A production scheduling method and storage medium based on optimized self-tuning genetic algorithm

Abstract

The present invention relates to the field of manufacturing, and discloses a production scheduling method and storage medium based on an optimized self-tuning genetic algorithm. The population in the genetic algorithm is initialized by a pre-optimization method, and the parameters in the genetic algorithm are adjusted by machine learning. Compared with the traditional method of scheduling production using a legacy algorithm, the method in the present invention has a faster convergence speed and can reduce unnecessary iterations, so that the sub-assembly line plan can be quickly adjusted. It solves the problem that the traditional genetic algorithm has a large number of iterations and a slow convergence speed, resulting in the sub-assembly production scheduling plan not being able to adapt well to the main line production situation.

Description

Technical Field

[0001] The present invention relates to the field of production and manufacturing, and in particular to a production scheduling method and a storage medium based on an optimized self-tuning genetic algorithm. Background Art

[0002] In the field of manufacturing, for some products with more complex processes, the production line generally consists of a main line and multiple sub-assembly lines. The main line is responsible for the production of the entire product, and the sub-assembly line is responsible for the production of certain parts. The parts produced by the sub-assembly line will go to the closing station of the main line and be installed on the product. There will be a smaller line-side warehouse next to the closing station, which is used to temporarily store a small number of sub-assembly parts to deal with some emergencies. For example, in automobile production, the body of the car is produced on the main line, and the door is produced on the sub-assembly line. After the door is produced, it will flow to the closing station of the main line and be installed on the body. In order to save production costs, modern workshops often adopt flexible production methods for production, that is, using the same production line to produce different products. In this way, since the main line will produce different products, the sub-assembly line must also select appropriate parts to produce according to the production situation on the main line to ensure that each product on the main line has corresponding parts supply. At the same time, since the products on the main line need to be inspected from time to time, and scrapped products may be taken off the main line, the production queue of the main line changes in real time, which requires that the production queue of the sub-assembly line can be adjusted in real time according to the situation of the main line.

[0003] For this production scheduling problem, the present invention provides a production scheduling method based on an optimized self-tuning genetic algorithm to perform production scheduling. Summary of the invention

[0004] In order to solve the above technical problems, the present invention provides a production scheduling method and storage medium based on an optimized self-tuning genetic algorithm.

[0005] In order to solve the above technical problems, the present invention adopts the following technical solutions:

[0006] A production scheduling method based on an optimized self-tuning genetic algorithm is proposed. The production scheduling problem of the assembly line is solved according to the change of the main line production queue. The optimization process of the self-tuning genetic algorithm includes the following steps:

[0007] Step 11: Select a reference station on the main line before the closing station, obtain all the production queues Q on the main line before the reference station, and determine the adaptability function Where diff(x[j], Q[j]) is the degree of type matching between the jth part x[j] in the assembly line production plan and the jth part Q[j] in the main line production queue. If the type of x[j] is the same as the type of Q[j], then diff(x[j], Q[j]) = 0; if the type of x[j] is different from the type of Q[j], then diff(x[j], Q[j]) = 1; EMS(x[j]) is the expected quantity of line inventory after the production of part x[j], and E is the ideal quantity of line inventory; cost(x[j-1], x[j]) is the cost from producing part x[j-1] to producing part x[j]; a, b, and c are penalty variables;

[0008] Step 12: Determine the initial population of the genetic algorithm by a pre-optimization method, and input the initial population into the genetic algorithm to obtain a test production schedule for the assembly line;

[0009] Step 13: Using historical data, repeat step 12, use different mutation rates and population individual numbers for the same production scheduling problem, and calculate the performance score of the genetic algorithm through the convergence time of the genetic algorithm and the fitness function value of the test production scheduling plan output by the genetic algorithm;

[0010] Step 14: Record the mutation rate, the number of individuals in the population, and the characteristics of the production scheduling problem used by the genetic algorithm with the highest performance score in each production scheduling problem to obtain training data; the characteristics of the production scheduling problem include the length of the production queue Q, the types of products included in the production queue Q, and the set values ​​of a, b, and c in the fitness function;

[0011] Step 15: Train the fully connected neural network through the training data; the input of the fully connected neural network is the characteristics of the production scheduling problem, and the output is the mutation rate and the number of individuals in the population;

[0012] Scheduling methods include:

[0013] Step 21: Input the characteristics of the actual production scheduling problem of the splitting and splicing line into the trained fully connected neural network to obtain the mutation rate and the number of populations;

[0014] Step 22: Input the mutation rate and number of individuals in the population obtained in step 21 into the genetic algorithm to obtain the actual production schedule of the assembly line.

[0015] Specifically, in step 13, when the performance score of the genetic algorithm is obtained by the convergence time of the genetic algorithm and the fitness function value of the output test production schedule, the convergence time of the genetic algorithm and the fitness function value of the output test production schedule are weightedly summed to obtain the performance score.

[0016] Specifically, the reference station selected in step 11 satisfies: the time T for the part on the main line to reach the closing station from the reference station is greater than the time t for the sub-assembly line to produce a part, that is, T>t; n is the number of processes required to produce a part on the assembly line, t sub(k) is the time required for process k, t prep It is the time it takes for the assembly line to prepare materials for the next round of production.

[0017] Specifically, in step 12, when the initial population of the genetic algorithm is determined by a pre-optimization method, the following steps are included:

[0018] Determine the total number of parts to be produced on the sub-assembly line based on the total number of parts produced on the main line;

[0019] Starting from the first part of the assembly line, the production type of the parts is determined in sequence, and the sequence of the obtained part production types is used as the first individual of the primary population; when determining the production type of the a-th part, the fitness function value A before the production of the a-th part is first calculated, and then the fitness function values ​​B of various production types after the production of the a-th part are calculated respectively, and the production type with the largest BA is selected as the production type of the a-th part;

[0020] Then, based on the first individual, the remaining individuals of the initial population are generated through mutation.

[0021] A computer-readable storage medium stores a computer program, which implements the production scheduling method when executed by a processor.

[0022] Compared with the prior art, the beneficial technical effects of the present invention are:

[0023] The production scheduling method in the present invention improves the traditional legacy algorithm, initializes the population in the genetic algorithm through a pre-optimization method, and adjusts the parameters in the genetic algorithm through machine learning. Compared with the traditional method of scheduling using the legacy algorithm, the method in the present invention has a faster convergence speed and can reduce unnecessary iterations, so that the sub-assembly line plan can be adjusted quickly, solving the problem that the traditional genetic algorithm has a large number of iterations and a slow convergence speed, resulting in the sub-assembly production scheduling plan not being able to adapt well to the main line production situation. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 It is a schematic diagram of the process of the production scheduling method of the present invention. DETAILED DESCRIPTION

[0025] A preferred embodiment of the present invention is described in detail below with reference to the accompanying drawings.

[0026] In the present invention, the upstream direction of the main line and the sub-joining line is the front, and the downstream direction is the rear.

[0027] For ease of understanding, the relevant terms in genetic algorithms are introduced as follows:

[0028] Genetic algorithm: Genetic algorithm is a heuristic algorithm designed based on the process of species evolution in nature. In this algorithm, each possible solution is an individual. Individuals evolve by combining with each other to obtain new individuals and mutating. After completing a specified number of evolutions, the best individual is selected as the final solution to the problem.

[0029] Individual: The individual in the present invention refers to each possible assembly line production schedule, represented by x, where x is a sequence containing all parts production types. For example, if 4 A parts and 4 B parts are to be produced, then x=AAAABBBBB; x[j] represents the jth part in the sequence x.

[0030] Population: A population is the set of all individuals; population S = {x1, x2, ..., x N}, N is the number of individuals in the population, which is a parameter that needs to be set, x i is the i-th individual in the population.

[0031] Fitness function: The fitness function is represented by f(x), which is a function used to measure the quality of a solution. The better a solution is, the larger its fitness function value is.

[0032] Generation: Generation is similar to the generation in biological organisms. Each time individuals are combined with others, the new individuals obtained are the next generation of individuals.

[0033] Mutation: In each generation, the new individual obtained by merging two individuals may mutate. For example, if individual x1=AAAA and x2=BBBB, the first half of x1 and the second half of x2 are taken to obtain the new individual x3=AABB. When x3 is combined with other individuals to generate the next generation of individuals, there is a certain probability that any element will change, such as the third element changing from B to A. Genetic algorithms control the probability of an individual mutating through the mutation rate.

[0034] This embodiment takes automobile production as an example. The car body is produced on the main line and the car doors are produced on the sub-assembly line. A car has a car body and multiple doors. These multiple doors are called a group of doors. Considering the order matching problem, the number of parts produced on the main line is consistent with the number of production groups on the sub-assembly line.

[0035] like Figure 1 As shown, the scheduling method based on the optimized self-tuning genetic algorithm in the present invention formulates a scheduling plan for the scheduling problem of the assembly line according to the changes in the main line production queue, wherein the optimization process of the self-tuning genetic algorithm includes the following steps:

[0036] S11: Select a reference station on the main line before the closing station, obtain the entire production queue Q of the main line before the reference station, and determine the adaptability function

[0037] Wherein diff(x[j], Q[j]) is the degree of type matching between the jth part x[j] in the assembly line production schedule and the jth part Q[j] in the main line production queue. If the type of x[j] is the same as the type of Q[j], then diff(x[j], Q[j]) = 0; if the type of x[j] is different from the type of Q[j], then diff(x[j], Q[j]) = 1; EMS(x[j]) is the expected quantity of line inventory after the production of part x[j], and E is the ideal quantity of line inventory; cost(x[j-1], x[j]) is the cost from producing part x[j-1] to producing part x[j], such as the loss of production time and additional material costs caused by switching the type of parts produced. The specific contents of the cost can be selected according to the actual production situation of the workshop.

[0038] a, b, and c are three penalty variables with values ​​between -1, 0. Their specific values ​​are determined according to the production scheduling tendencies of different production lines. For example, if you want the production queue of the assembly line to be as consistent as possible with the main line, you can make the value of a closer to -1 than b (with a larger penalty for inconsistent sequences).

[0039] The reference station selected in step 11 satisfies: the time T for the part on the main line to reach the closing station from the reference station is greater than the time t for the sub-assembly line to produce one part, that is, T>t; n is the number of production processes of the assembly line, t sub(k) is the time required for production process k, t prep It is the time it takes for the assembly line to prepare materials for the next round of production, excluding the time it takes to switch the type of parts being produced.

[0040] S12: Determine the initial population of the genetic algorithm through a pre-optimization method, and input the initial population into the genetic algorithm to obtain a test production schedule for the assembly line.

[0041] Specifically, the method for determining the initial population of the genetic algorithm by a pre-optimization method comprises the following steps:

[0042] Determine the total number of parts to be produced on the sub-assembly line based on the total number of parts produced on the main line;

[0043] Starting from the first part of the assembly line, the production type of the parts is determined in sequence, and the sequence of the obtained part production types is used as the first individual of the primary population; when determining the production type of the a-th part, the fitness function value A before the production of the a-th part is first calculated, and then the fitness function values ​​B of various production types after the production of the a-th part are calculated respectively, and the production type with the largest BA is selected as the production type of the a-th part;

[0044] Then, based on the first individual, the remaining individuals of the initial population are generated through mutation.

[0045] The genetic algorithm itself uses a completely random method to produce the first generation population. The individuals in the first generation population are very different from the optimal solution. The first individual x1 obtained by the above method is already a better solution. Other individuals are obtained by mutation based on x1, which can also ensure the diversity of solutions. Compared with ordinary genetic algorithms, it can improve the convergence speed and reduce the solution time.

[0046] S13: Using historical data, repeat step 12, use different mutation rates and population individual numbers for the same production scheduling problem, and calculate the performance score of the genetic algorithm through the convergence time of the genetic algorithm and the fitness function value of the test production scheduling plan output by the genetic algorithm.

[0047] Specifically, according to the actual production situation, the convergence time of the genetic algorithm and the adaptability function value of the output test production scheduling plan are weighted and summed to obtain the performance score.

[0048] S14: Record the mutation rate, the number of individuals in the population, and the characteristics of the production scheduling problem used by the genetic algorithm with the highest performance score in each production scheduling problem to obtain training data; the characteristics of the production scheduling problem include the length of the production queue Q, the types of products included in the production queue Q, and the set values ​​of a, b, and c in the fitness function.

[0049] S15: Train a fully connected neural network (MLP) through training data; the input of the fully connected neural network is the characteristics of the production scheduling problem, and the output is the mutation rate and the number of individuals in the population.

[0050] Scheduling methods include:

[0051] S21: Input the characteristics of the actual production scheduling problem of the split-and-assembled line into the trained fully connected neural network to obtain the mutation rate and the number of populations.

[0052] S22: Input the mutation rate and number of individuals in the population obtained in S21 into the genetic algorithm, and combine the actual production information and process requirements (such as the production queue Q of the main line, the expected quantity of line-side inventory, and the ideal quantity of line-side inventory E) to obtain the actual production schedule of the assembly line through the genetic algorithm.

[0053] A computer-readable storage medium stores a computer program, which implements the production scheduling method when executed by a processor.

[0054] It is obvious to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential features of the present invention. Therefore, from any point of view, the embodiments should be regarded as exemplary and non-limiting, and the scope of the present invention is defined by the appended claims rather than the above description, and it is intended that all changes falling within the meaning and scope of the equivalent elements of the claims are included in the present invention, and any reference numerals in the claims should not be regarded as limiting the claims involved.

[0055] In addition, it should be understood that although the present specification is described in terms of implementation modes, not every implementation mode includes only one independent technical solution. This narrative method of the specification is only for the sake of clarity. Those skilled in the art should regard the specification as a whole. The technical solutions in each embodiment may also be appropriately combined to form other implementation modes that can be understood by those skilled in the art.

Claims

1. A production scheduling method based on an optimized self-tuning genetic algorithm, which formulates a production scheduling plan for the production scheduling problem of the assembly line according to the changes in the main line production queue, and is characterized by: The optimization process of the self-tuning genetic algorithm includes the following steps: Step 11: Select a reference station on the main line before the closing station, obtain all the production queues Q on the main line before the reference station, and determine the adaptability function Where diff(x[j],Q[j]) is the degree of type matching between the jth part x[j] in the assembly line production plan and the jth part Q[j] in the main line production queue. If the type of x[j] is the same as the type of Q[j], then diff(x[j],Q[j]) = 0; if the type of x[j] is different from the type of Q[j], then diff(x[j],Q[j]) = 1; EMS(x[j]) is the expected quantity of line inventory after the production of part x[j], and E is the ideal quantity of line inventory; cost(x[j-1],x[j]) is the cost from producing part x[j-1] to producing part x[j]; a, b, and c are penalty variables; Step 12: Determine the initial population of the genetic algorithm by a pre-optimization method, and input the initial population into the genetic algorithm to obtain a test production schedule for the assembly line; Step 13: Using historical data, repeat step 12, use different mutation rates and population individual numbers for the same production scheduling problem, and calculate the performance score of the genetic algorithm through the convergence time of the genetic algorithm and the fitness function value of the test production scheduling plan output by the genetic algorithm; Step 14: Record the mutation rate, the number of individuals in the population, and the characteristics of the production scheduling problem used by the genetic algorithm with the highest performance score in each production scheduling problem to obtain training data; the characteristics of the production scheduling problem include the length of the production queue Q, the types of products included in the production queue Q, and the set values ​​of a, b, and c in the fitness function; Step 15: train a fully connected neural network using training data; the input of the fully connected neural network is the characteristics of the production scheduling problem, and the output is the mutation rate and the number of individuals in the population; Scheduling methods include: Step 21: Input the characteristics of the actual production scheduling problem of the splitting and splicing line into the trained fully connected neural network to obtain the mutation rate and the number of populations; Step 22: Input the mutation rate and the number of individuals in the population obtained in step 21 into the genetic algorithm to obtain the actual production schedule of the assembly line; The reference station selected in step 11 satisfies: the time T for the part on the main line to reach the closing station from the reference station is greater than the time t for the sub-assembly line to produce a part, that is, T>t; n is the number of processes required to produce a part on the assembly line, t sub(k) is the time required for process k, t prep It is the time for the assembly line to prepare materials for the next round of production; In step 12, when the initial population of the genetic algorithm is determined by a pre-optimization method, the following steps are included: Determine the total number of parts to be produced on the sub-assembly line based on the total number of parts produced on the main line; Starting from the first part of the assembly line, the production type of the parts is determined in sequence, and the sequence of the obtained part production types is used as the first individual of the primary population; when determining the production type of the a-th part, the fitness function value A before the production of the a-th part is first calculated, and then the fitness function values ​​B of various production types after the production of the a-th part are calculated respectively, and the production type with the largest BA is selected as the production type of the a-th part; Then, based on the first individual, the remaining individuals of the initial population are generated through mutation.

2. The production scheduling method based on the optimized self-tuning genetic algorithm according to claim 1 is characterized in that: In step 13, when the performance score of the genetic algorithm is obtained by the convergence time of the genetic algorithm and the fitness function value of the output test production schedule, the convergence time of the genetic algorithm and the fitness function value of the output test production schedule are weightedly summed to obtain the performance score.

3. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the production scheduling method according to any one of claims 1 to 2 is implemented.

Images