Method for solving flexible job shop scheduling problem based on clustering and reconstruction multi-task optimization algorithm

Abstract

The invention discloses a method for solving a flexible job-shop scheduling problem based on a clustering and reconstruction multi-task optimization algorithm, and the method comprises the following steps: taking the scheduling problem of each production line as an independent task, enabling the task to correspond to a population, extracting a descriptive feature vector of each task, and calculating the scheduling problem of each production line according to the descriptive feature vector; a group of initial solutions are randomly generated for each task to serve as individuals in a population, each individual is coded through a chromosome, and the chromosome corresponds to a machine allocation and process sorting scheme; calculating an adaptive value of each individual; individuals are periodically updated through an intra-cluster elite migration strategy, cooperation benefits of tasks and different clusters are periodically evaluated by adopting an inter-cluster adjustment and reconstruction strategy, a cluster structure is dynamically reconstructed, and finally an optimal scheduling scheme corresponding to each task is output. Coevolution of search knowledge among tasks is realized through a multi-task optimization framework, so that a complex flexible job shop scheduling problem is efficiently solved.

Description

Technical Field

[0001] This invention relates to the field of workshop scheduling technology, and specifically to a method for solving the flexible workshop scheduling problem based on a clustering and reconstruction multi-task optimization algorithm. Background Technology

[0002] With the transformation of manufacturing towards flexibility and intelligence, the Flexible Job-Shop Scheduling Problem (FJSP) has become a core challenge in production system optimization. Building upon traditional job-shop scheduling, this problem allows each process to be performed on multiple selectable machines, with processing times varying depending on machine characteristics, thus involving two levels of decision-making: machine allocation and process sequencing. The optimization task typically involves minimizing the maximum makespan while satisfying constraints such as process sequence and machine exclusivity. FJSP has been proven to be an NP-hard problem, with its solution space growing exponentially with the number of jobs and machines, making it difficult to find an exact solution in polynomial time.

[0003] Existing research mostly employs metaheuristic algorithms for solving FJSP problems, such as genetic algorithms (GA), simulated annealing (SA), and tabu search. Researchers have also gradually proposed new algorithms to optimize the results of FJSP, such as Anh-Dung NGUYEN et al.'s proposal to combine genetic algorithms with Google OR-Tools (GA-ORT), and ZhouWei et al.'s proposed improved genetic algorithm (IGA), which enhances local search capabilities by introducing simulated annealing. These methods have all achieved a certain degree of success, but their optimization processes are limited to single-task scenarios and do not consider collaborative optimization in multi-task environments. In real-world production environments, enterprises often need to handle multiple orders or production line scheduling tasks simultaneously. These tasks often share similarities in structure, process, or resource constraints, providing potential for improving overall optimization efficiency through knowledge transfer. However, traditional algorithms are mostly based on single-task scenarios, exhibiting significant shortcomings when dealing with such multi-task problems: simply solving each task independently fails to leverage inter-task similarities, leading to low computational resource utilization; combining all tasks into a single large-scale problem for unified solution can easily trigger "negative transfer" due to the lack of effective task similarity identification and knowledge transfer mechanisms. This means that optimization knowledge from one task interferes with the solution of another, degrading the quality of the solution. For example, while Azab et al.'s Discrete Event Simulation (DES) model can simulate production line dynamics, it cannot adaptively identify task similarities; and while IGA improves single-task search efficiency, it still lacks cross-task optimization capabilities. Therefore, in the face of flexible production demands for multiple varieties, a multi-task optimization scheme that can effectively identify task clusters and realize positive knowledge transfer is needed. Summary of the Invention

[0004] To overcome the shortcomings and deficiencies of existing technologies, this invention provides a method for solving the Flexible Job Shop (FJSP) scheduling problem based on a clustering and reconstruction multi-task optimization algorithm. This invention models the FJSP as a multi-task optimization problem, and uses a clustering and reconstruction multi-task optimization algorithm to identify similar scheduling problems through task clustering analysis. Furthermore, it utilizes inter-cluster reconstruction strategies to achieve knowledge transfer. By combining feature clustering, intra-cluster co-evolution, and inter-cluster dynamic reconstruction mechanisms, it achieves efficient and adaptive scheduling optimization. Under the premise of adhering to model rule constraints, it can still derive design schemes with smaller maximum completion times, effectively avoiding negative transfer phenomena and significantly improving the solution efficiency and quality of FJSP in a multi-task context.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: This invention provides a method for solving the flexible job shop scheduling problem based on a clustering and reconstruction multi-task optimization algorithm, comprising the following steps: The scheduling problem of each production line is treated as an independent task and corresponds to a population. The descriptive feature vector of each task is extracted, and a set of initial solutions is randomly generated for each task as individuals in the population. Each individual is encoded by a chromosome, which corresponds to the machine allocation and process sequencing scheme. Calculate the fitness value for each individual; Based on the descriptive feature vector of each task, all tasks are divided into multiple clusters; Genetic algorithm evolution is performed independently for each task within each cluster. At a set generation, knowledge transfer within the cluster is performed, and the elite individuals of the tasks within the cluster are transferred and replaced with the worst individuals. Every set number of generations, a portion of the tasks are sampled as exploration tasks. The benefits of the external elite individuals to the current task are calculated. Based on the benefit comparison, the task grouping is dynamically restructured until the set fitness value evaluation number is reached. Then, the optimal scheduling scheme corresponding to each task is output.

[0006] As a preferred technical solution, the descriptive feature vector for each task includes the number of orders, the number of product types, the number of machines, and the average process time.

[0007] As a preferred technical solution, the fitness value of each individual is calculated, specifically including: ; ; in, C represents the fitness value. max (X) represents the maximum completion time calculated by the scheduling scheme. Let X represent the penalty coefficient, O(X) represent the degree to which the scheduling scheme violates the operation order constraint, and M(X) represent the degree to which the scheduling scheme violates the machine mutual exclusion constraint.

[0008] As a preferred technical solution, the operation sequence constraint is expressed as follows: ; Among them, S i(j+1) This indicates that the assignment J is being performed. i The start time of the (j+1)th operation. Indicates operation O ij Completion time, Indicates the number of operations; Machine mutual exclusion constraints are represented as: ; in, Indicates operation O ij The start time, Indicate homework The The completion time of each operation, where H represents a constant. Let i represent the sequence variable, j be the job index, and k be the operation index.

[0009] As a preferred technical solution, all tasks are divided into multiple clusters based on the descriptive feature vector of each task, specifically including: Based on the K-Means clustering algorithm, all tasks are divided into K clusters according to the Euclidean distance of the feature vectors in the multivariate space, and tasks whose Euclidean distance between feature vectors meets the set range are grouped into the same cluster.

[0010] As a preferred technical solution, when setting the algebra, knowledge transfer within the cluster is performed, replacing the worst individual with the best individual within the cluster. Specifically, this includes: Set intra-cluster knowledge transfer conditions, determine whether the fitness value of the external elite individual in this task is better than the fitness value of the worst individual in this task. If the condition is met, for each task in the cluster, select the top N elite individuals with the highest fitness values ​​from the current population and put them into the candidate pool. Use the environment of the current task to evaluate the external elite individuals in the candidate pool and calculate their fitness values. If an elite individual from outside the current task environment has a fitness value that is better than that of the worst individual in the current population, then the elite individual is used to replace the worst individual in the current population. Otherwise, no replacement is performed, and the number of cases where no replacement has occurred is counted. When the count of cases where no knowledge transfer has occurred within a cluster reaches a set threshold, inter-cluster adjustment and reconstruction are performed.

[0011] As a preferred technical solution, the benefit of external elite individuals to the current task is calculated, and the task grouping is dynamically restructured based on the benefit comparison, specifically including: For each task T i Several exploration tasks T are randomly selected from those not belonging to the cluster. j From task T i Obtain elite individuals from the current population. best And calculate the performance of elite individuals on task T i The profit is expressed as: ; in, This indicates that the elite individuals from outside the region are in task T. i fitness value in the environment and Task T i At the current population's historical best and worst fitness values; Calculate its origin from each external cluster C x The average benefit AvgBenefit (T) i C x If cluster C exists best This ensures that its average return satisfies: ; in, If the preset gain threshold is used, then task T will be... i From its current cluster C current Migration to cluster C best middle.

[0012] As a preferred technical solution, the scheduling scheme satisfies the process sequence constraint, machine exclusivity constraint, and delivery date constraint. The process sequence constraint means that each operation within the same job is processed sequentially according to a predetermined process route. The machine exclusivity constraint means that the same machine can only perform one operation at any given time. The delivery date constraint means that all jobs must be completed before the final delivery deadline.

[0013] The present invention also provides a computer-readable storage medium storing a program, which, when executed by a processor, implements the above-described method for solving the flexible job shop scheduling problem based on a clustering and reconstruction multi-task optimization algorithm.

[0014] The present invention also provides a computer device, including a processor and a memory for storing processor-executable programs, wherein when the processor executes the program stored in the memory, it implements the above-described method for solving the flexible job shop scheduling problem based on a clustering and reconstruction multi-task optimization algorithm.

[0015] Compared with the prior art, the present invention has the following advantages and beneficial effects: (1) The initial clustering grouping of this invention is based on the K-Means algorithm. First, the tasks are grouped according to the descriptive feature vector of each multi-production line scheduling task (such as the number of orders, the number of product types, the number of machines, the average process time, etc.). Then, the K-Means algorithm will find a set of cluster centers through iterative calculation and assign each task to the cluster represented by the cluster center closest to it, realizing the initial K clusters. This process aims to maximize the similarity of task features within the cluster and maximize the difference between different clusters. Through this grouping, tasks with similar features are grouped into the same cluster, thereby creating conditions for efficient knowledge transfer between them in a logical way and reducing the possibility of negative transfer between different tasks from the source.

[0016] (2) The present invention adopts an intra-cluster elite knowledge transfer strategy. Within each cluster, an efficient knowledge transfer strategy is executed periodically to accelerate co-evolution. The core operation of this knowledge transfer strategy is to select several elite individuals with the best current fitness of each task in the cluster and enter them into the candidate pool. Each task evaluates the elite individuals in the candidate pool. Each task will select the elite individual that benefits it the most (i.e., the one that can improve the fitness of the task the most) from the candidate pool according to its own optimization needs and directly replace the worst individual in the population. This process realizes the directional transfer of high-quality scheduling schemes between different tasks, so that lagging tasks can directly learn advanced experience, thereby significantly accelerating the convergence speed of the entire cluster.

[0017] (3) The present invention adopts an inter-cluster adjustment and reconstruction strategy, which is different from intra-cluster knowledge transfer aimed at accelerating convergence. The strategy of the present invention focuses on global exploration to prevent the algorithm from getting stuck in local optima due to the initial immature clustering grouping. The inter-cluster exploration mechanism is initiated periodically (or when intra-cluster knowledge transfer is detected to be stagnant). The exploration task is specified by probability sampling and the benefits of elite individual transfer with tasks in non-clusters are evaluated. When a better task cooperation relationship is confirmed, the cluster structure will be dynamically reconstructed so that the task grouping can adapt to the evolution process, thereby enhancing the global optimization capability of the algorithm. Attached Figure Description

[0018] Figure 1 This is a Gantt chart of the flexible workshop scheduling scheme of the present invention; Figure 2 This is a flowchart illustrating the process of solving the flexible job shop scheduling problem using a multi-task optimization algorithm based on clustering and reconstruction, as described in this invention. Figure 3 This is a schematic diagram of intra-cluster knowledge transfer in this invention; Figure 4 This is a schematic diagram of the inter-cluster reconstruction strategy of the present invention. Detailed Implementation

[0019] 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. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0020] Example 1 like Figure 1 As shown, the FJSP problem consists of components such as jobs, operations, and machines. In this model, it is assumed that the basic data of the FJSP problem is given, including the job set, machine set, and processing time of each operation. The main objective of the FJSP optimization problem is to select a suitable processing machine for each operation and determine the processing order of the operations on the machines in order to minimize the completion time of all jobs under the following three constraints (makespan). (a) Machine availability constraint: Each operation must be performed on one of the available machines. (b) Operation sequence constraint: Operations within the same task must be processed sequentially in a predetermined order; (c) Machine mutual exclusion constraint: Each machine can only process one operation at a time; Among them, machine availability constraints define the range of machines that can be selected for each operation, while operation order constraints and machine mutual exclusion constraints are necessary constraints to ensure the feasibility of the FJSP problem. Assume there is an FJSP problem containing N... j One assignment, each assignment J i By N i It consists of N operations, totaling N. m Each machine, each operation O ij In machine M k The processing time is T. ijk The function f is used to map the solution X (machine allocation and ordering of all operations) to the completion time of the entire scheduling scheme. The model can be defined by the following formula: (1) (2) (3) (4) (5) (6) The total completion time in the model is defined by formula (1), where It is a binary variable, representing an operation of O. ij Is it in machine M? k Upper processing (Formula 2). and These are operations O ij The start and finish times; Formula (3) ensures the processing time constraint, where T ijk Indicates operation O ij In machine M k The processing time is determined by formula (4), which guarantees the operation sequence constraint, where S i(j+1) Indicates assignment J i The start time of the (j+1)th operation; Formula (5) ensures machine mutual exclusion constraints, where Indicate homework The The completion time of each operation, H is a large constant used to invalidate constraints when they are no longer needed. It is a sequence variable. In the above formula, i is the job index, j is the operation index, and k is the machine index. The machine availability constraint is shown in formula (2), and the operation order and machine mutual exclusion constraint are guaranteed by formulas (4) and (5).

[0021] In the FJSP problem, different scheduling tasks have different descriptive characteristics (such as the number of orders, the number of product types, the number of machines, and the average process time), which affect the complexity of scheduling and the structure of the solution space. Therefore, in the model, tasks are clustered according to feature vectors. Tasks with more similar features are more likely to be grouped into the same cluster, thus providing a basis for knowledge transfer. This invention combines genetic algorithms with a multi-task optimization framework, assigning each task to a population. Each individual in the population has a chromosome vector, the dimension of which is related to the total number of operations for that task. The chromosome vector encodes the machine allocation and sorting scheme for each operation. The fitness of each individual is determined by the maximum completion time of the scheduling scheme. At the same time, each population maintains a set of chromosomes during evolution and updates them periodically through an intra-cluster elite migration strategy. To mitigate the negative impact of inaccurate initial clustering, an inter-cluster adjustment and reconstruction strategy is also adopted. The collaborative benefits of tasks with different clusters are periodically evaluated, and the cluster structure is dynamically reconstructed to avoid local optima. During the optimization process, although the scheduling problem is discrete, the chromosomes are encoded with real values, and the design scheme is continuously optimized through genetic operations (such as crossover and mutation). By using a multi-task optimization framework to achieve the co-evolution of knowledge search among tasks, the complex flexible job shop scheduling problem can be solved efficiently.

[0022] This embodiment treats the scheduling problem of each production line as an independent task, closely linking them through knowledge transfer and cluster reconstruction strategies to jointly form a complete flexible job shop scheduling multi-task optimization system, thereby obtaining the final optimal design scheme with the smallest maximum completion time. Different tasks merely mean different problem instances, but each task solves the exact same core problem—the flexible job shop scheduling problem, which involves allocating and sorting machines for each operation to minimize the maximum completion time.

[0023] like Figure 2 As shown, the specific steps include: S1: Problem Initialization: Using genetic algorithm as the basic optimization algorithm, initialize the scheduling optimization task of multiple production lines. The scheduling problem of each production line is an independent task. Extract the descriptive feature vector of each task, including the number of orders, the number of product types, the number of machines, the average process time, etc. According to the definition of the FJSP model and multi-task optimization algorithm, each scheduling task is modeled as an independent optimization problem and corresponds to a population. During population initialization, a set of initial solutions is randomly generated for each task, i.e., individuals in the population. Each individual is encoded by a chromosome, which represents a complete machine allocation and process sequencing scheme for that task. The quality of an individual is directly evaluated by the maximum completion time (makespan) of its scheduling scheme; at the same time, descriptive feature vectors for each task are extracted, including the number of orders, the number of product types, the number of machines, the average process time, etc., for subsequent task clustering and grouping, and the number of algorithm evaluations (Fes) is initialized to zero.

[0024] S2: Calculate the fitness value. The fitness value of each task is determined by the objective function value (maximum completion time) of its scheduling scheme. The objective function value of the scheduling scheme is used as the fitness value of the individual task for evaluation. At the same time, the scheduling scheme must satisfy the process sequence constraint, machine exclusivity constraint, and delivery date constraint. The process sequence constraint means that each operation within the same job must be processed in the order of the predetermined process route. The machine exclusivity constraint means that the same machine can only execute one operation at any time. The delivery date constraint means that all jobs must be completed before their final delivery deadline.

[0025] The fitness value of each individual is determined by the maximum completion time C of the scheduling scheme X it represents. max The decision must be made. Simultaneously, the satisfaction of the problem constraints by the scheduling scheme must be considered. To facilitate algorithm processing, the degree of constraint violation is incorporated into the fitness value calculation as a penalty term. The final individual fitness value is calculated according to the following formula: (7) (8) Among them, C max O(X) represents the maximum completion time calculated by scheduling scheme X, as defined in formulas (1) and (6); O(X) represents the degree to which scheme X violates the operation sequence constraints (corresponding to formula (4)), for example, calculating all violations The number or time overlap of the constraint operation pairs, where S i(j+1)k C represents the start time of the (j+1)th operation of job i on machine k. ijk M(X) represents the completion time of the j-th operation of job i on machine k; M(X) represents the degree to which solution X violates the machine mutual exclusion constraint (corresponding to formula (5)), for example, calculating the number of operation pairs with overlapping processing times on the same machine or the total overlapping time; γ is a large penalty coefficient designed to significantly distinguish feasible solutions from infeasible solutions and guide the search toward the feasible region.

[0026] S3: Initial clustering and grouping: Using the K-Means clustering algorithm, all tasks are divided into K clusters based on the Euclidean distance of the feature vectors in the multivariate space, and tasks that are close to each other and have similar features are initially grouped. To achieve effective knowledge transfer between tasks, similar tasks must first be grouped. Using the K-Means clustering algorithm, all tasks are divided into K clusters based on their descriptive feature vectors (including order quantity, product type, machine count, average process time, etc.). After clustering, tasks with smaller Euclidean distances between their feature vectors are grouped into the same cluster, laying the foundation for subsequent intra-cluster knowledge transfer. This step is performed once at the beginning of the algorithm to complete the initial clustering of tasks.

[0027] S4: As Figure 3 As shown, the strategy of intra-cluster task knowledge transfer and independent evolution is implemented: the genetic algorithm evolution operation, including selection, crossover and mutation, is performed independently for each task in each cluster. Intra-cluster knowledge transfer is performed at a fixed number of generations, that is, the transfer of elite individuals of tasks in the cluster and the replacement of the worst individuals. In this embodiment, each task is independently optimized using a genetic algorithm framework. Simultaneously, to promote knowledge sharing among similar tasks within the cluster, a periodic intra-cluster knowledge transfer mechanism is implemented. Every fixed number of evaluation generations (g generations), it is determined whether the intra-cluster knowledge transfer condition is met, i.e., whether the fitness value of an external elite individual in this task is better than the fitness value of the worst individual in this task. If the condition is met, for each task within the cluster, the top N elite individuals with the highest fitness values ​​are selected from the current population and placed into a candidate pool. Subsequently, the external elite individuals in the candidate pool are evaluated using the environment of the current task (i.e., the specific machine, specific job, and specific processing time of the current task, etc.), and their fitness values ​​are calculated. If the fitness value of an external elite individual in the current task environment is better than the fitness value of the worst individual in the current population, then the elite individual replaces the worst individual in the current population; otherwise, no replacement is performed, and the number of cases where no replacement occurs is counted. When the count of cases where no knowledge transfer occurs within the cluster reaches a set threshold, inter-cluster adjustments will be performed. This strategy allows excellent scheduling schemes to be migrated between similar tasks, accelerating overall convergence within the cluster.

[0028] S5: Inter-cluster adjustment and reconstruction: Every certain number of generations, a portion of the tasks are sampled by probability as exploration tasks. Tasks are randomly selected from the non-dominant clusters, and the benefits of the external elite individuals to the current task are evaluated. The benefit information is recorded, and cluster reconstruction and task migration are carried out when the conditions are met. like Figure 4 As shown, to overcome the limitations of intra-cluster evolution and utilize high-quality cross-cluster scheduling information, a dynamic inter-cluster reconstruction mechanism is introduced. The mechanism works as follows: every fixed number of generations or when the number of times intra-cluster knowledge transfer stagnates reaches a certain level, for each task T... i (T) i The cluster it belongs to is C(T) i ), randomly select several exploration tasks T from those not belonging to its cluster. j ( From T j Obtain elite individuals from the current population. best And assess the foreign individual's contribution to task T. i The Benefit is calculated using the following formula: (9) in, This indicates that the elite individuals from outside the region are in task T. i fitness value in the environment and Task T i This formula quantifies the performance of an introductory individual relative to task T, based on the current population's historical best and worst fitness values. i The degree of improvement in the quality of the current solution; Subsequently, the average value of the gains generated by these individuals from different external clusters is calculated, and the cluster reconstruction process is initiated: for task T i Calculate its origin from each external cluster C x The average benefit AvgBenefit (T) i C x If an external cluster C exists... best This ensures that its average return satisfies: (10) in, If the preset gain threshold is used, then task T will be... i From its current cluster C current Migration to cluster C best This dynamic reconstruction mechanism, based on quantitative benefit comparison, ensures the adaptability of task grouping and can effectively utilize high-value knowledge across clusters.

[0029] When the specified number of fitness evaluations (50,000 times) is reached, the optimal scheduling scheme found for each task is output.

[0030] During operation, every certain number of generations or when intra-cluster knowledge transfer stagnates, a check is performed to determine if the inter-cluster reconstruction conditions are met. Tasks are sampled from each cluster with a certain probability as exploration tasks. For each task, elite individuals from its non-dominant clusters are introduced into the current task population for evaluation. If an individual introduced from a particular cluster is repeatedly proven to significantly improve the current task, the task is moved from its original cluster to the cluster with the highest benefit. Through this dynamic adjustment mechanism of task cluster affiliation, the algorithm can adaptively discover and utilize high-value scheduling knowledge across clusters, enhancing its global exploration capabilities.

[0031] This invention aims to minimize the maximum completion time while satisfying constraints such as process sequence, machine exclusivity, and delivery time. Addressing the issue of large task numbers and poor scalability of conventional algorithms in flexible job shop scheduling, this invention proposes a knowledge transfer mechanism based on clustering and dynamic reconstruction. It utilizes task feature vectors for rapid clustering, grouping similar tasks to achieve efficient intra-cluster knowledge transfer. Furthermore, by introducing an inter-cluster exploration mechanism, the potential benefit of external elite individuals to the current task is periodically evaluated, and task grouping is dynamically reconstructed based on benefit comparison. This adaptively discovers and utilizes high-value optimization knowledge across clusters. The combination of intra-cluster evolution and inter-cluster exploration and reconstruction effectively balances depth optimization and breadth exploration, thus maintaining excellent convergence performance and global optimization capabilities in complex multi-task environments.

[0032] Example 2 This embodiment provides a storage medium, which may be a ROM, RAM, disk, optical disk, or other storage medium. The storage medium stores one or more programs. When the programs are executed by a processor, they implement the method of solving the flexible job shop scheduling problem based on the clustering and reconstruction multi-task optimization algorithm of Embodiment 1.

[0033] Example 3 This embodiment provides a computing device, which may be a desktop computer, laptop computer, smartphone, PDA handheld terminal, tablet computer or other terminal device with display function. The computing device includes a processor and a memory. The memory stores one or more programs. When the processor executes the program stored in the memory, it implements the method of solving the flexible job shop scheduling problem based on the clustering and reconstruction multi-task optimization algorithm of Embodiment 1.

[0034] The above embodiments are preferred embodiments of the present invention, but the embodiments of the present invention are not limited to the above embodiments. Any changes, modifications, substitutions, combinations, or simplifications made without departing from the spirit and principle of the present invention shall be considered equivalent substitutions and shall be included within the protection scope of the present invention.

Claims

1. A method for solving flexible job shop scheduling problem based on clustering and reconstruction multi-task optimization algorithm, characterized in that, The method comprises the following steps: The scheduling problem of each production line is taken as an independent task, and a population corresponding to each task is extracted. A set of initial solutions of each task is randomly generated as individuals in the population. Each individual is coded by a chromosome, and the chromosome corresponds to a machine allocation and process sequencing scheme; The fitness value of each individual is calculated; All tasks are divided into multiple clusters according to the descriptive feature vectors of each task; The genetic algorithm evolution operation is independently performed on each task in each cluster. The migration of elite individuals in the cluster and the replacement of the worst individual are performed when the number of generations is set. Every certain number of generations, a part of tasks are sampled as exploration tasks. The yield of the external elite individual to the current task is calculated. The task grouping is dynamically reconstructed based on the yield comparison. When the set adaptive value evaluation times are reached, the optimal scheduling scheme corresponding to each task is output.

2. The method for solving flexible job shop scheduling problem based on clustering and reconstruction multi-task optimization algorithm according to claim 1, characterized in that, The descriptive feature vector of each task includes the number of orders, the number of product categories, the number of machine tables, and the average process time.

3. The method of solving flexible job shop scheduling problem based on clustering and reformulation multi-task optimization algorithm according to claim 1, characterized in that, The fitness value of each individual is calculated, specifically including: ； ； wherein, represents an adaptation value, C max (X) represents the maximum completion time calculated by the scheduling scheme, represents a penalty coefficient, O(X) represents the degree of violation of the operation sequence constraint by the scheduling scheme, and M(X) represents the degree of violation of the machine mutual exclusion constraint by the scheduling scheme.

4. The method for solving flexible job shop scheduling problem based on clustering and reconstruction multi-task optimization algorithm according to claim 3, characterized in that, The operation sequence constraint is represented as: ； wherein S i(j+1) denotes the start time of the j+1th operation of the job J i denotes the completion time of the operation O ij denotes the number of operations;​​ The machine mutual exclusion constraint is represented as: ； wherein, denotes the start time of operation O ij , denotes the completion time of the jth operation of job i, H denotes a constant, denotes the completion time of the jth operation of job i, H denotes a constant, denotes the sequence variable, i is the job index, j is the operation index, and k is the machine index.​ 5. The method for solving flexible job shop scheduling problem based on clustering and reformulation multi-task optimization algorithm according to claim 1, characterized in that, All tasks are divided into multiple clusters according to the descriptive feature vectors of each task, specifically including: Based on the K-Means clustering algorithm, the Euclidean distance of all tasks in the multi-dimensional space is divided into K clusters according to the feature vectors. Tasks with a Euclidean distance between feature vectors within a set range are divided into the same cluster.

6. The method for solving flexible job shop scheduling problem based on clustering and reformulation multi-task optimization algorithm according to claim 1, characterized in that, The knowledge migration within the cluster is performed when the number of generations is set. The migration of elite individuals in the cluster and the replacement of the worst individual are performed, specifically including: The cluster knowledge migration condition is set. It is judged whether the fitness value of the external elite individual in the task is better than the fitness value of the worst individual in the task. If it is satisfied, the top N elite individuals in the current population of each task in the cluster are selected and put into a candidate pool. The external elite individual in the candidate pool is evaluated in the current task environment, and its fitness value is calculated. If the fitness value of a certain external elite individual in the current task environment is better than the fitness value of the worst individual in the current population, the elite individual is used to replace the worst individual in the current population. Otherwise, no replacement is performed. The situation where no replacement occurs is counted. When the count of no knowledge migration within the cluster reaches a set threshold, the cluster adjustment and reconstruction are performed.

7. The method of solving flexible job shop scheduling problem based on clustering and reformulation multi-task optimization algorithm according to claim 1, characterized in that, The yield of the external elite individual to the current task is calculated. The task grouping is dynamically reconstructed based on the yield comparison, specifically including: For each task T i Several exploration tasks T are randomly selected from those not belonging to the cluster. j From task T i Obtain elite individuals from the current population. best And calculate the performance of elite individuals on task T i The profit is expressed as: ； wherein, represents the fitness value of the foreign elite individual in task T i under the environment, and are the historical best and worst fitness values of the current population in task T i respectively. computing its benefit from each foreign cluster C x the average benefit AvgBenefit (T i , C x ) it would receive if it were to join cluster C best such that its average benefit satisfies: ； wherein, is a preset gain threshold, then task T i from its current cluster C current migrate to cluster C best .

8. The method for solving flexible job shop scheduling problem based on clustering and reformulation multi-task optimization algorithm according to claim 1, characterized in that, The scheduling scheme satisfies the process sequence constraint, the machine exclusivity constraint and the delivery period constraint. The process sequence constraint represents that each operation within the same job is processed in a predetermined process route sequence. The machine exclusivity constraint represents that a same machine can execute only one operation at any time. The delivery period constraint represents that all jobs are completed before the final delivery deadline.

9. A computer readable storage medium storing a program, characterized in that, The program is executed by the processor to implement the method for solving the flexible job shop scheduling problem based on the clustering and reconstruction multi-task optimization algorithm of any one of claims 1-8.

10. A computer device comprising a processor and a memory for storing a processor executable program, characterized in that, The processor implements the method for solving the flexible job shop scheduling problem based on the clustering and reconstruction multi-task optimization algorithm according to any one of claims 1-8 when executing the program stored in the memory.

Images