A green optimization scheduling method based on distributed assembly production process of automobile parts

Abstract

The present application relates to a kind of green optimization scheduling method based on distributed assembly production process of automobile parts, belong to intelligent optimization scheduling technical field of manufacturing process.The present application establishes scheduling model for the processing, assembly process of automobile parts, and proposes a kind of super-heuristic three-dimensional distribution estimation algorithm for optimization solution for this model;Among them, scheduling model is established according to each part and the machining process and machining time contained in its parts, and the optimization goal is determined as minimizing automobile part maximum completion time and carbon emission simultaneously.The present application can obtain the optimal scheme of automobile parts production manufacturing process scheduling problem in a short time, so as to achieve the purpose of energy saving and emission reduction, reduce time cost.

Description

Technical Field

[0001] This invention relates to a green optimization scheduling method based on the distributed assembly production process of automotive parts, belonging to the field of intelligent optimization scheduling technology for manufacturing production processes. Background Technology

[0002] In the context of industrial globalization, various manufacturing industries have emerged. Automobiles, as modern industrial products, are not simply assembled in a single factory; instead, their components are outsourced to various manufacturers for collaborative processing. This leads to the distributed assembly production scheduling problem for automotive parts, namely, the production scheduling problem of minimizing time and cost by selecting which outsourced manufacturers to process parts. The high-efficiency operation of outsourced manufacturers in various industries inevitably leads to high efficiency, which in turn results in high energy consumption. Typically, the main source of energy is fossil fuels, and burning fossil fuels releases large amounts of carbon dioxide into the atmosphere, making global warming a serious environmental problem. To reduce carbon dioxide emissions and conserve energy, low-carbon scheduling in the manufacturing industry is imperative. The processing and production of automotive parts generally falls under the classic flow shop scheduling problem, which has attracted widespread attention from researchers both domestically and internationally due to its broad industrial and economic applications. The traditional flow shop scheduling problem considers a set of machines with identical production processes, and the optimization objective is to find a sequence of operations that satisfies one or a set of optimization indices. However, Okwudire, Chinedum, and others proposed that distributed manufacturing (DM) allows geographically dispersed production, typically small-scale and close to end users, leading to research on the Distributed Permutation Flow-shop Scheduling Problem (DPFSP). As early as 2010, Naderi and Ruiz began researching DPFSP. In 2013, Hatami further studied the Distributed Assemble Permutation Flow-shop Scheduling Problem (DAPFSP), an extension of DPFSP and also an NP-hard problem. The DAPFSP discussed in this patent is an extension of DAPFSP in terms of low-carbon energy saving; it is a complex scheduling problem involving adjustable equipment speeds and two coupled stages, requiring collaborative optimization of completion time and total energy consumption.

[0003] Compared to traditional scheduling problems, low-carbon scheduling achieves energy conservation, emission reduction, and cost reduction through comprehensive and coordinated optimization of resource allocation, process arrangement, and processing modes. In terms of machine speed control within low-carbon scheduling, machine processing speed can be dynamically adjusted according to different processing technologies. Low-speed processing consumes less energy and takes longer, while higher-speed processing produces the opposite effect. The processing of automotive parts generates significant energy consumption. Different parts are numbered, and each part contains several components. The different processing sequences of these components will affect the completion time of the batch of parts to varying degrees. Currently, few companies in automotive parts processing consider directly assembling completed parts. This invention directly assembles parts belonging to the same component in the same factory, significantly shortening the company's production cycle.

[0004] At the algorithmic level, the Hyper Heuristic 3-dimensional Estimation of Distribution Algorithm (HH3DEDA) is chosen. Compared with traditional heuristic algorithms, the hyperheuristic algorithm employs a higher-level heuristic strategy. It not only considers traditional neighborhood search and greedy selection but also introduces more complex optimization strategies and metaheuristic algorithms, such as genetic algorithms and tabu search. These strategies and algorithms cooperate to fully utilize various information in the search space, thereby achieving better allocation results. The hyperheuristic 3-dimensional distribution estimation algorithm described in this invention typically consists of high-level policy domain operations and low-level problem domain operations. The high-level policy domain operations guide the global search direction of the algorithm, while the low-level problem domain operations are responsible for local information in the search space. Because the hyperheuristic algorithm employs complex optimization strategies and metaheuristic algorithms, its computational complexity is high, enabling it to handle problems of different scales and complexities. Furthermore, the algorithm involved in this invention employs multiple optimization strategies and metaheuristic algorithms, enabling global search and optimization to obtain a better allocation scheme. Summary of the Invention

[0005] This invention provides a green optimization scheduling method based on the distributed assembly production process of automotive parts, which can provide automotive parts companies with reasonable processing solutions in a relatively short period of time, thereby enabling each processing machine to be used in a reasonable manner, thus achieving the goals of energy saving, environmental protection, and shortening the construction period.

[0006] The technical solution of this invention is: a green optimization scheduling method based on the distributed assembly production process of automotive parts. First, a scheduling model for the processing and assembly of automotive parts is established, with the optimization objective being to simultaneously minimize the completion time and energy consumption of the automotive parts production process. Since each automotive part contains multiple different types of components, and each component has multiple processing steps, a hyperheuristic three-dimensional distribution estimation algorithm is proposed to optimize the solution based on the processing time of each process on each machine.

[0007]

[0008]

[0009]

[0010]

[0011]

[0012]

[0013]

[0014]

[0015] TCE=ε×TEC (20)

[0016] TCE(π,V)=minTCE (21)

[0017]

[0018] Among them, y jv (t) is a binary symbol: it represents that machine j in any factory f is in processing mode at time t with speed v, and is assigned a value of 1; it is assigned a value of 0 when in standby mode. F represents that there are F factories of the same type, and each factory contains a processing stage and an assembly stage; Δ δ This indicates that there are a total of δ parts to be processed. Indicates the parts contained in each accessory; M m This indicates that there are m processing devices in the processing stage; Δ represents the component located in factory f. l The actual processing speed of machine j when the first part is being processed.

[0019] The model must satisfy the following constraints:

[0020] (1) Each part can be assigned to any factory for processing;

[0021] (2) Each factory contains at least one component to be processed;

[0022] (3) In each assembly line workshop, all parts must be processed in sequence, that is, first processed on equipment M1, then processed on equipment M2, and so on, until the last equipment M1. m The processing is completed on the upper part;

[0023] (4) Parts belonging to the same accessory must be processed continuously. Processing only a part of the accessory is not in line with the assembly conditions.

[0024] (5) Components Δ of factory f l The operation of the part at position i on machine j is represented as follows: Once you operate Startup must be performed without interruption; For operation Processing time;

[0025] (6) Once the accessories Δ l Once all parts are machined, they are immediately delivered to assembly machine M in the same factory. A Assembly can begin if all parts of the component have been machined and the assembly machine is idle. Once the assembly machine is started, it cannot be interrupted. l Indicates accessory Δ l In assembly machine M A Assembly time on

[0026] (7) Each machine can only perform one operation at a time;

[0027] Formula (1) represents the completion time required for the part at position 1 in factory f to be processed on machine M1; Formula (2) represents the completion time of the part at position i in factory f on machine 1; Formula (3) represents the completion time of the part at position j in factory f; Formula (4) represents the completion time of the part at position i in factory f on machine j; Formula (5) represents the completion time of the accessory at position l in factory f; Formula (8) represents the maximum completion time of the accessory in factory f.

[0028] Formulas (7) and (8) represent the maximum completion time C, respectively. max Total Energy Consumption (TEC, pp) jv Represents operation O i,j When processing is performed on machine j∈M at speed v∈V, the corresponding processing time is p. ij =p ij / v The corresponding power consumption per unit time is pp jv sp jThis means that the machine will only completely shut down when all operations are complete, and each machine j∈M will be in standby mode when not performing any tasks, with a power consumption sp per unit time. j Where TEC is the total energy consumption, ε is the carbon emission per unit energy consumption, and formula (11) represents that the encoding of an individual in the problem domain consists of two parts: one part is the workpiece processing sequence π; the other part is the workpiece processing speed reference table V.

[0029] Among them O i,j ∈O represents the i-th operation to be processed on each machine, v(i,j)∈V represents the running speed of the i-th operation on machine j. For ease of notation, the feasible solution is represented by (π,V), and the corresponding completion time and total energy consumption are denoted as C. max According to formulas (9) and (10), the only difference between carbon emissions and total energy consumption is a constant coefficient. Optimizing TEC or TCE will produce equivalent results.

[0030] As a further aspect of this invention, to enhance the local search capability of the algorithm, 12 low-level heuristic operations are designed, namely:

[0031] LLH1: Randomly swap any two components within a critical factory;

[0032] LLH2: Take a component from a critical factory and randomly insert it into another location;

[0033] LLH3: Randomly reverses the order of parts within a critical factory;

[0034] LLH4: Randomly swap parts between two factories;

[0035] LLH5: Take a component from a critical factory and randomly insert it into another factory;

[0036] LLH6: Randomly swap any two components within a critical factory;

[0037] LLH7: Select an accessory from the critical factory, then select a part from the sequence of parts contained in that accessory, and randomly insert it into another position in the sequence;

[0038] LLH8: Select any component from the critical factory, then select any two parts from the sequence of parts contained in that component, and randomly swap the positions of these two parts;

[0039] LLH9: Select any component from the critical factory, then select a segment from the sequence of parts contained in that component, and reverse that segment;

[0040] LLH10: Randomly select two machines to perform a sudden change in processing speed;

[0041] LLH11: Randomly select two machines to perform a processing speed exchange operation;

[0042] LLH12: Randomly select two machines to perform processing speed insertion operations.

[0043] As a further aspect of the present invention, the specific steps of the green optimization scheduling method based on the distributed assembly production process of automotive parts are as follows:

[0044] Step 1: Determine the encoding and decoding method: Use a three-dimensional vector encoding method, namely factory sequence, machine sequence, and part sequence; the underlying problem domain individual π is the part processing sequence, and the decoded individual contains the part sequence π assigned to each factory. f and the sequence of parts contained in each accessory. j Two parts;

[0045] Step 2: Initialize the underlying problem domain population and the 3D probability model; set the population size to popsize (hereinafter referred to as ps);

[0046] Step 3: Design the low-level heuristic operation LLH;

[0047] Step 4: Using feedback from the lower-level problem domain, sample and generate a high-level policy domain population P. G With the population size set to ps, for the high-level policy domain, each individual I in the population consists of 12 low-level heuristic operations. When decoding individuals in the high-level policy domain, the low-level heuristic operations in the high-level policy domain individuals are executed sequentially from left to right for the scheduling solutions of the low-level problem domain. After each low-level heuristic operation is executed, the new solution is compared with the original solution. If the new solution is better than the old solution, the new solution replaces the old solution; otherwise, the original solution is retained, and the remaining low-level heuristic operations are executed. After all low-level heuristic operations in the high-level policy domain individuals are executed, the fitness value of the high-level policy domain individuals is calculated. This fitness value is the optimal fitness value of the updated low-level problem domain individuals.

[0048] Step 5: Use the operation selection function to evaluate the population in the lower problem domain. At this time, there are two positional situations: one is that the individual in the higher policy domain is not initially located at position 1, and the other is that the individual in the higher policy domain is initially located at position 1.

[0049] Step 6: Decode all individuals in the high-level policy domain population and obtain a new solution based on the low-level heuristics in the high-level policy domain individuals;

[0050] Step 7: Select a predetermined proportion of high-quality individuals from the high-level strategy domain population to update the probability model of the three-dimensional distribution estimation algorithm, and generate a new next-generation elite population P through sampling. G (G+1);

[0051] Step 8: Calculate the three-dimensional matrix and update the three-dimensional probability matrix model simultaneously;

[0052] Step 9: Determine if the new solution is better than the old solution. If the termination condition is met, go to Step 4 if not. Otherwise, terminate the loop.

[0053] Step 10: Based on the decoding method, obtain the processing time of the parts on the processing machine and the assembly time of the accessories on the assembly machine, obtain the initial processing sequence of the parts and accessories, and generate the corresponding scheduling Gantt chart;

[0054] Step 11: Determine the critical path of the Gantt chart corresponding to the scheduling solution, where the critical path is the process path with the longest total processing time in the scheduling Gantt chart; at the same time, determine the critical path of each non-dominated solution in the non-dominated solution set. If there are multiple critical paths, randomly select one of them.

[0055] Step 12: Perform four neighborhood searches based on the critical path—exchange, insertion, reversal, and mutation—on the steps in the non-dominated solutions in sequence. If the new solution is better than the old solution, replace the old solution with the new non-dominated solution, redetermine the critical path, and then perform subsequent neighborhood search operations. If they are not mutually dominant, add the new solution to the non-dominated solution set, and continue the remaining neighborhood search operations on the current solution.

[0056] Step 13: Repeat the operations of Step 11 and Step 12 until all the processing steps corresponding to the obtained non-dominated solutions cannot shorten the total processing time;

[0057] Step 14: For each solution (π, V) in the non-dominated solution set, calculate... and Then we get

[0058] Step 15: Sort all non-dominated solutions in ascending order of α, and divide the non-dominated solution set Ω into Ω0. c and π e Two parts, where |Ω|=|Ω c |+|Ω e |;Ω c The nondominated solution α has a small value, requiring the maximum completion time C to be calculated. max Optimize, Ω e The nondominated solution α has a large value, so the total energy consumption TCE needs to be optimized.

[0059] Step 16: As shown in Step 15, processes outside the critical path are processes on the non-critical path. Based on the machine speed of the processes on the non-critical path, determine whether there is an allowable speed reduction margin; in Ω e For all non-dominated solutions, the energy-saving speed regulator strategy is then applied, which means keeping the machining speed of all parts on the critical path constant while adjusting the machining speed of parts on the non-critical paths, so that the same maximum completion time C is achieved. max To achieve the lowest possible total energy consumption (TCE) and further improve the quality of feasible scheduling solutions;

[0060] Step 17: Update the Pareto archive Ω * If the number of feasible solutions in the non-dominated solution set is greater than ps, then the crowding distance sorting criterion is used to sort all feasible scheduling solutions, while removing some solutions with the smallest crowding distance, until the number of feasible scheduling solutions in the population reaches ps.

[0061] This invention establishes a scheduling model for the processing and assembly of automotive parts, with the optimization objective of simultaneously minimizing the completion time and energy consumption of the automotive parts production and forming process, i.e., f = argmin{C max Since each automotive part contains multiple different types of components, and each component involves multiple processing steps, a hyperheuristic 3D distribution estimation algorithm is proposed for optimization based on the processing time of each machine. For the hyperheuristic 3D distribution estimation algorithm, the higher-level policy domain guides the lower-level problem domain as follows: 1. Setting the search space: The higher-level domain needs to set parameters such as the range and precision of the search space to ensure the algorithm can find the global optimum. 2. Setting the objective function: In the 3D distribution estimation problem, the objective function is usually the error between the estimated value and the actual value. The higher-level domain needs to design a suitable objective function to guide the algorithm in the correct direction. 3. Setting parameters: Hyperheuristic algorithms typically require setting many parameters, such as the number of iterations, population size, crossover probability, and mutation probability. Parameters also need to be set in the lower-level problem domain to ensure the algorithm can effectively explore the search space. 4. Assigning tasks: In the 3D distribution estimation algorithm, heuristic operations in the lower-level problem domain typically include local search and neighborhood search. The higher-level policy domain needs to assign different tasks to the lower-level problem domain based on factors such as the size and complexity of the search space to ensure the coordination and consistency of the entire algorithm. 5. Optimize low-level algorithms: In practice, the performance of low-level algorithms is often the bottleneck of hyperheuristic algorithms. Higher levels need to guide the optimization and improvement of lower levels to improve the overall performance and efficiency of the algorithm. This can be achieved by adjusting algorithm parameters and improving the algorithm structure.

[0062] The beneficial effects of this invention are:

[0063] 1. This invention establishes a scheduling model for the processing and assembly of automotive parts, and designs an energy-saving strategy based on machine process exchange and mutation to address the green optimization scheduling problem of distributed assembly production of automotive parts, which can significantly reduce energy consumption in the assembly line workshop.

[0064] 2. This invention adjusts the processing speed of the machine to achieve low-speed and low-consumption production of workpieces without affecting the production schedule. This approach not only contributes to sustainable development but also effectively reduces the cost of processing plants.

[0065] 3. This invention presents a hyperheuristic three-dimensional distribution estimation algorithm (HH3DEDA) designed for the green optimization scheduling problem in the distributed assembly production process of automotive parts. This algorithm utilizes heuristic search principles, continuously adjusting the search direction during the search process to effectively avoid local optima and quickly find the global optimum, maximizing efficiency while ensuring solution accuracy. Furthermore, HH3DEDA has broad applicability: it can estimate the joint distribution of three variables based on data within the problem size, which may represent position, velocity, or any other continuous variables in three-dimensional space. HH3DEDA can not only be used for distribution estimation of three-dimensional data but can also be extended to higher dimensions and is applicable to different types of data, demonstrating strong versatility. Compared to the two-dimensional probabilistic model used in EDA, the three-dimensional probabilistic model can simultaneously learn the block structure information and position information of high-quality individuals. When sampling and generating new individuals, it can more accurately determine the placement of high-quality block structures in the scheduling solution, thereby quickly guiding the algorithm to the high-quality solution region for searching and improving the algorithm's search efficiency.

[0066] 4. The hyperheuristic 3D distribution estimation algorithm proposed in this invention has advantages such as high efficiency, high accuracy, and wide applicability, and can be widely applied in fields such as data analysis, machine learning, and data mining.

[0067] 5. This invention can provide automotive parts companies with reasonable processing solutions in a relatively short period of time, allowing each processing machine to be used rationally, thereby achieving the goals of energy conservation, environmental protection, and shortening the construction period. Attached Figure Description

[0068] Figure 1 This is a schematic diagram of the overall structure of the present invention;

[0069] Figure 2 This is a schematic diagram of the hyperheuristic method of the present invention;

[0070] Figure 3 This is a schematic diagram of the high-level policy domain population sampling and update method of the three-dimensional probabilistic model of the present invention;

[0071] Figure 4 This is the initial Gantt chart of the present invention;

[0072] Figure 5 This is the critical path Gantt chart of the present invention;

[0073] Figure 6 This is a schematic diagram of an individual high-level strategy domain of the present invention;

[0074] Figure 7 This is a flowchart of the algorithm of the present invention. Detailed Implementation

[0075] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments.

[0076] like Figures 1-7 As shown, a green optimization scheduling method based on the distributed assembly production process of automotive parts is presented. A. Symbols and Explanation

[0077] (1) Symbol

[0078] Parameters: f: total number of factories; δ: total number of parts to be processed; n: total number of parts to be processed; n f : The number of parts allocated to factory f; n l : Belongs to product Δ l Number of parts; m: total number of machines; popsize: population size; ε: carbon emissions per unit of energy consumption (kg CO2 equivalent / kWh).

[0079] Subscript: f: Factory subscript f = 1, 2, ..., F; Δ: Parts sequence Δ l =Δ1,Δ2,...,Δ δ M: Machine sequence M = M1, M2, ..., M m i: Index of the part to be processed, i = 1, 2, ..., n; The sequence of part operations assigned to factory f and located at the i-th processing position: j: Machine index in all factories, j = 1, 2, ..., m; V: Variable speed sequence V = V1, V2, ..., V d .

[0080] variable: The component Δ in factory f l The operation of the parts required for its assembly during the i-th process on machine j; Machine J performs the operation Processing speed at that time; Assembly machine M A Processed parts Δ l The speed at that time; The component Δ in factory f l The processing time of the parts required for its assembly when the first process is carried out on machine 1; The component Δ in factory f l The processing time of the parts required for its assembly when the i-th process is processed on machine 1; The component Δ in factory f l The processing time of the first part when the parts required for its assembly are processed on machine j; The component Δ in factory f l The processing time of the i-th part when the parts required for its assembly are processed on machine j; Accessories Δ l In assembly machine M A Assembly time for processing; The component Δ in factory f l The completion time of the first process of the parts required for its assembly on machine 1; The component Δ in factory f l The completion time of the parts required for its assembly when the i-th (i≠1)th process is processed on machine 1; The component Δ in factory f l The completion time of the parts required for its assembly when the first operation is performed on machine j (j≠1); The component Δ in factory f l The completion time of the parts required for its assembly when the i-th (i-≠1) process is carried out on machine j (j≠1); Assembly completion time of component Δ1 located in factory f; The component Δ in factory f l Assembly completion time; C max : Maximum completion time; TEC: Energy consumption of all plants; (π,V): A feasible solution or individual.

[0081] Decision variable: y jv (t) is a binary symbol: it represents that machine j in any factory f is in processing mode at time t with speed v, and is assigned a value of 1; it is assigned a value of 0 when in standby mode; F represents that there are F factories of the same type, and each factory contains a processing stage and an assembly stage; Δ δ This indicates that there are a total of δ parts to be processed. Indicates the parts contained in each accessory; M m This indicates that there are m processing devices in the processing stage;

[0082] pp jv :operate The power consumption per unit time when processing at speed v∈V on machine j∈M; sp j This means that the machine will only completely shut down when all operations are complete. Each machine in j∈M will be in standby mode when not performing any tasks, with a power consumption of sp per unit time. j .

[0083] (2) Constraints:

[0084] 1. Each component can be assigned to any factory for processing.

[0085] 2. Each factory must contain at least one component to be processed.

[0086] 3. In each assembly line workshop, all parts must be processed sequentially, that is, first processed on machine M1, then on machine M2, and so on, until the last machine M... m The processing is completed on the upper part.

[0087] 4. Parts belonging to the same component must be processed continuously; processing only a portion of the components does not meet the assembly requirements.

[0088] 5. Components Δ of factory f l The operation of the part at position i on machine j is represented as follows: Once you operate Startup must be performed without interruption; For operation Processing time, For the parts Δ located in factory f l The actual processing time for the parts required for assembly to be processed on machine j during the i-th process.

[0089] 6. Once the accessories Δ l Once all parts are machined, they are immediately delivered to assembly machine M in the same factory. A Assembly can begin if all parts of the component have been machined and the assembly machine is idle. Once the assembly machine is started, it cannot be interrupted. Indicates accessory Δ l In assembly machine M A Assembly time on Indicates accessory Δ l In assembly machine M A The actual assembly time.

[0090] 7. Each machine can only perform one operation at a time.

[0091] B. Optimize the scheduling model and optimization objectives

[0092] Based on the relationship between the location of the processed parts, the location of the previous process for the same processed part, and the location of the processing machine, the technical solution of this invention is proposed: a green optimization scheduling method based on the automotive parts production and forming process. This method establishes a scheduling model for the processing and assembly of automotive parts, with the optimization objective being to simultaneously minimize the completion time and energy consumption of the automotive parts production and forming process, i.e., f = argmin{C max Since each automotive part contains multiple different types of components, and each component involves multiple processing steps, a hyperheuristic three-dimensional distribution estimation algorithm is proposed for optimization based on the processing time of each machine according to the processing steps.

[0093] The two optimization objectives of this invention are maximum completion time C. max The formula for calculating total energy consumption (TEC) is as follows:

[0094]

[0095]

[0096]

[0097]

[0098]

[0099]

[0100]

[0101]

[0102] TCE=ε×TEC (31)

[0103] TCE(π,V)=minTCE (32)

[0104]

[0105] In the above formula, F represents a total of F factories of the same type, and each factory includes a processing stage and an assembly stage; Δ δ This indicates that there are a total of δ parts to be processed. Indicates the parts contained in each accessory; M mLet m be the number of processing machines in the processing stage. Formula (1) represents the completion time required for the part at position 1 in factory f to be processed on machine M1; Formula (2) represents the completion time for the part in factory f to be processed on machine 1 for the i-th operation; Formula (3) represents the completion time for the part in factory f to be processed on machine j for the first operation; Formula (4) represents the completion time for the part at position i in factory f to be processed on machine j; Formula (5) represents the assembly completion time of the accessory at position l in factory f; Formula (6) represents the maximum completion time of the accessory in factory f; Formulas (7) and (8) are the maximum completion times C and C, respectively. max Total energy consumption TEC, of ​​which pp jv Represents operation O i,j When processing is performed on machine j∈M at speed v∈V, the corresponding processing time is p. ij =p ij / v The corresponding power consumption per unit time is pp jv ;sp j This means that the machine will only completely shut down when all operations are complete, and each machine j∈M will be in standby mode when not performing any tasks, with a power consumption of sp per unit time. j TEC represents total energy consumption, and ε represents carbon emissions per unit of energy consumption. Formula (11) represents the encoding of an individual in the problem domain, which consists of two parts: one part is the workpiece processing sequence π; the other part is the workpiece processing speed reference table V. i,j Let ∈O represent the i-th operation to be performed on each machine, and v(i,j)∈V represent the speed at which the i-th operation is performed on machine j. For ease of notation, the feasible solution is represented by (π,V), and the corresponding completion time and total energy consumption are denoted as C. max (π,V) and TEC(π,V). According to formulas (9) and (10), the only difference between carbon emissions and total energy consumption is a constant coefficient, and optimizing TEC or TCE will produce equivalent results. This invention is a typical multi-objective optimization and complex flow shop scheduling problem of adjusting machine speed. Since there is an obvious contradiction between the maximum completion time and total energy consumption, there is no absolutely optimal solution to this problem. By weighing the two objectives, a set of Pareto optimal (non-disadvantageous) solutions can be obtained, so that the multiple objective functions involved can be optimized simultaneously. The optimal scheduling scheme of this invention is a set of feasible solutions, not a single solution. The optimal or near-optimal solution set must be determined by specific multi-objective optimization techniques. At the same time, this invention has a close coupling relationship between the production and assembly stages. The machine speed directly determines the processing operation, and the processing stage affects the assembly stage, that is, the start time of the assembly stage is based on the completion time of the processing stage. The optimization objective should not only consider the optimal production efficiency criterion, but also meet the production requirements of low-carbon manufacturing. When machine M jWhen processing parts or accessories at a faster speed, the processing time will be shorter, but the total energy consumption will increase accordingly. Among them, v k1 ,v k2 ∈V, if but For any two feasible solutions (π', V') and (π”, V”), if C max (π',V')≥C max If (π”,V”) and TEC(π’,V’)≥TEC(π”,V”), then the feasible solution (π”,V”) dominates the feasible solution (π’,V’), and is symbolically represented as

[0106] like Figure 1 The diagram shows the overall structure of a distributed assembly and replacement flow workshop scheduling system for automotive parts processing, provided by this invention. Automotive parts refer to components such as engines, transmission systems, steering systems, running systems, braking systems, and the vehicle body. This invention delivers these automotive parts to corresponding distributors for processing and assembly. Taking the automotive braking system as an example, the entire braking system includes a power supply device, a control device, and brakes. This invention treats these devices as parts to be processed, and the assembly materials required for each device as parts to be processed. Figure 2 As shown, the hyperheuristic method of this invention consists of two parts: one part is a high-level policy domain population, in which each individual is composed of low-level heuristic operations; the other part is a low-level problem domain population, which is composed of problem domain decoding sequences. The high-level structure guides the low-level structure to update the solution, and then... Figure 3 The three-dimensional probabilistic model shown updates the population in the high-level policy domain. As high-level individuals sequentially execute low-level heuristic operations, the solutions in the low-level problem domain are also continuously updated, thereby finding the optimal solution set.

[0107] To address the two optimization objectives of cost minimization and energy consumption minimization required by this invention, the green optimization scheduling method based on the distributed assembly production process of automotive parts provided by this invention adopts the following steps:

[0108] Step 1: Determine the encoding and decoding method: Use a three-dimensional vector encoding method, namely factory sequence, machine sequence, and part sequence; the underlying problem domain individual π is the part processing sequence, and the decoded individual contains the part sequence π assigned to each factory. f and the sequence of parts contained in each accessory. j Two parts. For ease of understanding, Figure 4The Gantt chart for a feasible solution of the present invention includes 8 parts, 4 processing machines, 1 assembly machine, 2 factories, and 4 types of accessories. The first factory contains two accessories, where accessory pro2 contains 2 parts {Λ6,Λ1} and accessory pro4 contains 2 components, namely {Λ5,Λ3}; the second factory also includes two accessories, accessory pro1 contains parts {Λ7,Λ2} and accessory pro3 contains parts {Λ4,Λ8}, and the subsequences of all parts assigned to the two factories are π1=[6,1,5,3] and π2=[7,2,4,8].

[0109] Step 2: Initialize the underlying problem domain population and the three-dimensional probabilistic model. The population size is popsize, abbreviated as ps. The underlying problem domain population is initialized randomly to ensure the diversity and randomness of solutions. The initialization method for the three-dimensional probabilistic model is as follows:

[0110] Define P G For the high-level policy domain population of generation G in HH3DEDA, best_P G For P G A high-quality solution in the middle, sps is best_P G scale for best_P G The kth individual, Define a three-dimensional matrix Used to store information such as similar blocks and their distribution among excellent individuals in the Gth generation. The specific description is as follows:

[0111]

[0112]

[0113]

[0114]

[0115] Among them, in formula (12) This is an indicator function used to record best_P G The order of heuristic operations and the position of similar blocks in the k-th individual; Formula (13) is used to record the order of heuristic operations and the position of similar blocks of all high-quality solutions in the high-level policy domain. For best_P G The number of times the block structure [y,z] appears at position x; Formulas (14) to (15) describe the hierarchical structure of the three-dimensional matrix.

[0116] HH3DEDA uses based on Three-dimensional probability model To learn and accumulate the location information of similar blocks in high-quality solutions of high-level policy domains, so that for The probability distribution of similar blocks at position x for elements in the array is defined as follows:

[0117]

[0118] The sum of probabilities that similar blocks are distributed at position x in all high-quality solutions in the high-level policy domain is:

[0119]

[0120] The number of times similar blocks are distributed at position x in all high-quality solutions in the high-level policy domain is:

[0121]

[0122] Initialize the probability model:

[0123]

[0124] Step 3: Design Low-Level Heuristics (LLH): LLH1: Randomly swap any two parts within a critical factory. LLH2: Take a part from a critical factory and randomly insert it into another position. LLH3: Randomly reverse the order of parts within a critical factory. LLH4: Randomly swap parts between two factories. LLH5: Take a part from a critical factory and randomly insert it into another factory. LLH6: Randomly swap any two parts within a critical factory. LLH7: Select a part from a critical factory, then select a part from the sequence of parts contained in that part and randomly insert it into another position in the sequence. LLH8: Select any part from a critical factory, then select any two parts from the sequence of parts contained in that part and randomly swap their positions. LLH9: Select any part from a critical factory, then select a sequence from the sequence of parts contained in that part and reverse that sequence. LLH10: Randomly select two machines to perform a speed change operation. LLH11: Randomly select two machines to perform a speed swap operation. LLH12: Randomly select two machines to perform the processing speed insertion operation;

[0125] Step 4: Using feedback from the lower-level problem domain, sample and generate a high-level policy domain population P. GWith the population size set to ps, for the high-level policy domain, each individual I in the population consists of 12 low-level heuristic operations. When decoding individuals in the high-level policy domain, the low-level heuristic operations in the high-level policy domain individuals are executed sequentially from left to right for the scheduling solutions of the low-level problem domain. After each corresponding low-level heuristic operation is executed, the new solution is compared with the original solution. If the new solution is better than the old solution, the new solution replaces the old solution; otherwise, the original solution is retained, and the remaining low-level heuristic operations continue to be executed. When all low-level heuristic operations in the high-level policy domain individuals have been executed, the fitness value of the high-level policy domain individuals is calculated. This fitness value is the optimal fitness value of the updated low-level problem domain individuals. Figure 6 This is a schematic diagram of an individual in a high-level strategy domain;

[0126] Step 5: Use the operation selection function to evaluate the population in the lower problem domain. At this point, there are two positional situations: one is that the individual in the higher policy domain is not initially located at position 1, and the other is that the individual in the higher policy domain is initially located at position 1.

[0127] Step 5.1 For the first case: Assume Φ opt (Λ G,k ,i) is a selection function used to determine the position of Λ G,k The low-level heuristic operation occurring at the i-th position; due to similar blocks [Λ G,k (i-1),Λ G,k The probability of being selected (i) is stored in the three-dimensional probability model. In the middle, therefore Ф opt (Λ G,k i) Based on the three-dimensional probability model Sampling is performed, Ф opt (Λ G,k The specific operation process of i) is as follows:

[0128] Step 5.1.1: Randomly generate probability number r,

[0129]

[0130] Step 5.1.2: Use a roulette wheel to select the low-level heuristic operation L. c ,like

[0131]

[0132] Then c = 1, go to Step 5.1.3, if

[0133]

[0134] Then c = pos + 1.

[0135] Step 5.1.3: Return to Lc .

[0136] Step 5.2 Regarding the second case: For individuals in the high-level policy domain at initial position 1, there is no similar block [Λ]. G,k (i-1),Λ G,k (i)], and then Φ opt (Λ G,k i) Sampling cannot be performed at the initial position 1. Therefore, an initial sampling strategy is proposed for the initial position, which is described in detail below:

[0137] Step 5.2.1: Calculation

[0138]

[0139] Step 5.2.2: Determine Λ using the roulette wheel method. G,k The low-level heuristic operation L for initial position 1 in ps, k=1,... c ', random probability number r',

[0140]

[0141] Then c = 1, go to Step 5.2.3, if

[0142]

[0143] Then c = pos + 1.

[0144] Step 5.2.3: Let Λ G,k (1) (where k = 1, ..., ps) is L c '.

[0145] Step 6: Decode all individuals in the high-level policy domain population and obtain a new solution based on the low-level heuristics in the high-level policy domain individuals;

[0146] Step 7: Select the top ps×r high-quality individuals from the high-level policy domain population to update the probability model of the three-dimensional distribution estimation algorithm, and generate the next generation of a new elite population P through sampling. G (G+1);

[0147] According to the description in Step 5, P G The (G+1) update process is as follows:

[0148] Step 7.1: Let k = 1.

[0149] Step 7.2: Determine Λ based on the initial sampling strategy G,k Low-level heuristic operation L at initial position 1 c '.

[0150] Step 7.3: Let i = 2.

[0151] Step 7.4: Let Λ G,k (i)=Φ opt (Λ G,k ,i).

[0152] Step 7.5: Let i = i + 1. If i ≤ n, then go to step 4.

[0153] Step 7.6: Let k = k + 1. If k ≤ ps, then go to step 2.

[0154] Step 7.7: Output P G (G+1).

[0155] Step 8: Calculate the 3D matrix Simultaneously update the three-dimensional probability matrix model

[0156]

[0157] Step 8.1: Set the generation number G to 1;

[0158] Step 8.2: Calculation And update the probability model

[0159]

[0160] Step 8.3: Let G = G + 1. If G < maxG, then go to Step 8.2.

[0161] Step 9: Determine if the termination condition is met. If not, proceed to Step 4; otherwise, terminate the loop.

[0162] Step 10: Based on the decoding method, obtain the processing time of the parts on the machining machine and the assembly time of the accessories on the assembly machine to obtain the initial processing sequence of the parts and accessories, and generate the corresponding scheduling Gantt chart. (See...) Figure 4 ;

[0163] Step 11: Determine the critical path in the Gantt chart corresponding to the scheduling solution, such as... Figure 5 The seven operations O4, O2, O in factory 1 shown 2,1,4 O 2,1,3 O 2,1,2 O 2,1,1 O 2,6,1All operations are performed on the critical path, which is the process path with the longest total processing time in the scheduling Gantt chart; at the same time, the critical path of each non-dominated solution in the non-dominated solution set is determined, and if there are multiple critical paths, one of them is randomly selected.

[0164] Step 12: Perform four neighborhood searches based on the critical path—exchange, insertion, reversal, and mutation—on the steps in the non-dominated solutions in sequence. If the new solution is better than the old solution, replace the old solution with the new non-dominated solution, redetermine the critical path, and then perform subsequent neighborhood search operations. If they are not mutually dominant, add the new solution to the non-dominated solution set, and continue the remaining neighborhood search operations on the current solution.

[0165] Step 13: Repeat the operations of Step 11 and Step 12 until all the processing steps corresponding to the obtained non-dominated solutions cannot shorten the total processing time;

[0166] Step 14: For each solution (π, V) in the non-dominated solution set, calculate... and Then we get

[0167] Step 15: Sort all non-dominated solutions in ascending order of α, and divide the non-dominated solution set Ω into Ω0. c and Ω e Two parts, where |Ω|=|Ω c |+|Ω e |。 Ω c The nondominated solution α has a small value, requiring the maximum completion time (C) to be considered. max ) to be optimized, Ω e The non-dominated solution α has a large value, which requires optimization of the total carbon emissions (TCE).

[0168] Step 16: Gantt generated in Step 15 Figure 5 It can be seen that the processes outside the critical path are processes on the non-critical path. Based on the machine gear position of the processes on the non-critical path, it is determined whether there is an allowable speed reduction margin. In Ω e In this approach, an energy-saving speed regulator strategy is applied to all non-dominated solutions. This strategy maintains the machining speed of all parts on the critical path while adjusting the machining speed of parts on non-critical paths, achieving the same maximum completion time (C). max To achieve the lowest possible total carbon emissions (TCE) and further improve the quality of feasible scheduling solutions.

[0169] Step 17: Update the Pareto archive Ω *If the number of feasible solutions in the non-dominated solution set is greater than ps, then the crowding distance sorting criterion is used to sort all feasible scheduling solutions, while removing some solutions with the smallest crowding distance, until the number of feasible scheduling solutions in the population reaches ps.

[0170] The specific embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the present invention is not limited to the above embodiments. Within the scope of knowledge possessed by those skilled in the art, various changes can be made without departing from the spirit of the present invention.

Claims

1. A green optimization scheduling method based on the distributed assembly production process of automotive parts, characterized in that: First, a scheduling model for the processing and assembly of automotive parts is established, with the optimization objective of simultaneously minimizing the completion time and energy consumption of the automotive parts production process. Since each automotive part contains multiple different types of components, and each component has multiple processing steps, a hyperheuristic three-dimensional distribution estimation algorithm is proposed to optimize the solution based on the processing time of each process on each machine. (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) in, Binary symbols: represent arbitrary factories The machine in exist At all times with speed When in processing mode, assign a value of 1; when in standby mode, assign a value of 0. This indicates that they are jointly owned. There are several factories of the same type, and each factory includes both processing and assembly stages; Indicates shared ownership One part awaiting processing. This indicates the parts included in each accessory; Indicates the processing stage has Taiwan processing equipment; Representative at the factory accessories In the machine The actual processing speed of the machine when processing the first part; , which is a variable gear speed sequence; The model must satisfy the following constraints: (1) Each part can be assigned to any factory for processing; (2) Each factory contains at least one component to be processed; (3) In each assembly line workshop, all parts must be processed in sequence, that is, first in the equipment Processing on the equipment, and then on the machine Processing continues on the upper level, and so on, until the last piece of equipment. The processing is completed on the upper part; (4) Parts belonging to the same accessory must be processed continuously. Processing only a part of the accessory is not in line with the assembly conditions. (5) Factory accessories In the middle position The parts in the machine The operation on is represented as Once the operation Startup must be performed without interruption; For operation Processing time; (6) Once the accessories Once all parts are machined, they are immediately delivered to the assembly line in the same factory. Assembly can begin if all parts of the component have been processed and the assembly machine is idle. Once the assembly machine is started, it cannot be interrupted. Indicates accessories Assembly machine Assembly time on (7) Each machine can only perform one operation at a time; Formula (1) represents the factory The part in position 1 in the machine The required completion time during processing; Formula (2) represents the factory's... The Middle The completion time of parts at each location in machine 1; Formula (3) represents the factory's completion time. The first part in the machine The completion time; Formula (4) represents the factory's completion time. The Middle The parts in each position in the machine The completion time on the formula (5) represents the factory's completion time. In the middle position The completion time of the parts; Formula (6) represents the factory's completion time. Maximum completion time for components; Formulas (7) and (8) represent the maximum completion time, respectively. Total energy consumption , Representative operation In the machine Above speed During processing, the corresponding processing time is The corresponding power consumption per unit time is , This means the machine will only completely shut down once all operations are complete. Each machine will be in standby mode when not performing processing operations, and the power consumption per unit time is... ;in Total energy consumption, The carbon emissions per unit of energy consumption, represented by formula (11), indicate that the encoding of an individual in the problem domain consists of two parts: one part is the workpiece processing sequence. The other part is a workpiece processing speed comparison table. , in This indicates the number of processes to be performed on each machine. One operation, Indicates the first The operation is performed at the machine. The running speed on the above, for ease of symbol representation, allows for the use of feasible solutions. The corresponding completion time and total energy consumption are denoted as follows: and According to formulas (9) and (10), the only difference between carbon emissions and total energy consumption is a constant coefficient. or This will produce an equivalent result; To enhance the algorithm's local search capability, 12 low-level heuristic operations were designed, namely: LLH1: Randomly swap any two components within a critical factory; LLH2: Take a component from a critical factory and randomly insert it into another location; LLH3: Randomly reverses the order of parts within a critical factory; LLH4: Randomly swap parts between two factories; LLH5: Take a component from a critical factory and randomly insert it into another factory; LLH6: Randomly swap any two components within a critical factory; LLH7: Select an accessory from the critical factory, then select a part from the sequence of parts contained in that accessory, and randomly insert it into another position in the sequence; LLH8: Select any component from the critical factory, then select any two parts from the sequence of parts contained in that component, and randomly swap the positions of these two parts; LLH9: Select any component from the critical factory, then select a segment from the sequence of parts contained in that component, and reverse that segment; LLH10: Randomly select two machines to perform a sudden change in processing speed; LLH11: Randomly select two machines to perform a processing speed exchange operation; LLH12: Randomly select two machines to perform the processing speed insertion operation; The specific steps of the green optimization scheduling method based on the distributed assembly production process of automotive parts are as follows: Step 1: Determine the encoding and decoding method: Use a three-dimensional vector encoding method, namely factory sequence, machine sequence, and part sequence; the underlying problem domain individuals For the parts processing sequence, the decoded individual parts contain the parts sequence assigned to each factory. and the sequence of parts contained in each accessory Two parts; Step 2: Initialize the underlying problem domain population and the 3D probability model; set the population size to... , hereinafter referred to as ; Step 3: Design the low-level heuristic operation LLH; Step 4: Generate a high-level policy domain population by sampling feedback from the lower-level problem domain. The same population size is set to For high-level policy domains, each individual in the population It consists of 12 low-level heuristic operations. When decoding individuals in the high-level policy domain, the low-level heuristic operations in the high-level policy domain individuals are executed sequentially from left to right for the scheduling solutions in the low-level problem domain. After each low-level heuristic operation is executed, the new solution is compared with the original solution. If the new solution is better than the old solution, the new solution replaces the old solution; otherwise, the original solution is retained, and the remaining low-level heuristic operations are executed. After all low-level heuristic operations in the high-level policy domain individuals are executed, the fitness value of the high-level policy domain individuals is calculated. This fitness value is the optimal fitness value of the updated low-level problem domain individuals. Step 5: Use the operation selection function to evaluate the population in the lower problem domain. At this point, there are two positional situations: one is that the individual in the higher policy domain is not initially located at position 1, and the other is that the individual in the higher policy domain is initially located at position 1. Step 6: Decode all individuals in the high-level policy domain population and obtain a new solution based on the low-level heuristics in the high-level policy domain individuals; Step 7: Select a predetermined proportion of high-quality individuals from the high-level strategy domain population to update the probability model of the three-dimensional distribution estimation algorithm, and generate a new next-generation elite population through sampling. ; Step 8: Calculate the three-dimensional matrix and update the three-dimensional probability matrix model simultaneously; Step 9: Determine if the new solution is better than the old solution. If the termination condition is met, go to Step 4 if not. Otherwise, terminate the loop. Step 10: Based on the decoding method, obtain the processing time of the parts on the processing machine and the assembly time of the accessories on the assembly machine, obtain the initial processing sequence of the parts and accessories, and generate the corresponding scheduling Gantt chart; Step 11: Determine the critical path of the Gantt chart corresponding to the scheduling solution, where the critical path is the process path with the longest total processing time in the scheduling Gantt chart; at the same time, determine the critical path of each non-dominated solution in the non-dominated solution set. If there are multiple critical paths, randomly select one of them. Step 12: Perform four neighborhood searches based on the critical path in sequence for the steps in the non-dominated solution: exchange, insertion, reversal, and mutation. If the new solution is better than the old solution, replace the old solution with the new non-dominated solution, redetermine the critical path, and then perform the subsequent neighborhood search operations. If they are not mutually dominant, add the new solution to the non-dominated solution set, and continue the other neighborhood search operations on the current solution. Step 13: Repeat the operations of Step 11 and Step 12 until all the processing steps corresponding to the obtained non-dominated solutions cannot shorten the total processing time; Step 14: For each solution in the non-dominated solution set Calculate separately and Then we get ; Step 15: According to Sort all non-dominated solutions in ascending order, and then set the non-dominated solution set... Divided into and Two parts, of which ; Non-dominated solutions in The value is small, so the maximum completion time needs to be considered. Optimize, Non-dominated solutions in Large value, needs to be adjusted Optimize; Step 16: As shown in Step 15, processes outside the critical path are processes on the non-critical path. Based on the machine speed of the processes on the non-critical path, determine whether there is an allowable speed reduction margin; In the process, an energy-saving speed regulator strategy is applied to all non-dominated solutions. This means keeping the machining speed of all parts on the critical path constant while adjusting the machining speed of parts on non-critical paths, so that the maximum completion time remains the same. To the lowest possible This will further improve the quality of feasible scheduling solutions; Step 17: Update the Pareto Archive If the number of feasible solutions in the non-dominated solution set is greater than Then, the crowding distance sorting criterion is used to sort all feasible scheduling solutions, while some solutions with the smallest crowding distance are removed, until the number of feasible scheduling solutions in the population reaches a certain threshold. .

Images