A data-driven manufacturing system energy-saving buffer configuration optimization method

Abstract

This invention provides a data-driven method for optimizing the energy-saving buffer configuration of a manufacturing system. It simultaneously performs dynamic control of system production capacity and energy consumption, balancing global and local search capabilities, and significantly improves the solution quality for buffer configuration optimization. Compared to existing configuration methods, the solution quality is significantly improved. A machine learning-based surrogate model is used to replace discrete simulation software in evaluating buffer configuration candidate schemes during the local search phase, improving system throughput and energy consumption calculation efficiency. While significantly improving solution quality, the solution efficiency is comparable to existing configuration methods, achieving a solution quality improvement within a similar computation time. Furthermore, the database is updated in real time during the buffer configuration optimization process, ensuring the prediction accuracy of the surrogate model and realizing dynamic data-driven optimization.

Description

Technical Field

[0001] This invention relates to the field of production line optimization technology, and in particular to a data-driven method for optimizing the configuration of energy-saving buffers in manufacturing systems. Background Technology

[0002] Currently, manufacturing enterprises mostly use self-inspired methods for buffer design, which suffers from low computational efficiency and poor robustness in large-scale manufacturing systems. For personalized market customization and high-frequency production disturbances, it is impossible to efficiently configure buffer resources to achieve dynamic production capacity control. Furthermore, while existing buffer configuration optimization methods aim to improve system production capacity and reduce system construction costs, they cannot be used for energy efficiency control of manufacturing systems. Summary of the Invention

[0003] The purpose of this invention is to address the shortcomings of the aforementioned background technology by providing a multi-objective buffer configuration optimization method that simultaneously performs dynamic control of system production capacity and energy consumption, providing technical support for dynamic buffer configuration for personalized markets and high-frequency production disturbances, and improving the efficiency and robustness of buffer configuration optimization in manufacturing systems.

[0004] To achieve the above objectives, the present invention provides a data-driven method for optimizing the configuration of energy-saving buffers in manufacturing systems, comprising the following steps:

[0005] S1, satisfying b1+b2+…+b I-1 Given B, randomly generate N. dat Candidate solutions for manufacturing system buffer configuration

[0006] S2, establish a system simulation model and calculate system throughput and energy consumption;

[0007] S3, based on N dat Establish a database of candidate buffer configuration schemes and their corresponding system throughput and energy consumption; sort the candidate schemes in database D.

[0008] S4, randomly select N from database D. ini Each of the candidate solutions is used as an initial solution and placed into the set ∏. loc and ∏ cur ;

[0009] S5, establish the agent model;

[0010] S6, targeting ∏ loc A local domain search is performed on the candidate solutions to generate new candidate solutions;

[0011] S7 calculates the system throughput and energy consumption corresponding to the new candidate schemes generated in S6 using the proxy model;

[0012] S8 sorts the new candidate solutions generated in S6 and updates ∏. loc ;

[0013] S9, determine if the local neighborhood search stopping condition is met. If the stopping condition is met, proceed to S10; otherwise, return to S6 and proceed to S10 for ∏. loc Perform local domain search on candidate solutions;

[0014] S10, targeting ∏ loc and ∏ cur A global domain search is performed on the candidate solutions to generate new candidate solutions;

[0015] S11, calculate the system throughput and energy consumption corresponding to the new candidate scheme generated in S10;

[0016] S12, sorting the new candidate solutions generated in S10;

[0017] S13, will ∏ cur The candidate solution ranked first in the database is added to database D, and the candidate solution ranked last in database D is deleted.

[0018] S14, determine if the global neighborhood search stopping condition is met; if the stopping condition is met, proceed to S15; otherwise, return to S6, targeting ∏. loc Perform local domain search on candidate solutions;

[0019] S15, select the candidate scheme ranked first from database D as the final buffer configuration scheme.

[0020] Furthermore, in S3, candidate schemes with system throughput greater than the target throughput are sorted from low to high system energy consumption; the remaining candidate schemes are sorted from high to low system throughput.

[0021] Furthermore, the local neighborhood search method in S6 is as follows:

[0022]

[0023] Where i or k represents a randomly selected buffer, ΔB represents the change in capacity at a certain buffer position, and randint(C1, C2) represents the integers C1 or C2.

[0024] Alternatively, the local neighborhood search method in S6 is as follows:

[0025]

[0026] Where i or k represents a randomly selected buffer.

[0027] Alternatively, the local neighborhood search method in S6 is as follows:

[0028]

[0029]

[0030] Where i or k represents a randomly selected buffer.

[0031] In S6, one search method is randomly selected each time, and the process is repeated multiple times.

[0032] Furthermore, in S8, for N where the system throughput is greater than the target throughput hig There are N candidate solutions, sorted from low to high system energy consumption. hig >Ni ni Select the first N ini Update of candidate solutions ∏ loc Otherwise, select N. hi There are 10 candidate solutions. For the candidate solutions whose system throughput is less than the target throughput, they are sorted from highest to lowest system throughput, and the top N solutions are selected. ini -N hig There are 10 candidate solutions, combined with the selected N. hig Update of candidate solutions ∏ loc .

[0033] Furthermore, in S9, the stopping condition is set to the maximum allowed number of loops J. lo J loc Set to J loc =I-1, where I is the number of processing equipment in the manufacturing system.

[0034] Furthermore, in S10, each existing candidate solution is addressed sequentially. The new candidate solutions are as follows:

[0035]

[0036]

[0037] Where i is a randomly selected buffer, rand(C3, C4) represents a random number between constants C3 and C4, rand(C5, C6) represents a random number between constants C5 and C6, C7 is a constant, rand(C8, C9) represents a random number between constants C8 and C9, and b i (bes) represents the capacity of buffer i in the first-ranked candidate scheme selected by S8, b i (ran) represents the capacity of buffer i in a candidate scheme randomly selected from the candidate schemes selected by S8;

[0038] N selected for S8 ini Any one of the candidate solutions The bootstrap value is calculated using the corresponding system throughput and energy consumption, as follows:

[0039]

[0040] Among them, TH max and TH min These respectively indicate that S8 selected N. ini The maximum and minimum system throughput among the candidate schemes, TH n EC represents the system throughput corresponding to candidate solution n. max and EC min These respectively indicate that S8 selected N. ini The maximum and minimum energy consumption of the system corresponding to each candidate scheme, EC n This represents the system energy consumption corresponding to candidate scheme n;

[0041] if This generates a new candidate solution:

[0042]

[0043]

[0044] Where i is a randomly selected buffer, rand(C10, C11) represents a random number between the constants C10 and C11, and b i b represents the capacity of buffer i for scheme n among the candidate schemes selected by S8. i (ran1) and b i (ran2) represents the capacity of buffer i in two randomly selected candidate schemes from the candidate schemes selected by S8.

[0045] Furthermore, in S12, for N where the system throughput is greater than the target throughput... hi There are N candidate solutions, sorted from low to high system energy consumption. hig >Ni ni Select the first N ini Update of candidate solutions ∏ loc and ∏ cur Otherwise, select N. hig For candidate solutions where the system throughput is less than the target throughput, the top N solutions are selected and sorted from highest to lowest system throughput. ini -N hig There are 10 candidate solutions, combined with the selected N. hig Update of candidate solutions ∏loc and Π∏ cur .

[0046] Furthermore, the stopping condition in S14 is set to the maximum allowed number of loops J. glo J glo Set to 10I, where I represents the number of processing devices within the manufacturing system.

[0047] The above-described solution of the present invention has the following beneficial effects:

[0048] The data-driven energy-saving buffer configuration optimization method for manufacturing systems provided by this invention balances global and local search capabilities, significantly improving the solution quality of buffer configuration optimization in manufacturing systems. Compared with existing configuration methods, the solution quality is significantly improved. It employs a machine learning-based surrogate model to replace discrete simulation software in evaluating buffer configuration candidate schemes during the local search phase, improving system throughput and energy consumption calculation efficiency. While significantly improving solution quality, the solution efficiency remains comparable to existing configuration methods, achieving a solution quality improvement within a similar computation time. Furthermore, the database is updated in real time during the buffer configuration optimization process, ensuring the prediction accuracy of the surrogate model and realizing dynamic data-driven optimization.

[0049] Other beneficial effects of the present invention will be described in detail in the following detailed description section. Attached Figure Description

[0050] Figure 1 This is a flowchart of the steps of the present invention;

[0051] Figure 2 This is a schematic diagram of the principle of the present invention;

[0052] Figure 3 The manufacturing system of the present invention;

[0053] Figure 4 This is a schematic diagram illustrating the three local area search methods of the present invention. Detailed Implementation

[0054] The following specific examples illustrate the implementation of this disclosure. Those skilled in the art can easily understand other advantages and effects of this disclosure from the content disclosed in this specification. Obviously, the described embodiments are only a part of the embodiments of this disclosure, and not all of them. This disclosure can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of this disclosure. It should be noted that, in the absence of conflict, the following embodiments and features in the embodiments can be combined with each other. Based on the embodiments in this disclosure, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this disclosure.

[0055] It should be noted that various aspects of embodiments within the scope of the appended claims are described below. It will be apparent that the aspects described herein can be embodied in a wide variety of forms, and any particular structure and / or function described herein is merely illustrative. Based on this disclosure, those skilled in the art will understand that one aspect described herein can be implemented independently of any other aspect, and two or more of these aspects can be combined in various ways. For example, any number of aspects set forth herein can be used to implement the device and / or practice the method. Additionally, this device and / or method can be implemented using structures and / or functionalities other than one or more of the aspects set forth herein.

[0056] It should also be noted that the illustrations provided in the following embodiments are merely schematic representations of the basic concept of this disclosure. The drawings only show components relevant to this disclosure and are not drawn according to the actual number, shape, and size of the components in implementation. In actual implementation, the form, quantity, and proportion of each component can be arbitrarily changed, and the component layout may be more complex. Furthermore, specific details are provided in the following description to facilitate a thorough understanding of the examples. However, those skilled in the art will understand that the described aspects can be practiced without these specific details.

[0057] Embodiments of the present invention provide a data-driven method for optimizing the configuration of energy-saving buffers in a manufacturing system. The manufacturing system involved includes I-1 buffers and I processing equipment, such as... Figure 3 As shown (rectangles represent processing equipment, circles represent buffers). Here, 'i' represents a specific buffer or processing equipment. Each buffer has a rated capacity 'b'. i b can be stored temporarily. i Each component. Each processing equipment has a rated operating efficiency of μ. iThis indicates the number of parts that the equipment can process per unit time. Parts enter the manufacturing system from processing equipment 1, complete the processing program on each processing equipment, and then leave the system from equipment 1.

[0058] By configuring buffers of capacity B across I-1 buffers, system energy consumption is minimized while meeting the target throughput. This can be expressed by the following formula:

[0059] In b1+b2+…+b I-1 =B and the system throughput corresponding to the candidate scheme TH ≥ TH obj Under the premise of searching for the best buffer configuration candidate scheme Minimize the corresponding manufacturing system energy consumption EC. Where B represents the total system buffer capacity, TH represents the manufacturing system throughput, and TH... obj EC represents the target throughput of the manufacturing system design, and EC represents the system energy consumption.

[0060] like Figure 1 , Figure 2 As shown, the method specifically includes the following steps:

[0061] S1, satisfying b1+b2+…+b I-1 Given B, randomly generate N. dat Candidate solutions for manufacturing system buffer configuration

[0062] S2 uses discrete simulation software, such as Flexsim and Anera, to establish a system simulation model based on the manufacturing system's processing equipment and buffer configuration, and calculate the system throughput and energy consumption.

[0063] S3, based on N dat A database is established with candidate buffer configuration schemes and their corresponding system throughput and energy consumption; the candidate schemes in database D are sorted.

[0064] The sorting method is as follows: for candidate schemes whose system throughput is greater than the target throughput, they are sorted from low to high system energy consumption; for the remaining candidate schemes, they are sorted from high to low system throughput.

[0065] S4, randomly select N from database D. ini Each of the candidate solutions is used as an initial solution and placed into the set ∏. loc and ∏ cur .

[0066] S5 uses the buffer configuration scheme in database D as the sample tag data and the corresponding throughput and energy consumption as the tag data. It establishes a proxy model based on the ensemble learning algorithm - ExtraTreesRegressor.

[0067] In the Python simulation environment, the surrogate model can be built directly using the extreme random tree through the Skleam library; the database data capacity is set to 50I, where I is the number of processing equipment in the manufacturing system.

[0068] S6 proposes a local neighborhood search operator for ∏ loc A local domain search is performed on the candidate solutions to generate new candidate solutions.

[0069] There are three types of local domain search methods, such as... Figure 4 As shown, for each current candidate solution, a new candidate solution is generated by randomly selecting a search method each time, and this process is repeated multiple times.

[0070] The first method is:

[0071]

[0072] Where i or k represents a randomly selected buffer, ΔB represents the amount of capacity change at a certain buffer position, and randint(C1, C2) represents the integer C1 or C2. In this embodiment, C1 is 1 and C2 is 2.

[0073] The second method is:

[0074]

[0075] Where i or k represents a randomly selected buffer.

[0076] The third method is:

[0077]

[0078]

[0079] Where i or k represents a randomly selected buffer.

[0080] S7 uses the proxy model established in S5 to estimate the system throughput and energy consumption corresponding to the new candidate scheme generated in S6.

[0081] S8 sorts the new candidate solutions generated in S6.

[0082] Among them, for N where the system throughput is greater than the target throughput hig There are N candidate solutions, sorted from low to high system energy consumption. hig >N ini Select the first N ini Update of candidate solutions ∏ loc Otherwise, select N. higThere are 10 candidate solutions. For the candidate solutions whose system throughput is less than the target throughput, they are sorted from highest to lowest system throughput, and the top N solutions are selected. ini -N hig There are 10 candidate solutions, combined with the selected N. hig Update of candidate solutions ∏ loc .

[0083] S9 determines whether the local neighborhood search stopping condition is met.

[0084] The stopping condition is set to the maximum allowed number of loops J. loc Here, J loc Set to J loc =I-1, where I is the number of processing devices in the manufacturing system. If the stopping condition is met, proceed to S10; otherwise, return to S6, targeting ∏. loc Local domain search is performed on the candidate solutions.

[0085] S10, targeting ∏ loc and ∏ cur The candidate solutions undergo a global domain search to generate new candidate solutions.

[0086] First, for each of the existing candidate solutions... The new candidate solutions can be obtained as follows:

[0087]

[0088]

[0089] Where i is a randomly selected buffer, rand(C3, C4) represents a random number between constants C3 and C4, where C3 is 0.4 and C4 is 0.9, rand(C5, C6) represents a random number between constants C5 and C6, where C5 is 0 and C6 is 1, and the constant C is 0.3, rand(C8, C9) represents a random number between constants C8 and C9, where C8 is -1 and C9 is 1, and b i (bes) represents the capacity of buffer i in the first-ranked candidate scheme selected by S8, b i (ran) represents the capacity of buffer i in a candidate scheme randomly selected from the candidate schemes chosen by S8.

[0090] S8 chose N ini There are 1 candidate solutions, and each candidate solution is considered sequentially. The bootstrap value is calculated using the corresponding system throughput and energy consumption, as follows:

[0091]

[0092] Among them, TH max and TH min These respectively indicate that S8 selected N. ini The maximum and minimum system throughput among the candidate schemes, TH n EC represents the system throughput corresponding to candidate solution n. max and EC min These respectively indicate that S8 selected Ni ni The maximum and minimum energy consumption of the system corresponding to each candidate scheme, EC n This represents the system energy consumption corresponding to candidate scheme n.

[0093] if This generates a new candidate solution:

[0094]

[0095]

[0096] Where i is a randomly selected buffer, rand(C10, C11) represents a random number between the constants C10 and C11, where C10 is 0 and C11 is 1, and b i b represents the capacity of buffer i for scheme n among the candidate schemes selected by S8. i (ran1) and b i (ran2) represents the capacity of buffer i in two randomly selected candidate schemes from the candidate schemes selected by S8.

[0097] S11, calculate the system throughput and energy consumption corresponding to the new candidate scheme generated in S10.

[0098] S12, sorting the new candidate solutions generated in S10.

[0099] Among them, for N where the system throughput is greater than the target throughput hig There are N candidate solutions, sorted from low to high system energy consumption. hig >N ini Select the first N ini Update of candidate solutions ∏ loc and ∏ cur Otherwise, select N. hig For candidate solutions where the system throughput is less than the target throughput, the top N solutions are selected and sorted from highest to lowest system throughput. ini -N hig There are 10 candidate solutions, combined with the previously selected N. hig Update of candidate solutions ∏ loc and ∏ cur .

[0100] S13, will ∏ cur The candidate solution ranked first in the database is added to database D, and the candidate solution ranked last in database D is deleted.

[0101] S14, determine whether the global domain search stopping condition is met.

[0102] The stopping condition is set to the maximum allowed number of loops J. glo Here, J glo Set to 10I, where I represents the number of processing devices within the manufacturing system. If the stopping condition is met, proceed to S15; otherwise, return to S6, targeting ∏. loc Local domain search is performed on the candidate solutions.

[0103] S15, select the candidate scheme ranked first from database D as the final buffer configuration scheme.

[0104] This method balances global and local search capabilities, significantly improving the solution quality for buffer configuration optimization in manufacturing systems. Compared to existing configuration methods, the solution quality is significantly enhanced. This method employs machine learning to construct a surrogate model to replace discrete simulation software in evaluating buffer configuration candidate schemes during the local search phase, improving system throughput and energy consumption computational efficiency. While significantly improving solution quality, the solution efficiency remains comparable to existing configuration methods, achieving a solution quality improvement within a similar computation time. Furthermore, the database is updated in real-time during buffer configuration optimization, ensuring the prediction accuracy of the surrogate model and realizing dynamic data-driven technology.

[0105] To further verify the effectiveness of this method, numerical experiments were also conducted in this embodiment, comparing it with classic self-inspired methods: simulated annealing, tabu search, and genetic algorithms. The numerical experiment settings are shown in Table 1, and Table 2 presents the results. The results show that the computation time of this method is similar to that of simulated annealing and tabu search algorithms, but the solution quality is significantly improved; the computation time and solution quality of this method are significantly better than those of the genetic algorithm.

[0106] Table 1: Numerical Experiment Setup

[0107]

[0108]

[0109] Table 2: Numerical Experiment Results

[0110]

[0111] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A data-driven method for optimizing the configuration of energy-saving buffers in a manufacturing system, characterized in that, Includes the following steps: S1, satisfying b1+b2+…+b I-1 Given B, randomly generate N. dat Candidate solutions for manufacturing system buffer configuration S2, establish a system simulation model and calculate the system throughput and energy consumption; S3, based on N dat a database of buffer configuration candidates and corresponding system throughput and energy consumption; Sort the candidate solutions in database D; S4. Randomly select N candidate solutions from database D as initial solutions, and put them into sets Π and Π respectively. ini loc cur ;​​ S5, establish the agent model; S6, performing local domain search on the Π loc candidate solution to generate a new candidate solution; S7 calculates the system throughput and energy consumption corresponding to the new candidate schemes generated in S6 using the proxy model; S8, rank the new candidate solutions generated in S6, update Π loc ; S9, determine whether the local neighborhood search stopping condition is met; if the stopping condition is met, proceed to S10. Otherwise, go back to S6 and target Π. loc Perform local domain search on candidate solutions; S10, targeting Π loc and Π cur A global domain search is performed on the candidate solutions to generate new candidate solutions; S11, calculate the system throughput and energy consumption corresponding to the new candidate scheme generated in S10; S12, sorting the new candidate solutions generated in S10; S13, will Π cur The candidate solution ranked first in the database is added to database D, and the candidate solution ranked last in database D is deleted. S14, determine if the global neighborhood search stopping condition is met; if the stopping condition is met, proceed to S15; otherwise, return to S6, targeting Π. loc Perform local domain search on candidate solutions; S15, select the candidate scheme ranked first from database D as the final buffer configuration scheme.

2. The data-driven energy-saving buffer configuration optimization method for manufacturing systems according to claim 1, characterized in that, In S3, candidate schemes with system throughput greater than the target throughput are sorted from low to high system energy consumption; the remaining candidate schemes are sorted from high to low system throughput.

3. The data-driven energy-saving buffer configuration optimization method for manufacturing systems according to claim 2, characterized in that, The local neighborhood search method in S6 is as follows: Where i or k represents a randomly selected buffer, ΔB represents the change in capacity at a certain buffer position, and randint(C1,C2) represents the integer C1 or C2.

4. The data-driven energy-saving buffer configuration optimization method for manufacturing systems according to claim 2, characterized in that, The local neighborhood search method in S6 is as follows: Where i or k represents a randomly selected buffer.

5. The data-driven energy-saving buffer configuration optimization method for manufacturing systems according to claim 2, characterized in that, The local neighborhood search method in S6 is as follows: Where i or k represents a randomly selected buffer.

6. The data-driven energy-saving buffer configuration optimization method for manufacturing systems according to claim 2, characterized in that, In S8, for N where the system throughput is greater than the target throughput hig There are N candidate solutions, sorted from low to high system energy consumption. hig >N ini Select the first N ini Update of candidate solutions Π loc ;otherwise Select N hig There are 10 candidate solutions. For the candidate solutions whose system throughput is less than the target throughput, they are sorted from highest to lowest system throughput, and the top N solutions are selected. ini -N hig There are 10 candidate solutions, combined with the selected N. hig Update of candidate solutions Π loc .

7. The data-driven energy-saving buffer configuration optimization method for a manufacturing system according to claim 6, characterized in that, In S9, the stopping condition is set to the maximum allowed number of loops, J. loc J loc Set to J loc =I-1, where I is the number of processing equipment in the manufacturing system.

8. The data-driven energy-saving buffer configuration optimization method for a manufacturing system according to claim 1, characterized in that, S10 sequentially addresses each of the existing candidate solutions The new candidate solutions are as follows: Where i is a randomly selected buffer, rand(C3,C4) represents a random number between constants C3 and C4, rand(C5,C6) represents a random number between constants C5 and C6, C7 is a constant, rand(C8,C9) represents a random number between constants C8 and C9, and b i (bes) represents the capacity of buffer i in the first-ranked candidate scheme selected by S8, b i (ran) represents the capacity of buffer i in a candidate scheme randomly selected from the candidate schemes selected by S8; N selected for S8 ini Any one of the candidate solutions The bootstrap value is calculated using the corresponding system throughput and energy consumption, as follows: Among them, TH max and TH min These respectively indicate that S8 selected N. ini The maximum and minimum system throughput among the candidate schemes, TH n EC represents the system throughput corresponding to candidate solution n. max and EC min These respectively indicate that S8 selected N. ini The maximum and minimum energy consumption of the system corresponding to each candidate scheme, EC n This represents the system energy consumption corresponding to candidate scheme n; if This generates a new candidate solution: Where i is a randomly selected buffer, rand(C10,C11) represents a random number between the constants C10 and C11, and b i b represents the capacity of buffer i for scheme n among the candidate schemes selected by S8. i (ran1) and b i (ran2) represents the capacity of buffer i in two randomly selected candidate schemes from the candidate schemes selected by S8.

9. The data-driven energy-saving buffer configuration optimization method for a manufacturing system according to claim 8, characterized in that, S12 refers to N where the system throughput is greater than the target throughput. hig There are N candidate solutions, sorted from low to high system energy consumption. hig >N ini Select the first N ini Update of candidate solutions Π loc and Π cur ;otherwise Select N hig For candidate solutions where the system throughput is less than the target throughput, the top N solutions are selected and sorted from highest to lowest system throughput. ini -N hig There are 10 candidate solutions, combined with the selected N. hi Update of candidate solutions Π loc and Π cur .

10. The data-driven energy-saving buffer configuration optimization method for a manufacturing system according to claim 1, characterized in that, In S14, the stopping condition is set to the maximum allowed number of loops J. glo J glo Set to 10I, where I represents the number of processing devices within the manufacturing system.

Images