A distributed flexible workshop scheduling method and system with transportation process

By combining artificial immune algorithms and per-neighborhood search in distributed flexible operation workshop scheduling, the transportation time and energy consumption between machine tools are optimized, and the problem of not considering transportation time in the existing technology is solved, and the effective solution of DFJSP is realized and performance improvement.

CN114519455BActive Publication Date: 2024-11-26SHANDONG NORMAL UNIV
View PDF 5 Cites 0 Cited by

Patent Information

Application Number
CN202210026297.3
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2022-01-11
Publication Date
2024-11-26
Estimated Expiration
2042-01-11

AI Technical Summary

Technical Problem

The existing distributed flexible operation workshop scheduling method does not consider the transportation time between machine tools, and the algorithm effectiveness in DFJSP has not been verified, and there is a lack of an algorithm to solve the transportation time with the belt.

Method used

A hybrid meta heuristic algorithm based on artificial immune algorithm and peri-neighborhood search is proposed. Through the weighted sum method, the completion time and total energy consumption are optimized, and the order-based mixed integer linear planning model is established, the transportation time is considered, and the three-dimensional encoding and Various initialization strategies to avoid local optimum.

Benefits of technology

The problem of transportation time constraints in DFJSP is effectively solved, the maximum completion time and total energy consumption are optimized, and the effectiveness and performance of the algorithm are verified by comparison with other algorithms.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN114519455B_ABST
    Figure CN114519455B_ABST
Patent Text Reader

Abstract

The present invention provides a distributed flexible workshop scheduling method and system with a transportation process, comprising: obtaining information of workpieces to be processed and information related to the distributed flexible workshop; obtaining a scheduling result according to the information of the workpieces to be processed, the information related to the distributed flexible workshop and a preset distributed flexible workshop scheduling model; wherein the distributed flexible workshop scheduling model is set as a sequence-based mixed integer linear programming model that simultaneously optimizes completion time and total energy consumption using a weighted sum method; when determining the transportation time between machine tools, a virtual machine tool is preset, and the process of the virtual machine tool conveying each workpiece is set as the first process; the present invention takes into account the typical constraint condition of the transportation time between machine tools, and simultaneously optimizes the completion time and total energy consumption in a multi-objective problem using a weighted sum method, and establishes a sequence-based MILP mathematical model for the first time.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention belongs to the technical field of workshop scheduling, and in particular relates to a distributed flexible workshop scheduling method and system with a transportation process. Background Art

[0002] In recent years, with the increasing popularity of collaborative production among enterprises, distributed manufacturing systems have received widespread attention. Since the distributed scheduling problem has important theoretical and application value, it has been expanded to multiple directions such as distributed parallel machine scheduling problem, distributed permutation flow operation scheduling problem and distributed hybrid job shop scheduling problem. The Distributed Flexible Job-shop Scheduling Problem (DFJSP) is a typical branch of the Distributed Job-shop Scheduling Problem (DSSP), which has the characteristics of multiple factories working at the same time, and each factory represents a flexible job shop. The DFJSP problem contains three-dimensional optimization problems, including workpiece allocation problem, machine tool allocation problem and process optimization problem.

[0003] The inventors found that the existing methods for solving the distributed flexible job shop scheduling problem do not take into account the transportation time factor of the workpiece between machine tools, and the effectiveness of the currently studied algorithms and other meta-heuristic algorithms in DFJSP has not been verified, and there is no algorithm for solving the DFJSP problem with transportation time. Summary of the invention

[0004] In order to solve the above problems, the present invention proposes a distributed flexible workshop scheduling method and system with transportation process. The present invention proposes a hybrid metaheuristic algorithm based on artificial immune algorithm (AIA) and variable neighborhood search (VNS) to solve the DFJSP problem with transportation time (DFJSP with Transportation time, DFJSP-T).

[0005] In order to achieve the above object, the present invention is implemented through the following technical solutions:

[0006] In a first aspect, the present invention provides a distributed flexible workshop scheduling method with a transportation process, comprising:

[0007] Obtain information about workpieces to be processed and related information of distributed flexible workshops;

[0008] The scheduling result is obtained according to the information of the workpiece to be processed, the related information of the distributed flexible workshop and the preset distributed flexible workshop scheduling model;

[0009] Among them, the distributed flexible workshop scheduling model is set as a sequence-based mixed integer linear programming model that uses the weighted sum method to simultaneously optimize the completion time and total energy consumption; when determining the transportation time between machine tools, a virtual tool is preset, and the process of the virtual tool conveying each workpiece is set as the first process. The completion time of each subsequent process is equal to the completion time of the previous process plus the processing time of the subsequent process and the transportation time between the two machine tools, and the maximum completion time is obtained from the last process of all operations.

[0010] Furthermore, the distributed flexible workshop scheduling model is: minω·C max +(1-ω)·TEC, where ω is the weight of the target, C max is the maximum completion time obtained from the last process, and TEC is the total energy consumption.

[0011] Furthermore, the total energy consumption TEC is:

[0012]

[0013] Where n is the number of workpieces, op j is the number of processes for each workpiece j, u is the number of factories, m is the number of machine tools in each factory, j is the workpiece number, l is the process number, i is the machine tool number, f is the factory number, and X j,l,f,i,k is a binary decision variable, pt j,l,f,i For process O j,l Machining time on machine tool i in factory f, ec j,f,i is the unit energy consumption of machining workpiece j on machine tool i in factory f.

[0014] Furthermore, a hybrid artificial immune algorithm is used to solve the distributed flexible shop scheduling model, including the workpiece allocation vector, machine tool selection vector and process sequence vector.

[0015] Furthermore, the solution using the hybrid artificial immune algorithm at least includes initializing the population, recording the antibody with the highest affinity, evaluating the fitness value of individuals in the population by decoding, selecting the antibody with the highest fitness in the population and cloning it, updating the search strategy for the cloned antibody, and further updating the population through the immune suppression process.

[0016] Furthermore, if the operation to be processed is the first operation of the workpiece, the transportation time from the virtual machine tool to the current machine tool is re-analyzed, otherwise, various situations from one machine tool to another are considered simultaneously.

[0017] Furthermore, a neighborhood-based search strategy is used to update the search strategy of the cloned antibodies. The global search is performed on the job assignment vector. In the inner loop, the neighborhood structure is used to perform local search in each neighborhood structure, which is used for the process sequence vector and the machine tool selection vector respectively. For each cloned antibody, a local search operator is applied to obtain a solution, and the obtained solution is inserted into the workpiece assignment vector. If no satisfactory solution is found in the neighborhood structure, the global neighborhood structure is replaced and the local search is continued in the new neighborhood structure.

[0018] In a second aspect, the present invention further provides a distributed flexible workshop scheduling system with a transportation process, comprising:

[0019] The data acquisition module is configured to: obtain information of workpieces to be processed and information related to the distributed flexible workshop;

[0020] The scheduling module is configured to: obtain a scheduling result according to information of workpieces to be processed, information related to the distributed flexible workshop, and a preset distributed flexible workshop scheduling model;

[0021] Among them, the distributed flexible workshop scheduling model is set as a sequence-based mixed integer linear programming model that uses the weighted sum method to simultaneously optimize the completion time and total energy consumption; when determining the transportation time between machine tools, a virtual tool is preset, and the process of the virtual tool conveying each workpiece is set as the first process. The completion time of each subsequent process is equal to the completion time of the previous process plus the processing time of the subsequent process and the transportation time between the two machine tools, and the maximum completion time is obtained from the last process of all operations.

[0022] In a third aspect, the present invention further provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the distributed flexible workshop scheduling method with a transportation process described in the first aspect.

[0023] In a fourth aspect, the present invention further provides an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the steps of the distributed flexible workshop scheduling method with a transportation process described in the first aspect are implemented.

[0024] Compared with the prior art, the present invention has the following beneficial effects:

[0025] 1. This paper considers the typical constraint condition of transportation time between machine tools in DFJSP, and uses the weighted sum method to simultaneously optimize the completion time and total energy consumption in the multi-objective problem; based on different modeling ideas, a sequential MILP (Mixed Integer Linear Programming) mathematical model is established for the first time, and its effectiveness and feasibility are verified through a model example;

[0026] 2. Taking into account the complexity and uncertainty of the problem, the present invention adopts a three-dimensional coding method and a variety of initialization strategies to initialize the population; on this basis, an artificial immune algorithm and a hybrid algorithm based on variable neighborhood search are proposed to optimize the problem, and multiple search operators based on the problem are used to strengthen the solution to the problem; at the same time, a variety of neighborhood structures are combined to ensure that the algorithm will not fall into the local optimum; finally, by comparing with four other effective metaheuristic algorithms, the algorithm achieved better performance, verifying the effectiveness of the proposed algorithm. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] The drawings in the specification that constitute a part of this embodiment are used to provide a further understanding of this embodiment. The schematic embodiments of this embodiment and their descriptions are used to explain this embodiment and do not constitute improper limitations on this embodiment.

[0028] Figure 1 This is a schematic diagram of the distributed production scheduling layout of Example 1 of the present invention;

[0029] Figure 2 is a crossover operator POX operator in Embodiment 1 of the present invention;

[0030] Figure 3 is the crossover operator RPX operator of embodiment 1 of the present invention;

[0031] Figure 4 is a crossover operator TPX operator in Embodiment 1 of the present invention;

[0032] Figure 5 ANOVA value analysis of three initialization strategies in Example 1 of the present invention;

[0033] Figure 6 ANOVA value analysis of HAIA-NS and HAIA in Example 1 of the present invention;

[0034] Figure 7 This is the ANOVA value analysis of the five algorithms when ω=0.8 in Example 1 of the present invention;

[0035] Figure 8 This is the ANOVA value analysis of the five algorithms when ω=0.2 in Example 1 of the present invention;

[0036] Fig. 9 This is the ANOVA value analysis of the five algorithms when ω=0.5 in Example 1 of the present invention;

[0037] Fig.10 It is the convergence curve corresponding to the example-F2J10M5 of embodiment 1 of the present invention;

[0038] Fig.11 It is the convergence curve corresponding to the example-F2J15M5 of embodiment 1 of the present invention;

[0039] Fig.12 It is the convergence curve corresponding to the example-F2J20M3 of embodiment 1 of the present invention;

[0040] Fig.13 It is the convergence curve corresponding to the example-F3J20M5 of embodiment 1 of the present invention;

[0041] Fig.14 It is the convergence curve corresponding to the example-F2J40M3 of embodiment 1 of the present invention;

[0042] Fig.15 It is the convergence curve corresponding to the example-F3J40M5 of embodiment 1 of the present invention;

[0043] Fig.16 It is the convergence curve corresponding to the example-F3J60M5 of embodiment 1 of the present invention;

[0044] Fig.17 This is the convergence curve corresponding to the example F2J100M3 of embodiment 1 of the present invention. DETAILED DESCRIPTION

[0045] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0046] It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present application. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present application belongs.

[0047] Embodiment 1:

[0048] like Figure 1 As shown, this embodiment provides a distributed flexible workshop scheduling method with a transportation process, including:

[0049] Obtain information about workpieces to be processed and related information of distributed flexible workshops;

[0050] The scheduling result is obtained according to the information of the workpiece to be processed, the related information of the distributed flexible workshop and the preset distributed flexible workshop scheduling model;

[0051] Among them, the distributed flexible workshop scheduling model is set as a sequence-based mixed integer linear programming model that uses the weighted sum method to simultaneously optimize the completion time and total energy consumption; when determining the transportation time between machine tools, a virtual tool is preset, and the process of the virtual tool conveying each workpiece is set as the first process. The completion time of each subsequent process is equal to the completion time of the previous process plus the processing time of the subsequent process and the transportation time between the two machine tools, and the maximum completion time is obtained from the last process of all operations.

[0052] This embodiment studies the distributed flexible job shop scheduling problem with transportation time constraints (DFJSP-T), and the goal is to optimize the maximum completion time and total processing energy consumption at the same time; for this problem, this embodiment establishes a sequence-based mixed integer linear programming model. Considering the complexity and uncertainty of the problem, a three-dimensional coding method and multiple initialization strategies are used to initialize the population. On this basis, an artificial immune algorithm and a hybrid algorithm based on variable neighborhood search are proposed to optimize the problem, and multiple search operators based on the problem are used to strengthen the solution of the problem. At the same time, a variety of neighborhood structures are combined to ensure that the algorithm will not fall into the local optimum. Finally, by comparing with four other effective metaheuristic algorithms, the algorithm achieves better performance, verifying the effectiveness of the proposed algorithm; the specific process is as follows:

[0053] DFJSP is an extension and expansion of the Flexible Job-shop Scheduling Problem (FJSP), that is, a group of workpieces are assigned to multiple factories for processing. In this embodiment, when the workpiece is processed on the machine, the transportation time and the total energy consumption between the machines are considered at the same time. The weighted sum method is used to solve the multi-objective optimization problem, and multiple groups of different weights are assigned to optimize the two objectives. In addition, the process of the workpiece from the buffer to the machine tool is taken into account, and a virtual machine tool 0 is defined.

[0054] DFJSP consists of a set of artifacts {J1, J2, ..., J n} In multiple factories {F1, F2, ..., F u}, where each factory is the same; for each workpiece O j,l , there is a set of machine tools available Can be processed; once a process O j,l The processing is completed and the next process O j,l+1 The processing machine tool and the current process O j,l If the processing machine tools are different, workpiece j will be immediately transported from machine tool i' to machine tool i, and the transportation time is recorded as tt j,i’,i; At the same time, the energy consumption generated when the workpiece is processed on the machine tool is taken into account and each process O j,l The energy consumption is affected by the processing time pt j,l,i The energy consumption per unit time is recorded as ec j,i In this example, it is assumed that: (1) all factories are identical, such as the number of machines; (2) each workpiece can only be assigned to a specific factory; (3) all processes for each workpiece must be performed in the same factory; (4) each job can only be processed on one computer in a specified factory at the same time; (5) each machine cannot perform multiple processes at the same time; (6) except for virtual machine tool 0, all other machines in all factories are available at time 0; (7) each factory has sufficient transportation equipment to transfer work from one machine to another; and (8) only the energy consumption generated when the workpiece is processed in the machine tool is considered.

[0055] In this embodiment, various subscripts, sets, constants and variables are defined, as shown in Table 1:

[0056] Table 1 Parameters and definitions

[0057]

[0058]

[0059] In this embodiment, in order to solve DFJSP-T, a sequential MILP model is established; the optimization objective of DFJSP-T is defined as follows:

[0060] minω·C max +(1-ω)·TEC (1),

[0061] Among them, ω is the weight value of the target; the model requires 4 decision variables, among which Z j,f To assign each job to the appropriate factory, C j,l Used to define the completion time of each process, Y j,l,j’,l’ Used to determine the processing order of the process, and X j,l,f,i,i’ Ensure that each process is assigned to a machine tool in the factory. For this model, the total energy consumption (TEC) is calculated as follows:

[0062]

[0063] The constraints are defined as follows:

[0064]

[0065]

[0066]

[0067]

[0068]

[0069]

[0070]

[0071]

[0072]

[0073]

[0074]

[0075]

[0076]

[0077] Z j,f ,X j,l,f,i,i' ,Y j,l,j',l' ∈{0,1} (16)

[0078] Among them, constraint (3) aims to assign each job to a specific factory; constraints (4) and (5) ensure that all operations of each job are processed in the same factory; constraints (6) and (7) restrict each operation to select a processing machine from a set of available machine tools; constraint (8) ensures that except for the first operation of each job, each operation is executed in sequence and that if O is processed on machine i in factory f, j,l , then the previous process O j,l-1 The processing must be performed on machine i' in the same factory; since virtual machine 0 is considered, constraint (9) forces the first process of each workpiece to be delivered from virtual machine M0; in order to calculate the completion time, constraints (10) and (11) are summarized to define that the completion time of each process is equal to the completion time of the previous process plus the processing time of this process and the transportation time from machine i' to machine i; constraints (12) and (13) ensure that there is no overlap between all processes; constraint (14) determines the maximum completion time obtained from the last process of all jobs; finally, the value ranges of the four decision variables are declared in constraints (15) and (16).

[0079] In this embodiment, a hybrid artificial immune algorithm (HAIA) algorithm for solving the DFJSP-T problem under consideration is introduced. The main framework of the proposed HAIA is described in detail in Algorithm 1, as shown in Table 2:

[0080] Table 2 Algorithm 1: Hybrid artificial immune algorithm

[0081]

[0082] In this embodiment, the solution representation can be divided into three parts, namely, the job assignment (JA) part, the machine selection (MS) part and the operation sequence (OS) part; the scale of the solution is Among them, op j represents the number of processes for each workpiece j, and n represents the number of workpieces; the specific expression of the solution is as follows:

[0083] (1) Job assignment vector: This vector determines the assignment of each job to a specific factory, where the size of the vector is equal to the number of jobs and each number represents the ordinal number of the factory. For example, the first job is assigned to the first factory, and the remaining two jobs are assigned to the second factory.

[0084] (2) Machine selection vector: This vector ensures that each process is assigned to an available machine tool for execution, and the vector size is equal to Each number in the solution represents the serial number of the machine tool;

[0085] (3) Process sequence vector: This vector guarantees a feasible process sequence, where each number represents a workpiece, and the order in which each workpiece appears determines its process order. j times, which means that the artifact includes op j A process.

[0086] In order to achieve better convergence and diversity of the population, several initial strategies are used to initialize the solution; the initial strategies of the MS part are as follows:

[0087] (1) Random allocation strategy: For each process, an available machine tool is randomly selected for processing; this method has good randomness and maintains the diversity of the population;

[0088] (2) Minimum processing time strategy: For each process, a suitable machine tool with the minimum processing time is selected from a group of available machine tools for processing;

[0089] (3) Run the lowest energy consumption strategy: For each process, select a suitable machine tool with the lowest energy consumption from a group of available machine tools for processing;

[0090] (4) Global minimum fitness strategy: This strategy is an extension of the global minimum processing time strategy, which converts the processing time into the weighted value of multiple objectives; in this way, the generated solution can converge according to the weight coefficients we set as desired; the detailed steps are described in Algorithm 2 and shown in Table 3;

[0091] (5) Global-local minimum fitness strategy: This strategy is an extension of the global minimum fitness strategy, which separates each job and applies the global minimum fitness optimization to each job to select the appropriate machine tool.

[0092] Table 3 Global minimum fitness strategy

[0093]

[0094] In this embodiment, the initialization strategy of the OS part is described in detail as follows:

[0095] (1) Random sorting strategy: This rule is a simple and feasible method that randomly disrupts all processes to maintain the diversity of the population and randomly searches the entire solution space;

[0096] (2) Most number of operations remaining (MOR) strategy: The workpiece with the most remaining operations is preferably arranged in the operation sequence vector; if there are multiple workpieces with the same number of operations, a workpiece is randomly selected.

[0097] Since this problem involves transportation time and virtual machine tools, it is different from the decoding process in traditional FJSP; considering the current process to be processed, whether it is the first process or not, it can be divided into two categories; if the process to be processed is the first process of the workpiece, it is necessary to re-analyze the transportation time from virtual machine tool 0 to the current machine tool; otherwise, it is necessary to consider various situations from one machine tool to another. The entire decoding process is shown in Table 4:

[0098] Table 4 Decoding process

[0099]

[0100]

[0101] In this algorithm, in order to ensure that better antibodies are obtained and applied to subsequent immune operations such as cloning and immunosuppression, several antibodies with the highest affinity are selected, where the number of selected antibodies is represented by N. C For cloning operations, there are more antibody clones with high antibody affinity. Therefore, for each selected N C Antibody, clone number is defined as: N C-k+1, where k represents N C The kth antibody with higher affinity among the antibodies. In this way, the total number of clones after the cloning operation is equal to Nc(Nc+1) / 2.

[0102] like Figure 2 , Figure 3 and Figure 4 As shown in the figure, according to the characteristics of each part of the encoding scheme, the precedence operation crossover (POX) operator, the two-points crossover (TPX) operator and the random-point preservation crossover (RPX) operator are respectively used; it is worth noting that none of these three operators requires a repair mechanism; in the crossover link, the selection of parental individuals is a key factor affecting the quality of offspring individuals; therefore, one of the parents is selected from the cloned antibody vector CA, denoted as P1, and the other is the individual with the highest affinity selected from three random individuals by the binary tournament selection method, denoted as P2; in terms of elite selection, of the two newly generated offspring individuals O1 and O2, the better one is retained for subsequent operations.

[0103] In this embodiment, the POX operator is represented as LS1; the POX operator generates two offspring individuals by crossover operation of OS vectors of two parent individuals. All jobs are divided into two groups JSet1 and JSet2; jobs belonging to set JSet1 are copied from P1 to O1, and jobs belonging to set JSet2 are copied from P2 to O2; the blank positions in O1 and O2 are filled in the order of uncopied jobs in P2 and P1.

[0104] In this embodiment, the RPX operator is represented as LS2; the MS vector adopts the RPX operator, which works as follows: randomly generate a string consisting of 0s and 1s, whose length is equal to the length of the MS vector; copy the parents P1 and P2 to the two child individuals O1 and O2 respectively; if the number of the specified position is 0, replace the machine at that position.

[0105] In this embodiment, the TPX operator is represented as GS1; a new JA vector is generated by two-point crossover. The newly generated job combination can be searched in a global range to avoid falling into the local optimum; two points C1 and C2 in the JA vector are randomly selected; the parents P1 and P2 are copied to the two child individuals O1 and O2 respectively; the jobs before point C1 and after point C2 in the JA vector are exchanged.

[0106] The mutation operation can effectively search the solution space by changing the neighborhood structure to achieve the effect of optimizing the target value, which is an effective method to improve the quality of the solution. In the embodiment, two mutation methods are designed to optimize the solution and complete the search of the solution space. At the same time, two properties of DFJSP are given and applied to the design of the mutation operator. The two properties are explained below:

[0107] Property 1. In DFJSP, each factory can be regarded as an independent FJSP. Proof: In the encoding stage, each workpiece is randomly assigned to a factory, and all processes of each workpiece are processed in the same factory, and the transportation of workpieces between factories is not considered. Therefore, for each factory, there is a set of workpieces processed in the factory, and each process of each workpiece has a set of machine tools available for processing, which is consistent with the definition of FJSP, so each factory can be regarded as a separate FJSP.

[0108] Property 2: If only the process sequence or machine tool selection of the key factory is changed, the completion time of other factories will not be affected. Proof: According to Property 1, each factory can be regarded as an independent FJSP, so each factory has an independent completion time. Therefore, the completion time can be further expressed as: C max =max{C F1 , C F2 , …, C Fu}, where C Ff represents the completion time of the fth factory. Obviously, each factory has a separate coding scheme π f , which consists of two parts, namely the operation sequence part and the machine tool selection part. The coding scheme of DFJPS can be expressed as π=π1∪π2∪…∪π u In short, once the key factory π c The encoding scheme changes, other schemes π f =π-π c The coding scheme will not be affected, nor will the completion time of other factories.

[0109] Based on property 1, mutation operators are divided into local search operators and global search operators, where global search is performed on the job assignment vector and local search is performed on the process sequence vector and machine selection vector. Since the completion time criterion of the key factory determines the maximum completion time of DFJSP, we propose three mutation operators for the key factory, including two local search operators based on intra-factory mutation and a global search operator for inter-factory workpiece exchange. In addition, based on properties 1 and 2, after the mutation operator is applied to the key factory, there is no need to decode other factories again. Therefore, the algorithm complexity of the decoding part is reduced from O(n) to O(1), where n represents the number of factories.

[0110] Based on the above properties, several global search operators and local search operators are proposed to solve this problem. Since the maximum completion time is determined by the critical path of the solution, the maximum completion time can only be optimized by adjusting the operations on the critical path. To this end, two mutation operators based on the critical path are used to optimize the solution. The mutation operators are as follows:

[0111] In the embodiment, two mutation methods are designed to optimize the solution and complete the search of the solution space. One is based on random search operators, including LS3, LS4 and GS2. In addition, since the completion time of the key plant determines the maximum completion time of the project, the maximum completion time can only be optimized by adjusting the operations on the critical path. Therefore, we propose a local search operator based on the key plant, including LS5, LS6 and GS4.

[0112] Among them, random process permutation operator (LS3): generates a new neighborhood solution by exchanging the elements at two randomly selected and unequal positions in the OS vector. Random machine tool replacement operator (LS4): obtains a neighborhood solution by replacing the processing machine tool of any process. Key process exchange operator (LS5): similar to LS3, it generates a new neighborhood solution by exchanging two processes in the key factory. Key machine tool exchange operator (LS6): similar to LS4, it generates a new neighborhood solution by replacing the machine tool of the process in the key factory. Random workpiece exchange operator (GS2): the reasonable combination of jobs has a crucial influence on the convergence of the solution. On this basis, a random job exchange operator is designed to achieve global mutation. The specific operation can be attributed to exchanging randomly selected jobs between the two factories f1 and f2 in the JA vector. This operator ensures that the global search is carried out in a random direction and completes the traversal of the search space. Random workpiece insertion operator (GS3): In order to prevent uneven job distribution, a random workpiece insertion operator is proposed to balance the job distribution between factories. The details are as follows: (1) Find the factory f with the most and the least processes max and f min ; (2) in f max (3) Change the processing factory of workpiece j to f in the JA vector. min Critical workpiece insertion operator (GS4): In order to avoid uneven processing time of factories, a critical workpiece insertion operator is proposed to balance the maximum completion time and the minimum completion time of the two factories. The details are as follows: (1) Find the factory f with the longest and shortest completion time max , f min ; (2) Find the job j with the longest completion time in the key factory; (3) Change the factory of job j to f in the JA vector min .

[0113] In order to give full play to the potential of the population and explore the solution space, a search strategy based on the variable neighborhood search (VNS) algorithm is adopted for further exploration; VNS is an efficient local search method and has been widely proven and applied to various combinatorial optimization problems; generally speaking, VNS involves two neighborhood operations and multiple neighborhood structures, and the optimal solutions obtained by two different neighborhood structures are often different; therefore, once the algorithm falls into the local optimal trap, it can be avoided by switching the neighborhood structure through two neighborhood operation systems.

[0114] In this embodiment, a set of neighborhood structures GS k (k=1,2,…,k max ) to replace the neighborhood structure; it is worth noting that these three operators are developed for the JA vector, which means that the global search is performed on the job assignment vector. In the inner loop, several neighborhood structures LS l (l=1,2,…,l max ) is used to perform local search in each neighborhood structure, where these operators are used for OS vectors and MS vectors respectively; for each cloned antibody, several local search operators are applied to obtain a better solution, and the solution is inserted into the CA vector; if no better solution is found in these neighborhood structures, the global neighborhood structure is replaced and the local search is continued in the new neighborhood structure. The framework structure of the search strategy based on VNS is shown in Table 5:

[0115]

[0116]

[0117] The inhibition process is used as a protection mechanism to eliminate low-affinity antibodies and ensure the efficiency of population evolution; after the cloning and mutation process, all parent antibodies and progeny antibodies are retained in a temporary antibody set CA; the inhibition process is to reselect a group of high-quality antibodies from the CA vector to suppress the iteration of low-fitness antibodies in the population, among which the N with the highest affinity is retained. C antibodies, eliminating another group of the worst antibodies, thereby renewing the population.

[0118] In the embodiment, the algorithm proposed above is experimented and the experimental part is analyzed, specifically:

[0119] In order to verify the effectiveness of the proposed algorithm on DFJSP-T, in this embodiment, the algorithm was written in C++ language, and experimental analysis was carried out on a host configured with 3.40GHz Intel Core i7 CPU and 16GB RAM; it should be noted that the experiments, including the verification of the effectiveness of the strategy and model, were carried out when the weight coefficient was equal to 0.8; in the multi-algorithm comparison part, several groups of weight coefficients of 0.2, 0.5 and 0.8 were analyzed and compared.

[0120] Since there is no data set containing DFJSP-T transportation time, this embodiment randomly generates two groups of DFJSP-T examples with different problem scales according to the actual production problems, including small-scale examples and large-scale examples. Each example is divided into three parts. The first number represents the number of factories, with a value of f = {2, 4}; the second number represents the number of workpieces, with a value range of 5-100, where the value of the small-scale example is n = {5, 10, 15, 20}, and the value of the large-scale example is n = {20, 40, 60, 80, 100}. The last number represents the number of machine tools, and the number of machine tools is m = {3, 5}. The number of processes for each workpiece is randomly determined, with a value range of m / 2 to m, and the processing time of each process, the transportation time between different machine tools, the machine utilization rate of each workpiece, and the unit energy consumption are designed. The processing time and transportation time are evenly distributed in the intervals [10, 30] and [5, 20], respectively. The number of available machine tools for each process is determined in the interval [m / 2, m-1], taking into account special factors such as machine failures, preventive maintenance, etc. The discrete uniform distribution in the interval [0.1, 0.5] gives the specific energy consumption when machining a workpiece.

[0121] In order to verify the effectiveness of the MILP model, this embodiment built a model in IBM ILOG CPLEX 12.7, and conducted a comparative experiment between the MILP model and the proposed algorithm on a set of small-scale instances. Since CPLEX uses a branch-and-bound algorithm when solving the mixed integer programming problem MILP, the model takes a long time to calculate. Therefore, the calculation time of the model is limited to 1000 seconds, and the number of threads is set to 3. It is worth noting that the fitness value of the algorithm is the minimum value after each independent run for 30 seconds.

[0122] As shown in Table 6, the scale of each example is given in the first column, and the symbols 'F', 'J' and 'M' represent the number of factories, workpieces and machine tools, respectively. The third column lists the best fitness values ​​between the two models and the proposed algorithm. For each model, the fitness value, the number of solutions and the relative target gap (Gap) are compared in this embodiment, where Gap represents the difference between the optimal integer solution obtained within a certain period of time and the optimal solution that may exist in the solution space. It is worth noting that if the Gap value is equal to 0, the solution is the best individual in the solution space. The formula for calculating the Gap indicator is as follows:

[0123]

[0124] Among them, F integer represents the optimal integer solution obtained within a specific time, and F Best Represents the optimal solution that may exist in the solution space. At the same time, in order to compare the fitness values ​​of the model and algorithm, the relative percentage increase (RPI) is used as the performance metric, and its calculation formula is as follows:

[0125]

[0126] Among them, F current is the fitness value obtained by the given algorithm, and F Best It is the best fitness value among all compared models and algorithms.

[0127] Table 6 Performance comparison between MILP model and HAIA

[0128]

[0129]

[0130] From Table 6, we can see that both the model and the proposed algorithm can complete the search space search and find the optimal solution in a short time. However, as the instance size gradually increases, the model constraints and variables also increase. The MILP model cannot find the optimal solution in a reasonable time. From the last row, we can see that the average RPI value of the algorithm is much smaller than the RPI value of the sequential-based model.

[0131] In order to verify the effectiveness of the proposed initialization strategy, two other comparison algorithms are designed in this embodiment. Their initialization strategies both adopt the corresponding parts of the improved genetic algorithm proposed by Giovanni and Pezzella; (hereinafter referred to as IGA), and the ABC algorithm proposed by Wu et al., which are represented by HAIA-DI1 and HAIA-DI2 respectively.

[0132] The comparison results are shown in Table 7, where the second column lists the instance scale and the third column provides the best fitness value among the three algorithms. The fitness and RPI values ​​of the three algorithms with different initial strategies are shown in the following six columns. It can be seen from Table 7 that as the instance scale increases, the performance of the algorithm becomes more and more prominent. The last row calculates the average RPI value, which shows that the proposed initial strategy achieves better performance than the other two initial strategies. In order to intuitively show the differences between the three algorithms, multi-factor analysis of variance (ANOVA) was used to evaluate the performance of the algorithms. Figure 5 As shown, since the p-value is equal to 8.4166e-06 and is much smaller than 0.05, the difference is proven to be significant.

[0133] Table 7 Performance comparison of three initialization strategies

[0134]

[0135]

[0136] Faced with a complex solution space, the encoding scheme is divided into three parts, among which the workpiece allocation JA vector is the key part to determine the solution quality. To this end, a VNS-based search strategy is proposed to change the neighborhood structure of the JA vector. In order to verify the effectiveness of the VNS-based search strategy, a comparison algorithm HAIA-NS without a search strategy is designed in this embodiment, where the variable k max It is equal to zero, indicating that there is no neighborhood handover in the JA part.

[0137] The calculation results are shown in Table 8, where the second column indicates the scale of the instance and the third column shows the best value obtained between HAIA-NS and HAIA. The next two columns describe the average values ​​of the two algorithms after 30 independent runs, and the last two columns list the RPI values ​​of the two algorithms respectively. The results show that the algorithm using the VNS-based search strategy obtains better results in 15 of the given 16 instances; the last row calculates the average RPI values ​​of the two compared algorithms to be 1.73 and 0.04, respectively, indicating that the VNS-based search strategy is effective in improving the algorithm.

[0138] Figure 6 An ANOVA comparison of the two algorithms is provided, where the p-value is equal to 7.0034e-07, which is much less than 0.05. Therefore, there is a significant difference between the two algorithms, further demonstrating the effectiveness of the strategy.

[0139] Table 8. Performance comparison of HAIA-NS and HAIA

[0140]

[0141] In order to verify the effectiveness of the algorithm, four other algorithms were used as comparison algorithms, including IGA, ABC, hybrid genetic algorithm (hereinafter referred to as HGA) proposed by Chang et al., and enhanced memetic algorithm (MA) proposed by Gong et al. (hereinafter referred to as EMA). The study has proved the effectiveness of these four algorithms in solving the DFJSP problem. In order to avoid the randomness of algorithm operation and fair comparison, each algorithm in each instance was run independently 30 times, and the average values ​​were taken for comparison.

[0142] In order to verify the effectiveness of the algorithm, four other algorithms were used as comparison algorithms, including IGA, ABC, hybrid genetic algorithm (hereinafter referred to as HGA) proposed by Chang et al., and enhanced memetic algorithm (MA) proposed by Gong et al. (hereinafter referred to as EMA). The study has proved the effectiveness of these four algorithms in solving the DFJSP problem. In order to avoid the randomness of algorithm operation and fair comparison, each algorithm in each instance was run independently 30 times, and the average values ​​were taken for comparison. Two sets of examples of different sizes were used to verify the effectiveness of the algorithm, and the performance of the algorithm under different weight coefficients was analyzed.

[0143] The comparison results of multiple algorithms under different weight coefficients are shown in Tables 9, 10 and 11. Table 9 provides a multi-algorithm performance comparison with a weight coefficient equal to 0.8, where the first column indicates the instance size. In order to be able to analyze the best convergence values ​​of other algorithms when the weight coefficient ω = 0.8 and compare them with the results obtained by CPLEX, the minimum value of each algorithm is also listed in each column. The next ten columns respectively represent the average values ​​obtained from 30 runs of each comparison algorithm, and the last five columns are the RPI values ​​obtained based on the average values ​​of the five comparison algorithms. In the other two tables, the third column gives the best value obtained from the average values ​​of the following five comparison algorithms, and the last five columns give the RPI values ​​of these five comparison algorithms. The other two Tables 10 and 11 show the multi-algorithm performance comparison with weight coefficients equal to 0.2 and 0.5, respectively.

[0144] According to the comparison results, we can see that: in the given 32 instances, HAIA achieves better results. As the instance size increases, the superiority of the algorithm becomes more prominent. From the last row of the three tables, the RPI value of the algorithm is much smaller than that of the other four algorithms. In addition, Figure 7 Figure 8 and Fig. 9 The variance analysis comparison results of various algorithms are given. Under different weight coefficient settings, the algorithm proposed in this embodiment is better than the other four algorithms, and the significance of the difference between the compared algorithms is far less than 0.05.

[0145] Table 9 Comparative analysis of multiple algorithms when weight coefficient ω = 0.8

[0146]

[0147]

[0148] Table 10 Comparative analysis of multiple algorithms when weight coefficient ω = 0.2

[0149]

[0150]

[0151] Table 11 Comparative analysis of multiple algorithms when weight coefficient ω = 0.5

[0152]

[0153]

[0154] In order to compare the results intuitively, eight instances are randomly selected in this embodiment, including instance-F2J10M5, instance-F2J15M5, instance-F2J20M3, instance-F3J20M5, instance-F2J40M3, instance-F3J40M5, instance-F3J60M5 and instance-F2J100M3, and their corresponding convergence curves are shown as follows: Fig.10 , Fig.11 , Fig.12 , Fig.13 , Fig.14 , Fig.15 , Fig.16 and Fig.17 shown.

[0155] From the comparison results, HAIA has better performance than other efficient metaheuristic algorithms. The reasons can be summarized as follows: the various initial strategies proposed play a vital role in initializing a better population and exploring the solution space, especially in large-scale instances. Combining the two characteristics of the problem, several problem-related crossover and mutation operators are proposed to optimize the problem. A VNS-based search strategy is designed to find the optimal solution. This strategy ensures that the algorithm will not fall into the local optimum by continuously updating the neighborhood structure.

[0156] This embodiment considers a typical constraint in DFJSP, namely the transportation time between machine tools, and uses the weighted sum method to simultaneously optimize the completion time and total energy consumption in the multi-objective problem; based on different modeling ideas, a sequence-based MILP mathematical model is established for the first time, and its effectiveness and feasibility are verified through a small-scale example of CPLEX; a three-dimensional coding method is designed to initialize and decode the problem; combined with a VNS-based search strategy, a hybrid artificial immune algorithm is proposed to solve the problem, which integrates multiple problem-based crossover and mutation operators; the effectiveness of the initial strategy and the VNS-based strategy is verified through a small-scale example; in addition, the proposed algorithm has also achieved outstanding results in large-scale examples, further proving the effectiveness of the proposed strategy and algorithm.

[0157] Embodiment 2:

[0158] This embodiment provides a distributed flexible workshop scheduling system with a transportation process, including:

[0159] The data acquisition module is configured to: obtain information of workpieces to be processed and information related to the distributed flexible workshop;

[0160] The scheduling module is configured to: obtain a scheduling result according to information of workpieces to be processed, information related to the distributed flexible workshop, and a preset distributed flexible workshop scheduling model;

[0161] Among them, the distributed flexible workshop scheduling model is set as a sequence-based mixed integer linear programming model that uses the weighted sum method to simultaneously optimize the completion time and total energy consumption; when determining the transportation time between machine tools, a virtual tool is preset, and the process of the virtual tool conveying each workpiece is set as the first process. The completion time of each subsequent process is equal to the completion time of the previous process plus the processing time of the subsequent process and the transportation time between the two machine tools, and the maximum completion time is obtained from the last process of all operations.

[0162] The working method of the system is the same as the distributed flexible workshop scheduling method with transportation process in Example 1, and will not be repeated here.

[0163] Embodiment 3:

[0164] This embodiment provides a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the steps of the distributed flexible workshop scheduling method with a transportation process described in Embodiment 1 are implemented.

[0165] Embodiment 4:

[0166] This embodiment provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, the steps of the distributed flexible workshop scheduling method with a transportation process described in Embodiment 1 are implemented.

[0167] The above description is only a preferred embodiment of the present embodiment and is not intended to limit the present embodiment. For those skilled in the art, the present embodiment may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present embodiment shall be included in the protection scope of the present embodiment.

Claims

1. A distributed flexible workshop scheduling method with transportation process, characterized in that: include: Obtain information about workpieces to be processed and related information of distributed flexible workshops; The scheduling result is obtained according to the information of the workpiece to be processed, the related information of the distributed flexible workshop and the preset distributed flexible workshop scheduling model; The distributed flexible workshop scheduling model is set as a sequence-based mixed integer linear programming model that uses a weighted sum method to simultaneously optimize the completion time and total energy consumption; when determining the transportation time between machine tools, a virtual machine tool is preset, and the process of the virtual machine tool conveying each workpiece is set as the first process. The completion time of each subsequent process is equal to the completion time of the previous process plus the processing time of the subsequent process and the transportation time between the two machine tools, and the maximum completion time is obtained from the last process of all operations; A hybrid artificial immune algorithm is used to solve the distributed flexible shop scheduling model, including workpiece allocation vector, machine tool selection vector and process sequence vector; The solution using the hybrid artificial immune algorithm at least includes initializing the population, recording the antibody with the highest affinity, evaluating the fitness value of individuals in the population by decoding, selecting the antibody with the highest fitness in the population and cloning it, updating the search strategy for the cloned antibody, and further updating the population through the immune suppression process; The initialization population includes: The initialization strategy for the machine tool selection MS part is as follows: (1) Random allocation strategy: For each process, an available machine tool is randomly selected for processing; (2) Minimum processing time strategy: For each process, a suitable machine tool with the minimum processing time is selected from a group of available machine tools for processing; (3) Run the lowest energy consumption strategy: For each process, select a suitable machine tool with the lowest energy consumption from a group of available machine tools for processing; (4) Global minimum fitness strategy: This strategy is an extension of the global minimum processing time strategy, which converts the processing time into the weighted value of multiple objectives; (5) Global-local minimum fitness strategy: This strategy is an extension of the global minimum fitness strategy, which separates each job and applies the global minimum fitness optimization to each job to select the appropriate machine tool; The initialization strategy of the OS part of the process sequence is detailed as follows: (1) Random sorting strategy: This strategy is a simple and feasible method that randomly perturbs all processes to maintain the diversity of the population and randomly searches the entire solution space; (2) Maximum remaining number of steps strategy: The workpiece with the most remaining steps is arranged in the process sequence vector; if there are multiple workpieces with the same number of steps, a workpiece is randomly selected; The total energy consumption TEC is: Where n is the number of workpieces, op j is the number of processes for each workpiece j, u is the number of factories, m is the number of machine tools in each factory, j is the workpiece number, l is the process number, i is the machine tool number, f is the factory number, and X j,l,f,i,k is a binary decision variable, pt j,l,f,i For process O j,l Machining time on machine tool i in factory f, ec j,f,i is the unit energy consumption of workpiece j processed by machine tool i in factory f; The constraints are defined as follows: Z j,f ,X j,l,f,i,i' ,Y j,l,j',l' ∈{0,1}; Wherein, j, j' is the workpiece number subscript; l, l' is the process number subscript; i, i', i" is the machine tool number subscript; f is the factory number subscript; B is a positive integer; C j,l For process O j,l Completion time; C max is the maximum completion time; Z j,f is a binary decision variable. If process Oj,l is processed on machine tool i in factory f and its previous process is processed on machine tool i' in the same factory, then the value is 1, otherwise it is 0; Y j,l,j’,l’ is a binary decision variable, which takes the value of 1 if process Oj, l is processed before process Oj', l', otherwise it takes the value of 0.

2. A distributed flexible workshop scheduling method with a transportation process as claimed in claim 1, characterized in that: The distributed flexible workshop scheduling model is: minω·C max +(1-ω)·TEC, where ω is the weight of the target, C max is the maximum completion time obtained from the last process, and TEC is the total energy consumption.

3. A distributed flexible workshop scheduling method with a transportation process as claimed in claim 1, characterized in that: If the operation to be processed is the first operation of the workpiece, the transportation time from the virtual machine tool to the current machine tool is re-analyzed, otherwise, various situations from one machine tool to another are considered at the same time.

4. A distributed flexible workshop scheduling method with a transportation process as claimed in claim 1, characterized in that: A neighborhood-based search strategy is used to update the search strategy of the cloned antibodies. The global search is performed on the job assignment vector. In the inner loop, the neighborhood structure is used to perform local search in each neighborhood structure, which is used for the process sequence vector and the machine tool selection vector respectively. For each cloned antibody, a local search operator is applied to obtain a solution, and the obtained solution is inserted into the workpiece assignment vector. If no satisfactory solution is found in the neighborhood structure, the global neighborhood structure is replaced and the local search is continued in the new neighborhood structure.

5. A distributed flexible workshop scheduling system with transportation process, characterized in that: include: The data acquisition module is configured to: obtain information of workpieces to be processed and information related to the distributed flexible workshop; The scheduling module is configured to: obtain a scheduling result according to information of workpieces to be processed, information related to the distributed flexible workshop, and a preset distributed flexible workshop scheduling model; The distributed flexible workshop scheduling model is set as a sequence-based mixed integer linear programming model that uses a weighted sum method to simultaneously optimize the completion time and total energy consumption; when determining the transportation time between machine tools, a virtual machine tool is preset, and the process of the virtual machine tool conveying each workpiece is set as the first process. The completion time of each subsequent process is equal to the completion time of the previous process plus the processing time of the subsequent process and the transportation time between the two machine tools, and the maximum completion time is obtained from the last process of all operations; A hybrid artificial immune algorithm is used to solve the distributed flexible shop scheduling model, including workpiece allocation vector, machine tool selection vector and process sequence vector; The solution using the hybrid artificial immune algorithm at least includes initializing the population, recording the antibody with the highest affinity, evaluating the fitness value of individuals in the population by decoding, selecting the antibody with the highest fitness in the population and cloning it, updating the search strategy for the cloned antibody, and further updating the population through the immune suppression process; The initialization population includes: The initialization strategy for the machine tool selection MS part is as follows: (1) Random allocation strategy: For each process, an available machine tool is randomly selected for processing; (2) Minimum processing time strategy: For each process, a suitable machine tool with the minimum processing time is selected from a group of available machine tools for processing; (3) Run the lowest energy consumption strategy: For each process, select a suitable machine tool with the lowest energy consumption from a group of available machine tools for processing; (4) Global minimum fitness strategy: This strategy is an extension of the global minimum processing time strategy, which converts the processing time into the weighted value of multiple objectives; (5) Global-local minimum fitness strategy: This strategy is an extension of the global minimum fitness strategy, which separates each job and applies the global minimum fitness optimization to each job to select the appropriate machine tool; The initialization strategy of the OS part of the process sequence is detailed as follows: (1) Random sorting strategy: This strategy is a simple and feasible method that randomly perturbs all processes to maintain the diversity of the population and randomly searches the entire solution space; (2) Maximum remaining number of steps strategy: The workpiece with the most remaining steps is arranged in the process sequence vector; if there are multiple workpieces with the same number of steps, a workpiece is randomly selected; The total energy consumption TEC is: Where n is the number of workpieces, op j is the number of processes for each workpiece j, u is the number of factories, m is the number of machine tools in each factory, j is the workpiece number, l is the process number, i is the machine tool number, f is the factory number, and X j,l,f,i,k is a binary decision variable, pt j,l,f,i For process O j,l Machining time on machine tool i in factory f, ec j,f,i is the unit energy consumption of workpiece j processed by machine tool i in factory f; The constraints are defined as follows: Z j,f ,X j,l,f,i,i' ,Y j,l,j',l' ∈{0,1}; Wherein, j, j' is the workpiece number subscript; l, l' is the process number subscript; i, i', i" is the machine tool number subscript; f is the factory number subscript; B is a positive integer; C j,l For process O j,l Completion time; C max is the maximum completion time; Z j,f is a binary decision variable. If process Oj,l is processed on machine tool i in factory f and its previous process is processed on machine tool i' in the same factory, then the value is 1, otherwise it is 0; Y j,l,j’,l’ is a binary decision variable, which takes the value of 1 if process Oj, l is processed before process Oj', l', otherwise it takes the value of 0.

6. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of the distributed flexible workshop scheduling method with a transportation process as described in any one of claims 1 to 4 are implemented.

7. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the steps of the distributed flexible workshop scheduling method with transportation process as described in any one of claims 1 to 4 are implemented.

Citation Information

Patent Citations

  • Parallel machine group scheduling method and system based on improved artificial immune algorithm

    CN107730072A

  • Flexible job shop scheduling method and system with transmission and switching time

    CN112016801A

  • Flexible workshop scheduling optimization method and system considering crane transportation process

    CN112286149A

  • Distributed flexible workshop scheduling method and system with cranes

    CN112348314A

  • Scheduling method and system based on hybrid variable neighborhood search and gravitational search algorithm

    US20190079975A1