Multi-task optimization method for multi-user target requirements of distributed manufacturing system

Abstract

The embodiment of the invention discloses a multi-task optimization method for multi-user target requirements of a distributed manufacturing system, and relates to the technical field of intelligent manufacturing. According to the scheme, the method comprises the steps that a distributed manufacturing platform receives manufacturing demand information uploaded by a client terminal and establishes a manufacturing task set; performing matching processing on the manufacturing task set and the manufacturing resource set to obtain a candidate service set corresponding to the manufacturing task; and determining service information matched with the sub-tasks of the manufacturing tasks in the manufacturing task set from the candidate service set, obtaining a feasible service combination corresponding to the manufacturing tasks, and then sending order information to a corresponding enterprise terminal according to the obtained feasible service combination. The method can accurately and effectively reflect the effect that the scheduling scheme optimizes different user targets, alleviates the calculation pressure when a scheduling control system of the distributed manufacturing system deals with task scheduling caused by too many tasks and targets, and promotes the improvement of the operation efficiency of the distributed manufacturing system.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent manufacturing, and particularly relates to a multi-task optimization method for multi-user target requirements of a distributed manufacturing system. BACKGROUND

[0002] Most of the existing distributed manufacturing schemes focus on meeting the needs of service demanders, which may ignore the objective conditions of service providers and platforms, resulting in problems such as insufficient supply of manufacturing resources, reduced quality, and unreasonable scheduling schemes. It is worth noting that the objectives of different users in a distributed manufacturing system are usually independent, so the requirements of each user can be regarded as an independent optimization task with specific objectives and constraints, but these tasks must be implemented through the same scheduling scheme.

[0003] Currently, evolutionary algorithms are widely used to solve the service combination optimization problem in distributed manufacturing. However, most evolutionary algorithms can only handle one task at a time, and the number of objectives increases dramatically when all user requirements are treated as separate optimization objectives, making it difficult to simultaneously meet the objectives of different users. To solve this problem, the same or similar objectives of different users are usually aggregated into a unified global objective using methods such as weighting or summation, which helps to reduce the number of optimization objectives and simplify the complexity of the problem. Although this objective aggregation method simplifies the problem to some extent, it essentially transforms the multi-task optimization problem into a single-task optimization problem with a single objective or multiple objectives, so there is still an imbalance in resource allocation. Another type of solution method uses multi-task optimization algorithms such as MFEA (Multifactorial Evolutionary Algorithm), which can simultaneously solve the optimal service combination of manufacturing tasks from different customers, but it ignores the fact that manufacturing resources are limited. Because this type of method optimizes multiple independent manufacturing tasks simultaneously without considering the competition for resources among these tasks, it is not suitable for service combination optimization in real-world distributed manufacturing scenarios.

[0004] In summary, in addition to the competition for service resources, there is also a real need to coordinate the optimization of different users' objectives in a distributed manufacturing system. There is an urgent need to design and develop a mechanism that can accurately and effectively reflect the effect of scheduling scheme optimization on different user objectives, thereby alleviating the computational pressure of the scheduling control system of the distributed manufacturing system when dealing with excessive tasks and objectives, and promoting the improvement of the operating efficiency of the distributed manufacturing system. SUMMARY

[0005] The embodiments of the present invention provide a multi-task optimization method for multi-user target requirements in a distributed manufacturing system. This method can accurately and effectively reflect the effect of scheduling scheme optimization on different user targets, alleviate the computational pressure on the scheduling and control system of the distributed manufacturing system when dealing with too many tasks and targets, and promote the improvement of the operating efficiency of the distributed manufacturing system.

[0006] To achieve the above objectives, the embodiments of the present invention adopt the following technical solutions: A multi-task optimization method for multi-user objective requirements in distributed manufacturing systems, such as Figure 9 As shown, including: S1. The distributed manufacturing platform receives manufacturing demand information uploaded by client terminals and establishes a set of manufacturing tasks, wherein each manufacturing task in the set is further decomposed into at least one sub-task; for example, the service composition optimization process in a distributed manufacturing system can be described as follows: [The text abruptly ends here, so the translation stops as well.] Submit their requirements to the distributed manufacturing platform, and the platform will... The customer's needs are described as a series of manufacturing tasks, namely , Index it. Typically, each task... All can be decomposed into sequential structures Sub-tasks, namely , Index it.

[0007] S2. Accept manufacturing resource information uploaded by the supplier's enterprise terminal and establish a manufacturing resource set; for example: Each supplier's enterprise terminal, each enterprise terminal Provide to the platform Each manufacturing resource, platform, and these resources are encapsulated into a collection, namely... , Index it.

[0008] S3. Match the manufacturing task set with the manufacturing resource set to obtain a candidate service set for the corresponding manufacturing task; for example: the platform will... The manufacturing content and Matching manufacturing capabilities to construct Candidate service set , Index it.

[0009] S4. From the candidate service set, determine the service information that matches the sub-tasks of the manufacturing tasks in the manufacturing task set, obtain the feasible service combination corresponding to the manufacturing task, and then send order information to the corresponding enterprise terminal according to the obtained feasible service combination. The order information is used to initiate the manufacturing process for the enterprise. For example: the platform from For each Select services to obtain a feasible service combination for each manufacturing task.

[0010] In this embodiment, S4 includes: S41. Establish a three-layer multi-task optimization model, wherein the individuals in the upper layer of the three-layer multi-task optimization model correspond to customer terminals, and the individuals in the lower layer correspond to enterprise terminals; wherein the upper layer aims to maximize customer satisfaction, service quality, and minimize service costs, the lower layer aims to maximize revenue for suppliers, and the middle layer aims to minimize penalty costs, total costs, and maximize production flexibility, total revenue, total satisfaction, and total service quality for the distributed manufacturing platform.

[0011] S42. In the three-layer multi-task optimization model, after generating the initial population using random key encoding, the key is selected uniformly. For example, a novel decoding method including a starting offset and a selection interval can be used to ensure uniform key selection. For example, the initial population is generated using random key encoding, and a unique decoding method is designed. The specific process is as follows: For each individual... , defines it by express, Depend on indivual It is composed of random variables, and ;by Starting from Select the key and use it. As the selection interval, until a selection is made. A key as a task The encoding, if the selected position exceeds If the length is less than or equal to the given length, then the selection will start from the beginning. and The calculation is as follows:

[0012] Assigned to The manufacturing services are provided by mapping It means that, among them It is the first one in the selected keys The value at each position. Assume... It is 6. It is 0.7. This means choosing The fifth manufacturing resource is Provide services. In each subtask... After selecting the manufacturing service, the execution order of each subtask is determined according to the First-In-First-Out (FIFO) principle. The subtask with the earliest ideal start time is executed first. If multiple subtasks have the same start time, the subtask with the most subsequent subtasks is executed. If the number of subsequent subtasks is also the same, the subtask with the shortest required time is executed first.

[0013] S43. Optimize the goals of each customer and supplier and establish the skill vectors of each group of individuals; specifically, calculate the results of optimizing the scheduling scheme for each customer and supplier goal, rank the results, and use reverse normalization to calculate the skill vector of each individual for the ranking results of each goal.

[0014] S44. Use the projection length of the skill vector onto the reference vector to determine similarity, and divide the population individuals into elite individuals and non-elite individuals based on the similarity determination results.

[0015] S45. Adaptive differential evolution is performed on elite individuals and non-elite individuals respectively, and offspring populations are generated. In the evolution stage, different adaptive differential evolution methods are adopted for elite individuals and non-elite individuals to generate offspring populations. Elite individuals are guided to explore the unknown solution space, and non-elite individuals are guided to move closer to elite individuals. At the same time, an adaptive crossover rate combining linear decay and periodic oscillation is designed.

[0016] S46. Merge the generated offspring population with the original population, sort the distributed manufacturing platform targets in the merged population, and select the top-ranked individuals to form the next generation population. Repeat the above operations on the newly obtained population until the termination criteria are met.

[0017] Specifically, A set of manufacturing tasks consists of a series of manufacturing tasks corresponding to the manufacturing requirements of each customer. , The index of T; Decomposed into sequential structures Each subtask is represented as , for index, Indicates the first Sub-tasks; Total Enterprise terminals in a supplier company Provide to the platform A manufacturing resource, the manufacturing resource set is , is the index of, represents the mth manufacturing service in the manufacturing resource set. The platform matches the manufacturing content of with the manufacturing capacity of , constructs the candidate service set of , is its index. The platform selects services for each from .

[0018] In S41, the three-layer multi-task optimization model established includes: the personalized target model of the upper customer, the personalized target model of the lower supplier and the overall target model of the intermediate distributed manufacturing platform; The personalized target model of the upper customer includes: Wherein, min and max represent the minimum function and the maximum function respectively; 、 and represent the satisfaction, service quality and service cost of the customer respectively; and represent the upper limit and the lower limit of the delivery time required by the customer respectively, represent the delivery time required by the customer ; and represent two risk attitude coefficients respectively; and represent the punishment coefficients of task delay delivery and early delivery respectively; is a 1-0 coefficient for detecting whether can be manufactured on ; is a 1-0 coefficient for determining whether is manufactured on ; is the qualified rate of manufacturing service ; and are manufacturing cost and transportation cost respectively; is the income value of manufacturing service per unit time; is the workload of task , is the manufacturing capacity of service per unit time; is the index of the two service suppliers before and after the task transportation;​ denotes the i-th subtask whether the 1-0 coefficient of whether the i-th subtask can be manufactured on denotes the i-th subtask whether the 1-0 coefficient of whether the i-th subtask is decided to be manufactured on is the transportation cost per unit distance, is the distance between the service provider and , and is the distance between the provider and the customer ; denotes the i-th subtask, i.e., the last subtask whether the 1-0 coefficient of whether the i-th subtask, i.e., the last subtask can be manufactured on denotes the i-th subtask, i.e., the last subtask whether the 1-0 coefficient of whether the i-th subtask, i.e., the last subtask is decided to be manufactured on the personalized target model of the lower-layer provider, comprising: , wherein denotes the profit of the provider ; denotes the time spent by the manufacturing service to complete the subtask ; is the cost per unit time of the manufacturing service .

[0019] the overall target model of the middle-layer distributed manufacturing platform, comprising: , wherein PC, F, C, B, S and Q respectively denote the penalty cost, the scheduling scheme flexibility, the total cost, the total profit, the total satisfaction and the total service quality of the platform; and respectively are the unit penalty cost of early delivery and delayed delivery of the task; and respectively are the ability to cope with manufacturing task changes and manufacturing resource changes; , , and respectively denote the ability to cope with manufacturing task changes by relying on service diversity, the remaining manufacturing resources and cooperative enterprises; is the service diversity of , is the provider ​​​​The number of partner companies Is with The number of services with similar functions; , and They are respectively relying on The reliability of the facilities, similar manufacturing resource services, and the ability of partner companies to adapt to changes in manufacturing resources.

[0020] The constraints of the three-layer multi-task optimization model include: ,in, Indicates the first Sub-tasks, namely The start time, express Completion time, express Actual completion time Indicates manufacturing services Available service hours Indicates the first The transportation time for each sub-task.

[0021] S43 includes: Establish individual skill vectors ,in, Indicates task fitness. reflect Optimize suppliers Performance in terms of the target aspect; , and They represent In optimization and The ranking of the target aspects, and and These represent the minimum and maximum ranking values ​​for the corresponding targets, respectively.

[0022] S44 includes: defining a reference vector , Composed of the maximum task fitness, it reflects the skill vector corresponding to the most ideal scheduling scheme, i.e. .individual skill vectors and The similarity is due to exist The projection length determines the skill vector corresponding to the most ideal scheduling scheme, which is used as the reference vector. and calculate individual skill vectors exist the projection length on where, denotes the 2-norm of a vector, is the dot product of two vectors; the projection length The larger the indicates the more similar the individual the corresponding scheduling scheme is to the multi-task objective optimization ability is stronger.

[0023] Further, S45 includes: S451 according to the individual's descending order, select the top individuals as elite individuals, the rest are non-elite individuals.

[0024] S452 determines whether it is an elite individual. If it is not an elite individual, execute S453, otherwise, execute S454.

[0025] S453 randomly selects an elite individual and two different individuals and , and generates a mutation operator in the first way: where, is the non-elite individual adaptive scaling factor, and respectively represent its minimum maximum value, denotes the minimum in the non-elite individual, denotes the maximum in the non-elite individual. is a minimum value, is a curvature factor, g represents the number of iterations of the current algorithm, and G represents the maximum number of iterations of the algorithm.

[0026] S454 finds the worst-performing task, i.e. the minimum element in or , and locates the individual that performs best in optimizing the task objective. Then randomly select another elite individual , and generate a mutation operator in the second way: where, is the elite individual adaptive scaling factor, and respectively represent its minimum maximum value.

[0027] S455 pair and Crossing according to element index (at this time two parameters are expressed as and ), thereby generating offspring individuals , in the following manner: wherein, indicates the element index of the individual, , is an element index randomly selected from . is the crossover rate, is the minimum value of the crossover rate, controls the degree of deviation .

[0028] S456 pair The out-of-bound elements in

[0029] S457 repeat the above S452 to S456 until individuals are generated as offspring population, and N is a positive integer which can be set by a technician.

[0030] The embodiment of the application constructs a three-layer multi-task optimization model to realize the collaborative optimization of different user target demands under limited manufacturing resources, wherein the upper and lower layers of the model optimize the individualized targets of customers and suppliers, and the middle layer optimizes the overall target of the distributed manufacturing platform. A skill vector guided multi-task optimization method is designed to solve the multi-task optimization model. First, a unified solution space is constructed by random key coding, and a new decoding method is designed to improve the utilization rate of the key. Then, a class of skill vectors is designed to evaluate the ability of the scheduling scheme to meet the target demands of different users, and the scheduling scheme is ranked based on the skill vector. Finally, an adaptive differential evolution method is designed, different mutation guidance methods are adopted for scheduling schemes with different rankings, and a dynamic crossover rate combining linear decay and periodic oscillation is designed to balance the convergence and diversity of the optimization method. Therefore, the scheduling scheme can accurately and effectively reflect the effect of optimizing different user targets, relieve the computational pressure of the scheduling and control system of the distributed manufacturing system when dealing with too many tasks and targets, and promote the improvement of the operation efficiency of the distributed manufacturing system. BRIEF DESCRIPTION OF DRAWINGS

[0031] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0032] Figure 1 is a flowchart of the optimization method of the present invention.

[0033] Figure 2 This is a schematic diagram of the random key decoding process of the present invention.

[0034] Figure 3 This is a schematic diagram illustrating the construction of skill vectors according to the present invention.

[0035] Figure 4 This is a graph showing the adaptive scaling factor for the skill in this invention.

[0036] Figure 5 This is the adaptive crossover rate curve of the present invention.

[0037] Figure 6 Examples of embodiments of the present invention Indicator comparison chart.

[0038] Figure 7 Examples of embodiments of the present invention Indicator comparison chart.

[0039] Figure 8 is a comparison chart of the optimization targets of the distributed manufacturing platform in the embodiments of the present invention.

[0040] Figure 8(a) is a schematic diagram of the distributed manufacturing platform optimization target experimental case T03S02 of the present invention; Figure 8(b) is a schematic diagram of the distributed manufacturing platform optimization target experimental case T03S04 of the present invention; Figure 8(c) is a schematic diagram of the distributed manufacturing platform optimization target experimental case T03S08 of the present invention; Figure 8(d) is a schematic diagram of the distributed manufacturing platform optimization target experimental case T05S02 of the present invention; Figure 8(e) is a schematic diagram of the distributed manufacturing platform optimization target experimental case T05S04 of the present invention; Figure 8(f) is a schematic diagram of the distributed manufacturing platform optimization target experimental case T05S08 of the present invention; Figure 8(g) is a schematic diagram of the distributed manufacturing platform optimization target experimental case T08S02 of the present invention; FIG. 8(h) is a schematic diagram of an experimental case T08S04 of the distributed manufacturing platform optimization target of the embodiment of the present application; FIG. 8(i) is a schematic diagram of an experimental case T08S08 of the distributed manufacturing platform optimization target of the embodiment of the present application; FIG. 8(j) is a schematic diagram of an experimental case T15S02 of the distributed manufacturing platform optimization target of the embodiment of the present application; FIG. 8(k) is a schematic diagram of an experimental case T15S04 of the distributed manufacturing platform optimization target of the embodiment of the present application; FIG. 8(l) is a schematic diagram of an experimental case T15S08 of the distributed manufacturing platform optimization target of the embodiment of the present application; Figure 9 is a flowchart of the overall scheme of the present application. DETAILED DESCRIPTION

[0041] In order for those skilled in the art to better understand the technical solutions of the present application, the present application will be further described in detail below in combination with the drawings and specific embodiments. In the following, embodiments of the present application will be described in detail, and examples of the embodiments are shown in the drawings, in which the same or similar reference numerals represent the same or similar elements or elements having the same or similar functions throughout. The embodiments described below by reference to the drawings are exemplary and are only used to explain the present application, and cannot be interpreted as a limitation on the present application. Those skilled in the art can understand that, unless specifically stated, the singular forms "one", "said" and "the" used herein can also include the plural forms. It should be further understood that the phrase "comprising" used in the specification of the present application means that the features, integers, steps, operations, elements and / or components exist, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or groups thereof. It should be understood that when we say that an element is "connected" or "coupled" to another element, it can be directly connected or coupled to the other element, or there can be intermediate elements. In addition, "connected" or "coupled" used herein can include wireless connection or coupling. The phrase "and / or" used herein includes any one of the associated listed items and all combinations thereof. Those skilled in the art can understand that, unless otherwise defined, all terms (including technical terms and scientific terms) used herein have the same meaning as that generally understood by those skilled in the art in the field to which the present application belongs. It should also be understood that terms such as those defined in a general dictionary should be understood to have meanings consistent with those in the context of the prior art, and should not be interpreted with idealized or overly formal meanings unless defined as such.

[0042] As Figure 1As shown, in the embodiment of the present application, the multi-task optimization method for multi-user target demand in a distributed manufacturing system proposed by the present application acts on the scheduling control system of the distributed manufacturing system, and includes the following steps: 1) A three-layer multi-task optimization model is constructed, the upper layer takes maximizing satisfaction, service quality and minimizing service cost as the target of the customer, the lower layer takes maximizing revenue as the target of the supplier, and the middle layer takes minimizing penalty cost, total cost and maximizing production flexibility, total revenue, total satisfaction, and total service quality as the target of the distributed manufacturing platform.

[0043] 2) An initial population is generated by using random secret key coding, and a new decoding method containing an initial offset and a selection interval is used to ensure uniform selection of the secret key.

[0044] 3) The results of optimizing each customer target and supplier target of the scheduling scheme are calculated, and the results are ranked respectively, and the skill vector of each individual is calculated by using reverse normalization of the ranking results of each target.

[0045] 4) The similarity between the skill vector and the reference vector is judged according to the projection length of the skill vector on the reference vector, and the multi-task optimization ability of the population individuals is distinguished based on this, and is divided into elite individuals and non-elite individuals.

[0046] 5) Based on the evolution stage, different adaptive differential evolution methods are adopted for the elite individuals and the non-elite individuals to generate a child population, guiding the elite individuals to explore the unknown solution space, guiding the non-elite individuals to approach the elite individuals, and a self-adaptive crossover rate combining linear decay and periodic oscillation is designed.

[0047] 6) The child population obtained in the above step is combined with the original population, a dynamic decomposition strategy is used to sort the distributed manufacturing platform target of the combined population, and the individuals with high ranking are selected to form the next generation population.

[0048] 7) The above operation is repeated for the new obtained population until the termination criteria are met.

[0049] Further, the service combination optimization process in the distributed manufacturing system can be described as follows: A customer submits his demand to the distributed manufacturing platform, and the platform describes the demand of the customer as a series of manufacturing tasks, i.e. , , and indexes them. Usually, each task can be decomposed into a series of sub-tasks with sequential structure, i.e. , , and indexes them. There are enterprise terminal suppliers of the enterprise terminal, and each enterprise terminal A platform provides manufacturing resources, which are encapsulated into a set, i.e. , indexed by . The platform matches the manufacturing contents of with the manufacturing capabilities of , and constructs a candidate service set , indexed by . The platform selects services for each from , and considers various constraints (such as production, transportation, etc.), so as to obtain a feasible service combination for each manufacturing task. This process involves three parties of users, i.e. customers, suppliers, and the platform, and for each type of user, their respective different objectives are considered, and a three-layer multi-task optimization (TMTO) model is constructed as follows:

[0050]

[0051]

[0052]

[0053] wherein min and max represent minimum and maximum functions respectively; , and are the satisfaction, service quality, and service cost of the customer respectively; is the profit of the supplier ; , , , , and are the penalty cost, scheduling scheme flexibility, total cost, total profit, total satisfaction, and total service quality of the platform respectively and are the upper and lower limits of the delivery time required by the customer ; and are risk attitude coefficients, and and are the penalty coefficients of delayed delivery and early delivery respectively; and ​​​They are judgments Can and whether or not The 1-0 coefficient manufactured on the surface; It is a service The pass rate; and These are services Revenue and costs per unit of time; and These are manufacturing costs and transportation costs; It is a task The workload, and It is a service Manufacturing capacity per unit time; It is the transportation cost per unit distance. Service Provider and The distance between them Supplier With customers The distance between them; and These are the unit penalty costs for early delivery and late delivery of the task, respectively; and These are services The ability to adapt to changes in manufacturing tasks and resources; yes The diversity of services, Supplier The number of terminals of partner companies, Is with The number of services with similar functions; , and These are respectively the ability to cope with changes in manufacturing tasks by relying on service diversity, other manufacturing resources, and the terminal of partner companies; , and They are respectively relying on services The reliability of similar manufacturing resource services and the ability of partner companies' end-users to adapt to changes in manufacturing resources; and These are tasks The start and end times.

[0054] Furthermore, an initial population is generated using random key encoding, and a unique decoding method is designed. The specific process is as follows: For each individual , defines it by express, Depend on indivual The random variable consists of .

[0055] As shown in Figure 2 , the secret key is selected from as the starting point, and is selected as the interval until keys are selected as the encoding of the task . If the selected position exceeds the length of , select from the beginning. This method avoids the fact that the elements at the front of are more likely to be selected, and the elements in can be fully utilized. . and are calculated as follows:

[0056] The manufacturing service assigned to is represented by the mapping , where is the value of the th position in the selected secret key. Assuming is 6, is 0.7, , this means that the 5th manufacturing resource in is selected to provide services for .

[0057] After selecting the manufacturing service for each subtask , the execution order of each subtask is determined according to the first-in-first-out principle (FIFO). First, execute the subtask with the earliest ideal start time, if multiple subtasks have the same start time, execute the one with the most subsequent subtasks. If the number of subsequent subtasks is also the same, execute the subtask with the shortest required time first.

[0058] Further, a skill vector is designed to evaluate the multi-task optimization ability of different scheduling schemes. As shown in Figure 3 , in the multi-task optimization scenario, individuals oriented to single-task optimization perform well in a certain task, but easily lead to loss of population diversity and imbalance between tasks. Global mediocre individuals, although ensuring wide task coverage, have low optimization efficiency. The skill vector of an individual is represented as , where the element represents the task fitness, reflecting the performance of in optimizing the customer 's goal, and the element reflects in optimizing the supplier performance of the target aspect. and are determined by performing a reverse Min-Max normalization on the target ranking of all individuals, which is calculated as follows: wherein, and respectively represent the minimum and maximum values of the ranking of the corresponding target. The ranking of the target aspect of the optimization and , while and respectively represent the minimum and maximum values of the ranking of the corresponding target.

[0059] On the basis of defining the skill vector of the individual, a reference vector is defined, which is composed of the maximum value of the task fitness and reflects the skill vector corresponding to the most ideal scheduling scheme, i.e. The similarity of the skill vector of the individual to is determined by the projection length of on , and the calculation formula is as follows: wherein, represents the 2-norm of the vector, is the dot product of the two vectors. The greater the projection length , the more similar and are, and the stronger the optimization ability of the scheduling scheme corresponding to the individual to the multi-task target is.

[0060] Further, an adaptive genetic mechanism based on differential evolution is designed to generate the next generation of population, and the specific steps are as follows: According to the skill of the individual , the population is arranged in descending order, and the first individuals are selected as elite individuals, and the rest are classified as non-elite individuals. Determine whether is an elite individual. If it is not an elite individual, step 3 is executed, otherwise, step 4 is executed. Randomly select an elite individual and two different individuals and , and generate a mutation operator as follows: wherein, is a non-elite individual adaptive scaling factor, and respectively represent the minimum and maximum values. is a very small value of 1e-10, is a curvature factor. The adaptive scaling factor of the non-elite individual The design aims to accelerate the optimization of non-elite individuals while maintaining exploration capabilities; therefore, it is based on elite individuals. Build To guide optimization. The design is based on the sigmoid function, which guarantees... The value of gradually decreases over time, which can be used in the early stages of evolution. and Significant differences exist for wide-area search; later, in... Detailed exploration of the vicinity is conducted to avoid premature convergence. The exponential term is nonlinearly adjusted using the tangent function to ensure that... hour Following the trend of the sigmoid function, Ensure the domain of the tangent function. For example... Figure 4 As shown in part (a), when the curvature factor When the value is large, the algorithm emphasizes early-stage diversity and later-stage convergence, while when the value is small... It can better balance diversity and convergence. Figure 4 In part (b), Smaller individuals have smaller... These individuals perform poorly in multi-task optimization, which allows them to perform better in elite individuals. With the help of [the system / mechanism], optimization can be performed faster, accelerating the overall convergence of the population.

[0061] turn up Optimize the worst-performing task, i.e. The smallest element in the middle corresponds to or And identify the individual who performs best when optimizing the task objective. Then another elite individual was randomly selected. Generate the mutation operator as follows: : ,in An adaptive scaling factor for elite individuals. and These represent their minimum and maximum values, respectively. (Elite individuals'...) The design aims to explore more solution space regions while maintaining individual stability, therefore based on Self-construction . Design concept and Similar, but with a larger value, which helps... Escape the local area, but avoid deviating too far from yourself.

[0062] right and Crossing to generate offspring individuals , the generation is as follows: , wherein, represents the element index of the individual, , is an element index randomly selected from the range of , which ensures that at least one element comes from . is the crossover rate, is the minimum value of the crossover rate, controls the degree of deviation from . As shown in Figure 5 , the thus designed has linear decay and periodic oscillation characteristics, and the amplitude and period gradually decrease with the running of the algorithm. The early is large and changes slowly, which is beneficial to improve the optimization probability and population diversity; with the increase of the number of iterations, gradually decreases and the amplitude decreases and the frequency increases, which prevents premature convergence while maintaining population quality.

[0063] The out-of-bound elements in are processed as follows: Repeat the above steps until individuals are generated as the offspring population.

[0064] Further, the initial population and the obtained offspring population are merged, and for each individual in the merged population, the result of optimizing the distributed manufacturing platform target is calculated, and the optimization results are multi-objective sorted. According to the ranking, the top individuals are selected to form a new generation population.

[0065] Specific experimental cases: The number of customers has 4 possible values, i.e. , the number of suppliers has 3 possible values, i.e. . Therefore, there are 12 different scale experimental cases, each of which is represented as , for example, means a case with 8 customers and 4 suppliers. The values of the remaining parameters are randomly generated within the range specified in Table 1.

[0066] Table 1

[0067] Due to the large number of customers and suppliers, it is not feasible to list all their targets for direct comparison, so the invention defines the following two indicators to evaluate the ability to simultaneously optimize the targets of customers and suppliers. In contrast, the platform's targets are easy to compare directly. wherein, denotes the execution count of the algorithm, is its index. is the customer target quantity, is its index. is the provider optimization target quantity, denotes its index. denotes the individual achieved in the th run, th target value of , while and denote the worst and best target values over all runs, respectively. Similarly, denotes the individual achieved in the th run, th target value of , and also denote the worst and best target values over all runs, respectively. and denote the average of the th target of and the th target of , respectively, and denote the ranking of these average target values. The higher or the earlier, the better the algorithm.

[0068] Since the present invention is aimed at a multi-level optimization model, general multi-objective optimization algorithms or multi-task optimization algorithms cannot directly solve such complex models, so the method of the present invention is compared with a class of extended multi-factor optimization algorithms (EMFEA), each experiment is independently run 10 times. The random mating probability of EMFEA is set to 0.7, the population selection probability is set to 0.15, and the population size and maximum evolution stage are consistent with the method in the present invention. Figure 6 shows the comparison results of obtained by the two methods. It can be seen that the method proposed in the present invention has achieved higher average and median values in all 12 experimental cases, and has a significant advantage over the results of EMFEA. This shows that the scheduling scheme obtained by the method of the present invention performs better in optimizing the objectives of service demanders and providers. In addition, the The standard deviation of the mean is usually smaller than the corresponding standard deviation of EMFEA, which indicates that the proposed method provides better solution stability. Figure 7 The comparison results of the two methods in solving the obtained are shown. In addition to the experimental cases , the mean values of the objectives obtained by the method of the present application are better than EMFEA in other experimental cases. This indicates that the present application can effectively achieve more collaborative optimization of service demander and provider objectives. Figures 8(a) to 8(l) show the results obtained by the two methods in optimizing the objectives of the distributed manufacturing platform. In all 12 experimental cases, the method of the present application achieved the best results in terms of the mean values of at least five objectives. In addition, the method of the present application also achieved the best values of all objectives in experimental cases , and . These results show that the method of the present application can obtain scheduling schemes that can effectively balance the interests of different users, achieving collaborative optimization of platform, customer and supplier objectives.

[0069] The present application proposes a multi-task optimization method for the demand of multiple user objectives in a distributed manufacturing system, which can effectively solve the problem that different user demands cannot be fully considered in traditional distributed manufacturing task allocation methods. By combining a three-layer multi-task optimization model, random secret key coding, skill vector representation and an adaptive differential evolution strategy, the method not only effectively solves the dimension disaster problem when considering the individualized needs of customers, suppliers and platforms at the same time, but also achieves efficient collaborative optimization of different user objectives, ensuring the interests of all parties involved in the distributed manufacturing system, and has high application value and practicality.

[0070] Each of the embodiments in the specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other. Each embodiment focuses on the difference from other embodiments. In particular, the device embodiments are described more simply because they are basically similar to the method embodiments, and the relevant parts can be referred to the part of the method embodiments. The above is merely a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any changes or replacements within the technical range disclosed by the present application can be easily thought by those skilled in the art, which should be covered in the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A multi-task optimization method for multi-user objective requirements in a distributed manufacturing system, characterized in that, include: S1. The distributed manufacturing platform receives manufacturing demand information uploaded by the client terminal and establishes a set of manufacturing tasks, wherein each manufacturing task in the set of manufacturing tasks is further decomposed into at least one sub-task. S2. Accept manufacturing resource information uploaded by the supplier's enterprise terminal and establish a manufacturing resource set; S3. Match the manufacturing task set with the manufacturing resource set to obtain a candidate service set for the corresponding manufacturing task; S4. From the candidate service set, determine the service information that matches the sub-task of the manufacturing task in the manufacturing task set, obtain the feasible service combination corresponding to the manufacturing task, and then send the order information to the corresponding enterprise terminal according to the obtained feasible service combination.

2. The method according to claim 1, characterized in that, In S4, it includes: S41. Establish a three-layer multi-task optimization model, wherein the individuals in the upper layer of the three-layer multi-task optimization model correspond to the client terminal, and the individuals in the lower layer correspond to the enterprise terminal. S42. In the three-layer multi-task optimization model, the initial population is generated by random key encoding and then the key is selected uniformly. S43. Optimize individual customer and supplier objectives and establish skill vectors for various groups of individuals; S44. Use the projection length of the skill vector onto the reference vector to determine similarity, and divide the population individuals into elite individuals and non-elite individuals based on the similarity determination results. S45. Perform adaptive differential evolution on elite individuals and non-elite individuals respectively, and generate offspring populations; S46. Merge the generated offspring population with the original population, sort the distributed manufacturing platform targets of the merged population, and select the top-ranked individuals to form the next generation population.

3. The method according to claim 1 or 2, characterized in that, A set of manufacturing tasks consists of a series of manufacturing tasks corresponding to the manufacturing requirements of each customer. , The index of T; Decomposed into sequential structures Each subtask is represented as , for index, Indicates the first Sub-tasks; Total Enterprise terminals in a supplier company Provide to the platform Each manufacturing resource, the set of manufacturing resources is , for index, This represents the m-th manufacturing service in the set of manufacturing resources.

4. The method according to claim 3, characterized in that, In S41, the established three-layer multi-task optimization model includes: the personalized target model of the upper-layer customer, the personalized target model of the lower-layer supplier, and the overall target model of the middle-layer distributed manufacturing platform. The personalized target model for the upper-level customer includes: Where min and max represent the minimum and maximum value functions, respectively; , and Representing customers respectively Satisfaction, service quality, and service cost; and Representing customers respectively The upper and lower limits of the required delivery time. Indicates customer Required delivery time; and These represent two risk attitude coefficients, respectively. and These represent the penalty coefficients for delayed and early delivery of the task, respectively. It is a test Is it possible to The 1-0 coefficient manufactured on the surface; It is a decision Is it in The 1-0 coefficient manufactured on the surface; Manufacturing services The pass rate; and These are manufacturing costs and transportation costs; Manufacturing services Revenue per unit of time; It is a task The workload, It is a service Manufacturing capacity per unit time; Indexes of the two service providers involved in the mission transportation; Indicates the first Sub-tasks Can it be done? The 1-0 coefficient manufactured on the surface; Indicates the first Sub-tasks Whether to decide on The 1-0 coefficient manufactured on the surface; It is the transportation cost per unit distance. Service Provider and The distance between them Supplier With customers The distance between them; Indicates the first The last subtask Can it be done? The 1-0 coefficient manufactured on the surface; Indicates the first The last subtask Whether to decide on The 1-0 coefficient manufactured above; the personalized target model of the lower-level supplier, including: in, Indicates supplier Profits; Subtasks Manufacturing services The time taken to complete; Manufacturing services Cost per unit of time.

5. The method according to claim 4, characterized in that, The overall target model of the intermediate layer distributed manufacturing platform includes: Among them, PC, F, C, B, S and Q represent the platform's penalty cost, scheduling scheme flexibility, total cost, total profit, total satisfaction and total service quality, respectively. and These are the unit penalty costs for early delivery and late delivery of the task, respectively; and They are The ability to adapt to changes in manufacturing tasks and resources; , and These respectively represent the ability to adapt to changes in manufacturing tasks by relying on service diversity, other manufacturing resources, and partner companies; yes The diversity of services, Supplier The number of partner companies Is with The number of services with similar functions; , and They are respectively relying on The reliability of the facilities, similar manufacturing resource services, and the ability of partner companies to adapt to changes in manufacturing resources.

6. The method according to claim 5, characterized in that, The constraints of the three-layer multi-task optimization model include: in, Indicates the first Sub-tasks, namely The start time, express Completion time, express Actual completion time Indicates manufacturing services Available service hours Indicates the first The transportation time for each sub-task.

7. The method according to claim 2, characterized in that, S43 includes: Establish individual skill vectors ,in, Indicates task fitness. reflect Optimize suppliers Performance in terms of the target aspect; , and They represent In optimization and The ranking of the target aspects, and and These represent the minimum and maximum ranking values ​​for the corresponding targets, respectively.

8. The method according to claim 7, characterized in that, S44 includes: using the skill vector corresponding to the most ideal scheduling scheme as a reference vector. and calculate individual skill vectors exist Projected length ,in, Describes the 2-norm of a vector. The dot product of the two vectors; the projected length The larger the value, the more likely it is to indicate and The more similar the individuals The stronger the corresponding scheduling scheme is in optimizing multi-task objectives, the better.

Images