 in the original order and position. MS / O1 MS superior.
[0113] In addition, if Figure 7 As shown, step S500 is specifically as follows: the mutated parent adopts the individuals in the population after crossover, and the mutation probability is P m ; Use POS / PMS to represent the parent of OS / MS, and use OOS / OMS to represent the child of OS / MS.
[0114] In addition, in step S600: for the same processing plan, since the scheduling sequence is known, the order on the machine corresponding to the maximum completion time can be selected and moved to other machines for processing;
[0115] Since not every search can obtain a new solution that dominates the original scheduling solution, the following rules are set to constrain it:
[0116] Rule 1: To reduce the computational complexity of local search, set the search to a maximum of 5 times;
[0117] Rule 2: In order to avoid falling into the local optimum, set a 25% probability of accepting a solution that does not dominate the original solution in each search; that is, within 5 searches, if a new solution that dominates the objective function value of the original scheduling solution is found, then the new solution is set as the new offspring; otherwise, the new solution population after 5 searches is sorted by non-domination, congestion calculation, and population sorting, and the new solution with sequence number 1 in the new population is accepted as the new offspring with a probability of 25%;
[0118] Rule 3: To avoid local search taking too long, only individuals in the first frontier are set to participate in local search; if the number of individuals in the first frontier is greater than Searchsize, the first Searchsize individuals are selected for local search. Searchsize is a variable parameter value, and it is also a judgment parameter for local search to be enabled: when the number of individuals in the first frontier is greater than this value, a specified number of individuals are selected from the individuals in the first frontier for local search operation.
[0119] Simulation experiment, this paper first uses the LPP heuristic rule improved NSGA-Ⅱ algorithm (LPP Non-dominated Sorting Genetic Algorithm II, LNSGA-Ⅱ) to compare the optimization performance with the traditional multi-objective genetic algorithm (Multi-Objective Genetic Algorithm, MOGA), non-dominated sorting genetic algorithm (Non-dominated Sorting Genetic Algorithms, NSGA) and non-dominated sorting genetic algorithm II (Non-dominated Sorting Genetic Algorithm II, NSGA-Ⅱ), and then according to the designed examples, a large number of simulation experiments are carried out to explore the impact of different influencing factors on the algorithm performance. The experiment shows that the improved NSGA-Ⅱ algorithm is effective in solving the dual-objective model of maximum completion time and energy consumption, and can better achieve the expected goals. The simulation experiment is analyzed below.
[0120] Example generation, in order to verify the effectiveness of the algorithm in all aspects, this section uses the idea of split-plot to design simulation experiments. Combined with relevant literature and the application scenarios of this problem, the factors that affect the performance of the algorithm are mainly: the number of machines, the number of workpieces and the proportion of processing time in the process stage. Among them, the number of machines m∈{4,6,12}; the number of workpieces n∈{50,100,200}; the proportion of processing time in the process stage μ∈{0.7,0.5,0.3}. For the four parameters of the number of iterations, population size, crossover probability, and mutation probability of the population, the Taguchi method is used to set five levels for each parameter. The orthogonal table is shown in Table 3-1. Minitab software is used to conduct a multi-factor orthogonal experiment with IGD as the evaluation index, and the mean main effect diagram is obtained as shown in the following figure. Figure 3-7 As shown in the figure, according to the experiment, the crossover rate, mutation rate, population size, and number of iterations are 0.9, 0.25, 100, and 100 respectively.
[0121] Comparison of algorithm effects: According to the example design, a total of 27 (3*3*3) example problems were generated, and then 108 examples were generated based on the comparison of the four multi-objective intelligent algorithms in this paper. In order to minimize the influence of random factors, each example was repeated 10 times, that is, a total of 1080 experiments, and the results of each simulation run were recorded. To verify the effectiveness of the algorithm proposed in this chapter, the programming software used in this paper to solve the wafer manufacturing workshop scheduling problem is Matlab, version Matlab R2017a, and the program is run on a personal computer with a processor of AMD Ryzen 74800U@1.80GHz, an operating system of win10, and a running memory of 16GB.
[0122] As can be seen in Table 1, in the medium and high levels of the processing time proportion of the process stage, the LNSGA-Ⅱ algorithm tends to reduce the Cmax value during the optimization process, which may lead to relatively high energy consumption values. By comparing the Cmax values, it can be clearly seen that the LNSGA-Ⅱ algorithm performs better than the other three algorithms. In addition, the comprehensive comparison of Cmax and energy consumption results shows that when dealing with small-scale problems, the solutions obtained by the LNSGA-Ⅱ algorithm are better than the other three algorithms. For medium and large-scale problems, the LNSGA-Ⅱ algorithm performs more prominently in reducing the Cmax value. Overall, the LNSGA-Ⅱ algorithm can not only obtain better solutions than the other three algorithms, but also ensure the maximum completion time of the first optimization objective while achieving a good value for the other optimization objective. Therefore, it can be concluded that the LNSGA-Ⅱ algorithm has more advantages than other algorithms. It can effectively balance the global search capability and local search to obtain high-quality solutions.
[0123] Through comparative experiments, it can be clearly seen that the LNSGA-Ⅱ algorithm can obtain better results than the other three algorithms. This is mainly due to the improved strategy for the initial population: by adopting the LPP and EST heuristic algorithms, the machine with the lowest processing energy consumption is selected for each workpiece, and the workpiece allocation on the machine with the maximum completion time is optimized. These improvements enable the LNSGA-Ⅱ algorithm to show its significant advantages at the beginning of the iteration. By improving the elite retention strategy, it is ensured that the better workpiece allocation fragments can be retained in the optimal solution after each generation of iteration. At the same time, combined with the local search strategy for continuous adjustment and optimization, it not only increases the possibility of finding the global optimal solution, but also helps the algorithm to jump out of the local optimal solution, so that the final optimization result is more satisfactory.
[0124] Fig.10 This is the Pareto frontier found by LNSGA-Ⅱ. It can be seen from the figure that these non-dominated individuals are relatively evenly distributed in space. Figure 11-Figure 12This is the convergence curve of the LNSGA-Ⅱ algorithm in the optimization process. From the figure, we can see that the algorithm converges quickly in the early stage of iteration, and gradually finds the approximate optimal solution as the number of iterations increases. Finally, although the average value of the population changes slightly, the optimal individual remains stable and no longer changes.
[0125] Table 1 Comparison of algorithm results
[0126]
[0127] Table 1 Comparison of algorithm results
[0128]
[0129] The technical principle of the present invention is described above in conjunction with specific embodiments. These descriptions are only for explaining the principle of the present invention and cannot be interpreted as limiting the scope of protection of the present invention in any way. Based on the explanations herein, those skilled in the art can associate other specific implementations of the present invention without paying creative labor, and these methods will fall within the scope of protection of the present invention.
Claims
1. A multi-processing module wafer manufacturing scheduling method based on the improved NSGA-Ⅱ algorithm is characterized by: The steps include: S100: Based on the production scenarios, production processes, and problem characteristics of the multi-processing module wafer manufacturing scheduling problem, a mathematical model with maximum completion time and total workshop energy consumption as optimization objectives was constructed; S200, encoding the chromosome based on a double-layer encoding method of a workpiece processing sequence and a machine selection sequence, wherein the lengths of the workpiece processing sequence and the machine selection sequence are the same as the number of workpieces; S300: Based on the population initialization of the heuristic pheromone, the processing time and power of each workpiece on the machine vary, and the workpieces need to be preferentially assigned to the machine with the lowest processing energy consumption for processing. The average processing energy consumption of each workpiece is first calculated, and then the workpieces are assigned to the machine with the lowest processing energy consumption for processing. S400, crossover operation, introduces the uniform crossover operator in the crossover stage, and on this basis, optimizes the uniform crossover operator with reference to the sequential crossover operator, and proposes the sequential uniform crossover operator; the sequential uniform crossover operator is used to operate the order processing sequence of the chromosome, and the uniform crossover operator is used to operate the machine processing sequence; the parent of the crossover is selected using the tournament selection method, and the crossover probability P is set. c ; S500, mutation operation. In order to avoid falling into the local optimum during the iteration process and generate a solution that jumps out of the local optimum, it is necessary to use a random exchange operator to exchange the genes at two positions on the order processing sequence / machine processing sequence layer of each parent chromosome for mutation, and set the mutation probability to P. m ; S600, in order to speed up the optimization of the overall algorithm, a local search operator guided by problem characteristics is designed to minimize the maximum completion time and total energy consumption; S700, decoding operation, adopts a scheduling strategy that prioritizes the shortest processing time. When the total power of the workshop exceeds the set maximum instantaneous power threshold, the excess power value is first calculated, that is, the difference between the current total power of the workshop and the maximum instantaneous power. Then, all processing machine sets that can meet or exceed this excess power requirement are found, and the machine set can be adjusted. These sets may include only a single machine or multiple machines, but the number of machines in a set of multiple machines is strictly controlled within n-1, where n represents the total number of machines currently in operation. S800, improved elite retention strategy, modifies the multi-objective genetic algorithm, sets parameters, then directly adds N×α individuals to the new population, and finally, randomly selects the remaining N×(1-α) individuals from the suboptimal plane, and outputs specific indicator values after decoding.
2. The multi-processing module wafer manufacturing scheduling method based on the improved NSGA-II algorithm according to claim 1 is characterized in that: In step S100: In order to describe the problem and accurately establish the corresponding mathematical model, the following definitions are made: At the same time, any machine can only process one workpiece; Each workpiece can only be processed on one machine, and the machine cannot be preempted during processing; The first processing position of each machine can start processing without preparation; Failure of the processing machine or changes due to external factors are not taken into account; When solving wafer fabrication workshop scheduling problems, an objective function is needed to evaluate the pros and cons of various scheduling schemes. The optimization objectives are to minimize completion time, minimize total energy consumption, and maximize finished product yield. (1) The maximum completion time is the smallest, which can be expressed as: (2) The total energy consumption of the workshop is the smallest. The total energy consumption of the workshop can be expressed as: Formula (2.1) and Formula (2.2) are the two objective functions used to solve the wafer manufacturing workshop scheduling problem.
3. The multi-processing module wafer manufacturing scheduling method based on the improved NSGA-II algorithm according to claim 1 is characterized in that: Step S200 is specifically as follows: The first layer of chromosome is the order processing sequence, which can be expressed as {OS1, OS2, ..., OS i ,OS i+1 ,…,OS n }, where OS i ∈{1,2,…,n}, used to represent the serial number of the i-th processed workpiece; The second layer of chromosomes is the machine-processed sequence, which can be represented as {MS1, MS2,…, MS i ,MS i+1 ,…,MS n }, where MS i ∈{1,2,…,m}, used to represent the order OS i The assigned processing machine serial number; For the processing sequence of workpieces assigned to the same machine, first arrange the corresponding position numbers of these orders in the order processing sequence in ascending order, and then determine the processing order of these orders on the current machine based on their positions in the sorted list.
4. The multi-processing module wafer manufacturing scheduling method based on the improved NSGA-II algorithm according to claim 1 is characterized in that: In step S300: The average operating energy consumption of the workpiece is calculated using formula (4.1); S311: First, calculate the average processing energy consumption of each workpiece according to formula (4.1), and then sort the workpieces in the order of increasing energy consumption to obtain an ordered workpiece processing queue; S312: Allocate a machine to a workpiece waiting to be processed based on the principle of starting processing the workpiece earliest; if multiple machines are available to process the workpiece at the same time, allocate the workpiece to the machine with the lowest processing energy consumption for processing; S313: When machining the workpiece, the machine processing start time and end time are updated in real time; S314: Repeat the above steps until all workpieces are processed.
5. The multi-processing module wafer manufacturing scheduling method based on the improved NSGA-II algorithm according to claim 1 is characterized in that: Step S400 is specifically as follows: This paper uses the OUX crossover operator to operate the order processing sequence layer of the chromosome, and uses the UX crossover method to operate the machine processing sequence layer of the chromosome; using P1 OS / P1 MS and P2 OS / P2 MS Represents the two parents of order processing sequence / machine processing sequence, using O1 OS / O1 MS and O2 OS / O2 MS Two children representing order processing sequence / machine processing sequence respectively; The main steps of the two crossover operators are described as follows: Description of OUX crossover operator: S411: First, randomly generate a random number N in the range [0.2*Popsize, 0.5*Popsize] c , and then randomly select N c The genes at position J constitute subset set ; S412: From P1 OS / P2 OS Copy and J set The genes corresponding to the order in the O1OS / O2OS were copied in the same order and position. S413: P1 OS / P2 OS Never in O2 OS / O1 OS The genes that appeared above are placed in O2 according to their original order. OS / O1 OS Free space on the Description of the UX cross operator: S421: First, randomly generate a random number N in the range [0.2*Popsize, 0.5*Popsize] c , and then randomly select genes at Nc positions to form a subset M set ; S422: From P1 MS / P2 MS Copy and M set The genes corresponding to the order in O1 are copied to O1 in the same order and position. MS / O2 MS superior; S423: P1 MS / P2 MS The remaining genes are copied to O2 in the original order and position. MS / O1 MS superior.
6. The multi-processing module wafer manufacturing scheduling method based on the improved NSGA-II algorithm according to claim 1 is characterized in that: Step S500 is specifically as follows: The mutated parent adopts the individuals in the population after crossover, and the mutation probability is P m ; Use POS / PMS to represent the parent generation of OS / MS, and use OOS / OMS to represent the child generation of OS / MS.
7. The multi-processing module wafer manufacturing scheduling method based on the improved NSGA-II algorithm according to claim 1 is characterized in that: In step S600: For the same processing plan, since the scheduling sequence is known, we can select the order on the machine corresponding to the maximum completion time and move it to other machines for processing; Since not every search can yield a new solution that dominates the original scheduling solution, the following rules are set to constrain it: Rule 1: To reduce the computational complexity of local search, set the search to a maximum of 5 times; Rule 2: To avoid falling into a local optimum, set a 25% probability of accepting a solution that does not dominate the original solution in each search. That is, within 5 searches, if a new solution is found that dominates the objective function value of the original scheduling solution, then the new solution is set as the new offspring. Otherwise, the new solution population after 5 searches is sorted by non-domination, congestion calculation, and population sorting, and the new solution with sequence number 1 in the new population is selected as the new offspring with a 25% probability. Rule 3: To avoid local search taking too long, only individuals on the first frontier are allowed to participate in local search; if the number of individuals on the first frontier is greater than Searchsize, the first Searchsize individuals are selected for local search.
Citation Information
Patent Citations
Method for solving shop energy-saving scheduling problem based on improved NSGA-II algorithm
CN110543151A
Printing workshop multi-target scheduling optimization method based on improved NSGA-II
CN118246679A
Multi-objective optimization method and system for master production plan of casting parallel workshops
US20210373888A1
Cited By
Multi-station synchronous machining control method and system
CN120669641A
Automatic suspension conveying system for bicycle frames
CN121168732A
