Digital-analog dual-driven intelligent scheduling method and device across organizational machine workshops

Abstract

The application relates to a digital-analog dual-drive intelligent scheduling method and device across organizational machine workshops, wherein the method comprises the following steps: establishing a scheduling model across organizational machine workshops and performing vector coding on the scheduling model to obtain factory, process and machine vectors; a first and a second parent population are generated through a collaborative initialization strategy, a first child population is obtained by genetic evolution search on the first parent population, and a second child population is obtained by data-driven neighborhood search and recording of search knowledge; new first and second parent populations are generated according to the parent and child populations, and search knowledge of the new first parent population is obtained by multi-operator collaborative neighborhood search; a one-dimensional tensor is obtained by splicing the factory vector and the like, and a corresponding scheduling scheme is obtained in combination with a neighborhood action recommendation neural network model. Therefore, the problems that the existing cross-organizational workshop scheduling method is difficult to realize fast and efficient search while recording historical search experience, and the solving efficiency of the resource grouping complex machining workshop scheduling problem is low and the like are solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of cross-organizational workshop scheduling, and in particular relates to a digital-analog dual-drive intelligent scheduling method and device for cross-organizational machine workshops. BACKGROUND

[0002] A machine workshop is an important production environment in the process route of an aviation product, and the machine workshop is responsible for cutting and processing the hot-pressed composite material into the required shape and size. In modern manufacturing enterprises, a multi-park distributed manufacturing mode is adopted, an order is placed at the headquarters, a command center commands a transport team to send raw materials from a warehouse to different parks, the processing capacity of different parks is different, but they can be processed in parallel, in order to improve the production capacity, the enterprise selects to distribute raw materials to multiple parks for processing, which increases the complexity of the problem.

[0003] The existing technology considering the cross-organizational workshop scheduling problem still has obvious deficiencies in the method level.

[0004] In terms of method, the solution methods of distributed workshop scheduling problems can be divided into data-driven and knowledge-driven. Among them, the knowledge-driven algorithm generally uses intelligent optimization algorithm combined with knowledge-driven local search operation, which can realize fast global search, but cannot record the search knowledge of each generation, and the algorithm directly eliminates the old population after the population update iteration is completed, but the eliminated population contains historical search experience, which can effectively utilize it to provide knowledge for subsequent search of the algorithm; the data-driven algorithm generally uses reinforcement learning algorithm, which uses neural network to store the search strategy of each generation, but this kind of algorithm has slow search speed, random strategy search direction and lack of guidance.

[0005] In summary, the existing cross-organizational workshop scheduling method is difficult to realize fast and efficient search while recording historical search experience, and the solution efficiency of the resource grouping complex machine processing workshop scheduling problem is low, which needs to be solved urgently. SUMMARY

[0006] The present application provides a digital-analog dual-drive intelligent scheduling method and device for cross-organizational machine workshops to solve the problems that the existing cross-organizational workshop scheduling method is difficult to realize fast and efficient search while recording historical search experience, and the solution efficiency of the resource grouping complex machine processing workshop scheduling problem is low.

[0007] The first aspect embodiment of the application provides a digital-analog dual-drive intelligent scheduling method across organizational machine workshops, including the following steps: modeling an actual machine workshop production scene, establishing a cross-organizational machine workshop scheduling model corresponding to the machine workshop production scene, and vector encoding the cross-organizational machine workshop scheduling model to obtain a factory vector, a process vector and a machine vector corresponding to the cross-organizational machine workshop scheduling model; generating a first parent population and a second parent population corresponding to the cross-organizational machine workshop scheduling model according to a preset collaborative initialization strategy, and performing a genetic evolution search operation on the first parent population to obtain a first child population of the first parent population, while based on the second parent population and a preset data-driven neighborhood search strategy, obtaining a second child population and recording first search knowledge corresponding to the second child population; based on the first parent population and the second child population, generating a new first parent population, while based on the second parent population and the first child population, obtaining a new second parent population, and performing a multi-operator collaborative neighborhood search operation on the new first parent population to obtain second search knowledge corresponding to the new first parent population; horizontally splicing the factory vector, the process vector and the machine vector according to the first search knowledge and the second search knowledge to obtain a one-dimensional tensor, and inputting the one-dimensional tensor into a pre-trained neighborhood action recommendation neural network model to obtain a scheduling scheme corresponding to the machine workshop production scene that meets a preset requirement.

[0008] Optionally, in an embodiment of the application, the preset collaborative initialization strategy includes: selecting a parallel machine with the minimum processing time for each workpiece in the cross-organizational machine workshop scheduling model; or, accumulating the processing time of each workpiece in different factories to obtain a cumulative value corresponding to each workpiece, and arranging the cumulative values of all workpieces in descending order to generate a cumulative value sequence, so as to distribute each workpiece to the corresponding factory according to the cumulative value sequence; or, selecting a parallel machine with the minimum completion time in all factories for each workpiece; or, arranging each process in ascending order according to the delivery period of each process in the cross-organizational machine workshop scheduling model to obtain a process sequence, and distributing each workpiece to the corresponding factory based on the process sequence and a preset factory code.

[0009] Optionally, in an embodiment of the present application, the generating the first parent population and the second parent population corresponding to the cross-organizational machine shop scheduling model according to a preset cooperative initialization strategy comprises: calculating total tardiness and maximum completion time corresponding to the machine shop production scenario according to the factory vector, the process vector and the machine vector; initializing a plurality of target sub-populations randomly, and adjusting the factory vector, the process vector and the machine vector corresponding to each target sub-population in the plurality of target sub-populations through any initialization strategy in a plurality of preset initialization strategies; and merging the plurality of target sub-populations according to the adjusted factory vector, the process vector and the machine vector to obtain the first parent population and the second parent population.

[0010] Optionally, in an embodiment of the present application, the performing a genetic evolution search operation on the first parent population to obtain a first sub-population of the first parent population comprises: obtaining a mating pool corresponding to the first parent population based on a two-person tournament algorithm, and selecting a parent randomly from the mating pool through each individual in the first parent population; exchanging process processing sequences of the cross-organizational machine shop scheduling model according to a process-based crossover strategy and the parent, and exchanging machine selection information and factory information of the cross-organizational machine shop scheduling model by using a general crossover strategy to generate a first sub-population of the first parent population.

[0011] Optionally, in an embodiment of the present application, the preset data-driven neighborhood search strategy comprises: finding a key job with maximum tardiness in the cross-organizational machine shop scheduling model, and searching a first target job meeting a preset delivery period requirement, exchanging positions of the key job and the first target job, or inserting the key job into a position in front of the first target job; or randomly selecting any job in the cross-organizational machine shop scheduling model, and searching a second target job meeting the preset delivery period requirement, and exchanging positions of the second target job and the any job; or finding a key factory and a key path corresponding to the key factory of the cross-organizational machine shop scheduling model according to the maximum completion time, and randomly selecting a key process, and scanning completion times of all machines to obtain a target machine meeting a preset completion time requirement, so as to insert the key process into the target machine.

[0012] Optionally, in an embodiment of the present application, the one-dimensional tensor is input into a pre-trained neighborhood action recommendation neural network model to obtain a scheduling scheme corresponding to the machining plant production scene and satisfying a preset requirement, comprising: extracting a current state of the one-dimensional tensor, and inputting the current state into an evaluation network of the neighborhood action recommendation neural network model to obtain a Q value of the current state, while determining a Q value of a current policy according to the selected neighborhood action; extracting a next state of the one-dimensional tensor, and inputting the next state into a target network of the neighborhood action recommendation neural network model to obtain Q values of all actions of the next state; based on the Q values of all actions of the next state, obtaining a maximum Q value in all actions of the next state, and calculating a target value according to the maximum Q value, a preset discount factor and reward information of the one-dimensional tensor; calculating a root mean square error of the neighborhood action recommendation neural network model according to the Q value of the current action and the target value, and optimizing the evaluation network parameters through a preset back propagation algorithm, while updating the target network parameters using the optimized evaluation network parameters in a case where the target network parameters satisfy a preset update requirement; adjusting the root mean square error through the updated target network parameters, and obtaining the scheduling scheme corresponding to the machining plant production scene and satisfying the preset requirement in a case where the root mean square error satisfies a preset error condition.

[0013] The second aspect embodiment of the application provides a digital-analog dual-drive intelligent scheduling device across organization machine workshops, comprising: a modeling module, configured to model an actual machine workshop production scene, establish a cross-organization machine workshop scheduling model corresponding to the machine workshop production scene, and perform vector coding on the cross-organization machine workshop scheduling model to obtain a factory vector, a process vector and a machine vector corresponding to the cross-organization machine workshop scheduling model; a data-driven module, configured to generate a first parent population and a second parent population corresponding to the cross-organization machine workshop scheduling model according to a preset collaborative initialization strategy, perform a genetic evolution search operation on the first parent population to obtain a first child population of the first parent population, and obtain a second child population based on the second parent population and a preset data-driven neighborhood search strategy, and record first search knowledge corresponding to the second child population; a knowledge-driven module, configured to generate a new first parent population based on the first parent population and the second child population, obtain a new second parent population according to the second parent population and the first child population, and perform a multi-operator collaborative neighborhood search operation on the new first parent population to obtain second search knowledge corresponding to the new first parent population; and a splicing module, configured to horizontally splice the factory vector, the process vector and the machine vector to obtain a one-dimensional tensor according to the first search knowledge and the second search knowledge, and input the one-dimensional tensor into a pre-trained neighborhood action recommendation neural network model to obtain a scheduling scheme corresponding to the machine workshop production scene and satisfying a preset requirement.

[0014] Optionally, in an embodiment of the application, the data-driven module comprises: a first initialization strategy unit configured to select a parallel machine with the minimum processing time for each workpiece in the cross-organization machine workshop scheduling model; or a second initialization strategy unit configured to accumulate the processing time of each workpiece in different factories to obtain a cumulative value corresponding to each workpiece, arrange the cumulative values of all workpieces in descending order to generate a cumulative value sequence, and distribute each workpiece to a corresponding factory according to the cumulative value sequence; or a third initialization strategy unit configured to select a parallel machine with the minimum completion time in all factories for each workpiece; or a fourth initialization strategy unit configured to arrange each process in ascending order according to the delivery period of each process in the cross-organization machine workshop scheduling model to obtain a process sequence, and distribute each workpiece to a corresponding factory based on the process sequence and a preset factory code.

[0015] Optionally, in an embodiment of the present application, the data driving module further comprises: a calculation unit, configured to calculate total tardiness and maximum completion time corresponding to the machine shop production scene according to the factory vector, the process vector and the machine vector; an adjustment unit, configured to randomly initialize a plurality of target sub-populations, and adjust the factory vector, the process vector and the machine vector corresponding to each target sub-population in the plurality of target sub-populations through any initialization strategy in a plurality of preset initialization strategies; a merging unit, configured to merge the plurality of target sub-populations according to the adjusted factory vector, the process vector and the machine vector, to obtain the first parent population and the second parent population.

[0016] Optionally, in an embodiment of the present application, the data driving module further comprises: a selection unit, configured to obtain a mating pool corresponding to the first parent population based on a two-person tournament algorithm, and randomly select a parent from the mating pool through each individual in the first parent population; an exchange unit, configured to exchange process processing sequences of the cross-organizational machine shop scheduling model according to a process-based crossover strategy and the parent, and exchange machine selection information and factory information of the cross-organizational machine shop scheduling model by using a general crossover strategy, to generate a first offspring population of the first parent population.

[0017] Optionally, in an embodiment of the present application, the data driving module further comprises: a first neighborhood searching unit, configured to find a critical job with maximum tardiness in the cross-organizational machine shop scheduling model, and search a first target job meeting a preset delivery time requirement, exchange positions of the critical job and the first target job, or insert the critical job into a position in front of the first target job; or a second neighborhood searching unit, configured to randomly select any job in the cross-organizational machine shop scheduling model, and search a second target job meeting the preset delivery time requirement, and exchange positions of the second target job and the any job; or a third neighborhood searching unit, configured to find a critical factory and a critical path corresponding to the critical factory of the cross-organizational machine shop scheduling model according to the maximum completion time, randomly select a critical process, and scan completion times of all machines to obtain a target machine meeting a preset completion time requirement, to insert the critical process into the target machine.

[0018] Optionally, in an embodiment of the present application, the splicing module comprises: a first extraction unit configured to extract a current state of the one-dimensional tensor and input the current state into an evaluation network of the neighborhood action recommendation neural network model to obtain a Q value of the current state, and determine a Q value of a current policy according to the selected neighborhood action; a second extraction unit configured to extract a next state of the one-dimensional tensor and input the next state into a target network of the neighborhood action recommendation neural network model to obtain Q values of all actions of the next state; an acquisition unit configured to acquire a maximum Q value among all actions of the next state based on the Q values of all actions of the next state, and calculate a target value according to the maximum Q value, a preset discount factor and reward information of the one-dimensional tensor; an optimization unit configured to calculate a root mean square error of the neighborhood action recommendation neural network model according to the Q value of the current action and the target value, and optimize the evaluation network parameters through a preset back propagation algorithm, and update the target network parameters by using the optimized evaluation network parameters in a case where the target network parameters meet a preset update requirement; and an adjustment unit configured to adjust the root mean square error by using the updated target network parameters, and obtain a scheduling scheme that meets a preset requirement for the production scene of the machine shop in a case where the root mean square error meets a preset error condition.

[0019] An electronic device is provided in a third aspect of embodiments of the present application, and includes a memory, a processor, and a computer program stored in the memory and executable on the processor. The processor executes the program to implement the intelligent scheduling method of a digital-analog dual drive across machine shops of different organizations as described in the above embodiments.

[0020] A computer readable storage medium is provided in a fourth aspect of embodiments of the present application, and stores a computer program. The program is executed by a processor to implement the intelligent scheduling method of a digital-analog dual drive across machine shops of different organizations as described above.

[0021] Therefore, the embodiments of the present application have the following beneficial effects:

[0022] Embodiments of the present application can model an actual machine shop production scene, establish a cross-organizational machine shop scheduling model corresponding to the machine shop production scene, and vector encode the cross-organizational machine shop scheduling model to obtain a factory vector, a process vector, and a machine vector corresponding to the cross-organizational machine shop scheduling model; generate a first parent population and a second parent population corresponding to the cross-organizational machine shop scheduling model according to a preset collaborative initialization strategy, and perform a genetic evolution search operation on the first parent population to obtain a first child population of the first parent population, while obtaining a second child population based on the second parent population and a preset data-driven neighborhood search strategy, and recording first search knowledge corresponding to the second child population; generate a new first parent population based on the first parent population and the second child population, and obtain a new second parent population according to the second parent population and the first child population, and perform a multi-operator collaborative neighborhood search operation on the new first parent population to obtain second search knowledge corresponding to the new first parent population; horizontally splice the factory vector, the process vector, and the machine vector to obtain a one-dimensional tensor according to the first search knowledge and the second search knowledge, and input the one-dimensional tensor into a pre-trained neighborhood action recommendation neural network model to obtain a scheduling scheme corresponding to the machine shop production scene that meets a preset requirement. The present application models the production scene of an actual machine shop as a cross-organizational machine shop scheduling problem by considering various resource configurations and constraint conditions, thereby improving the manufacturing capacity of an aerospace product in a machine shop from the perspective of resource optimization, improving the solving efficiency of a resource grouping complex machine shop scheduling problem, and reducing the manufacturing cost of the aerospace product. Thus, the existing cross-organizational shop scheduling method is difficult to achieve fast and efficient search while recording historical search experience, and the solving efficiency of the resource grouping complex machine shop scheduling problem is low.

[0023] Additional aspects and advantages of the present application will be in part apparent and in part pointed out hereinafter. BRIEF DESCRIPTION OF DRAWINGS

[0024] The above and / or additional aspects and advantages of the present application will become apparent and be readily appreciated from the following description, including the accompanying drawings, wherein:

[0025] Figure 1 A flowchart of a cross-organizational machine shop digital model double-drive intelligent scheduling method according to an embodiment of the present application;

[0026] Figure 2 A Gantt chart of a cross-organizational machine shop scheduling model according to an embodiment of the present application;

[0027] Figure 3 An N1-N3 neighborhood structure diagram according to an embodiment of the present application;

[0028] Figure 4 A schematic diagram of an N4 neighborhood structure is provided for an embodiment of the present application;

[0029] Figure 5 A schematic diagram of a feedforward neural network is provided for an embodiment of the present application;

[0030] Figure 6 A schematic diagram of a deep Q network algorithm training process is provided for an embodiment of the present application;

[0031] Figure 7 A schematic diagram of the execution logic of a digital-analog dual-driven intelligent scheduling method for cross-organizational machine workshops is provided for an embodiment of the present application;

[0032] Figure 8 A schematic diagram of the logic architecture of a digital-analog dual-driven intelligent scheduling method for cross-organizational machine workshops is provided for an embodiment of the present application;

[0033] Figure 9 A schematic diagram of a digital-analog dual-driven intelligent scheduling device for cross-organizational machine workshops is provided for an embodiment of the present application;

[0034] Figure 10 A schematic diagram of the structure of an electronic device is provided for an embodiment of the present application.

[0035] Among them, 10- the digital-analog dual-driven intelligent scheduling device for cross-organizational machine workshops; 100- the modeling module, 200- the data-driven module, 300- the knowledge-driven module, 400- the splicing module; 1001- the memory, 1002- the processor, 1003- the communication interface. DETAILED DESCRIPTION

[0036] The embodiments of the present application are described in detail below, examples of which are shown in the accompanying drawings, wherein 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 referring to the accompanying drawings are exemplary and are intended to explain the present application, and cannot be understood as a limitation of the present application.

[0037] A cross-organizational machine shop numerical-analogical dual-drive intelligent scheduling method and device of embodiments of the present application are described below with reference to the accompanying drawings. In view of the problems mentioned in the above background art, the present application provides a cross-organizational machine shop numerical-analogical dual-drive intelligent scheduling method. In the method, the actual machine shop production scene is modeled, a cross-organizational machine shop scheduling model corresponding to the machine shop production scene is established, and the cross-organizational machine shop scheduling model is vector encoded to obtain a factory vector, a process vector, and a machine vector corresponding to the cross-organizational machine shop scheduling model. According to a preset collaborative initialization strategy, a first parent population and a second parent population corresponding to the cross-organizational machine shop scheduling model are generated, and a genetic evolution search operation is performed on the first parent population to obtain a first child population of the first parent population. Meanwhile, based on the second parent population and a preset data-driven neighborhood search strategy, a second child population is obtained, and a first search knowledge corresponding to the second child population is recorded. Based on the first parent population and the second child population, a new first parent population is generated, and according to the second parent population and the first child population, a new second parent population is obtained. A multi-operator collaborative neighborhood search operation is performed on the new first parent population to obtain a second search knowledge corresponding to the new first parent population. According to the first search knowledge and the second search knowledge, the factory vector, the process vector, and the machine vector are horizontally spliced to obtain a one-dimensional tensor, and the one-dimensional tensor is input into a pre-trained neighborhood action recommendation neural network model to obtain a scheduling scheme corresponding to the machine shop production scene that satisfies the preset requirements. The present application models the actual machine shop production scene as a cross-organizational machine shop scheduling problem by considering various resource configurations and constraint conditions, thereby improving the manufacturing capacity of the aircraft product in the machine shop from the perspective of resource optimization, improving the solving efficiency of the resource grouping complex machine shop scheduling problem, and reducing the manufacturing cost of the aircraft product. Thus, the existing cross-organizational shop scheduling method is difficult to achieve fast and efficient search while recording historical search experience, and the solving efficiency of the resource grouping complex machine shop scheduling problem is low.

[0038] Specifically, Figure 1 A flowchart of a cross-organizational machine shop numerical-analogical dual-drive intelligent scheduling method provided by embodiments of the present application.

[0039] As Figure 1 shown, the cross-organizational machine shop numerical-analogical dual-drive intelligent scheduling method includes the following steps:

[0040] In step S101, the actual machine shop production scene is modeled, a cross-organizational machine shop scheduling model corresponding to the machine shop production scene is established, and the cross-organizational machine shop scheduling model is vector encoded to obtain a factory vector, a process vector, and a machine vector corresponding to the cross-organizational machine shop scheduling model.

[0041] Those skilled in the art should understand that, in order to improve the manufacturing capacity of the aviation product in the machining workshop from the perspective of resource optimization, the embodiment of the present application models the actual machining workshop production scene, considers various resource configurations and constraint conditions, and finally models the actual machining workshop production scene into a cross-organizational machining workshop scheduling model.

[0042] It should be noted that, as shown in Figure 2 The cross-organizational machining workshop scheduling model in the embodiment of the present application can be described as follows: N workpieces need to be processed, each workpiece has only one process, i.e., a machining process; there are M parallel machines in the workshop in total, and there are F groups, each group is responsible for using M f parallel machines, and each parallel machine needs one worker to start processing; the optimization goal of the cross-organizational machining workshop scheduling model is to minimize the total tardiness (TT) and the maximum completion time C max , and the calculation expression is:

[0043]

[0044]

[0045] , wherein C i represents the completion time of order i; D i represents the delivery period.

[0046] In addition, the embodiment of the present application can also use a three-layer vector coding strategy to vector code the cross-organizational machining workshop scheduling model to cover all solutions of the solution space of the model. Specifically, the embodiment of the present application can first use permutation of 1-N to represent the order of the workpieces, for example, [5, 3, 1, 2, 4] represents the processing order of workpieces 1-5; secondly, a factory vector with a length of N is used to represent the group selection information of the workpieces, and each dimension marks the grouping information of the workpieces, for example, [1, 2, 1, 2, 2] represents the grouping information of workpieces 1-5; finally, a machine vector with a length of N is used to represent the machine selection of the workpieces, for example, [2, 1, 2, 2, 1] represents the machine selection information of workpieces 1-5, wherein workpiece 1 selects machine 2.

[0047] Therefore, the embodiment of the present application models the actual machining workshop production scene into a cross-organizational machining workshop scheduling model, and vector codes the cross-organizational machining workshop scheduling model, thereby providing reliable data basis for subsequent execution of multi-strategy collaborative initialization and other operations.

[0048] In step S102, a first parent population and a second parent population corresponding to the cross-organizational machine shop scheduling model are generated according to a preset cooperative initialization strategy, a genetic evolution search operation is performed on the first parent population to obtain a first offspring population of the first parent population, and a second offspring population is obtained based on the second parent population and a preset data-driven neighborhood search strategy, and first search knowledge corresponding to the second offspring population is recorded.

[0049] Further, the embodiment of the present application also generates a first parent population and a second parent population corresponding to the cross-organizational machine shop scheduling model through a cooperative initialization strategy. In addition, the embodiment of the present application also sets a search strategy of genetic evolution search and knowledge-driven local search to quickly converge in the solution space.

[0050] In actual execution process, the embodiment of the present application can perform a genetic evolution search operation on the first parent population to obtain a first offspring population of the first parent population, and the embodiment of the present application also needs to perform a data-driven neighborhood search processing on the second parent population to generate a second offspring population, and store first search knowledge corresponding to the second offspring population.

[0051] Optionally, in an embodiment of the present application, the preset cooperative initialization strategy includes: selecting a parallel machine with minimum processing time for each workpiece in the cross-organizational machine shop scheduling model; or, accumulating the processing time of each workpiece in different factories to obtain a cumulative value corresponding to each workpiece, and arranging the cumulative values of all workpieces in descending order to generate a cumulative value sequence, and distributing each workpiece to the corresponding factory according to the cumulative value sequence; or, selecting a parallel machine with minimum completion time in all factories for each workpiece; or, arranging each process in ascending order according to the delivery period of each process in the cross-organizational machine shop scheduling model to obtain a process sequence, and distributing each workpiece to the corresponding factory based on the process sequence and a preset factory code.

[0052] It should be noted that the cooperative initialization strategy in the embodiment of the present application includes the following four initialization strategy forms:

[0053] 1. Selecting a parallel machine with minimum processing time for all workpieces to reduce the completion time by reducing the processing time;

[0054] 2. Accumulating the processing time of all workpieces in different factories, arranging the cumulative values in descending order, and distributing them to different factories for processing according to the order to ensure that machines with long processing time are processed first;

[0055] 3. Selecting a parallel machine with minimum completion time in all factories for each workpiece;

[0056] 4. Arrange all processes in ascending order of delivery time, and then allocate workpieces to different factories according to factory code FA, so that workpieces with earlier delivery time are processed first, thereby finally reducing total tardiness.

[0057] Thus, the embodiment of the application provides a solid theoretical basis for a high-quality initial population by determining the above-mentioned cooperative initialization strategy.

[0058] Optionally, in an embodiment of the application, the first parent population and the second parent population corresponding to the cross-organizational machine shop scheduling model are generated according to a preset cooperative initialization strategy, and the method comprises the following steps: calculating total tardiness and maximum completion time corresponding to a machine shop production scenario according to a factory vector, a process vector and a machine vector; initializing a plurality of target sub-populations randomly, and adjusting the factory vector, the process vector and the machine vector corresponding to each target sub-population in the plurality of target sub-populations by using any initialization strategy in a plurality of preset initialization strategies; and merging the plurality of target sub-populations according to the adjusted factory vector, the process vector and the machine vector to obtain the first parent population and the second parent population.

[0059] After obtaining the encoding vector, the embodiment of the application can distribute workpieces to different teams according to the factory vector, and take out processes in order according to the process vector; secondly, machine information of the workpieces is obtained according to the machine vector, the workpieces are arranged on corresponding machines for processing, and the starting time and the completion time are calculated; finally, after all the workpieces are processed, the tardiness of each workpiece is calculated, all tardinesses are added to obtain total tardiness TT, and maximum completion time C max .

[0060] Then, the embodiment of the application can initialize a plurality of target sub-populations randomly, for example, 5 sub-populations are initialized randomly, and the sizes of the sub-populations are respectively ps / 5; secondly, for each sub-population, different vector encodings (i.e. factory vector, process vector and machine vector) are adjusted by using an initialization strategy; finally, the plurality of target sub-populations are merged into one population with a size of ps, so that the first parent population and the second parent population are obtained by using the initialization method.

[0061] Thus, the embodiment of the application effectively enhances the convergence of the initial population by initializing the sub-population by using the above-mentioned cooperative initialization strategy, thereby integrating a high-quality initial population.

[0062] Optionally, in an embodiment of the present application, the genetic evolution search operation is performed on the first parent population to obtain a first offspring population of the first parent population, including: based on a two-person tournament algorithm, obtaining a mating pool corresponding to the first parent population, and randomly selecting a parent from the mating pool by each individual in the first parent population; according to a precedence operation crossover (POX) strategy and parent exchange, exchanging the processing sequence of the cross-organizational machine shop scheduling model, and exchanging the machine selection information and the factory information of the cross-organizational machine shop scheduling model by using a universal crossover (UX) to generate the first offspring population of the first parent population.

[0063] In the process of implementation, according to the genetic evolution search strategy, the two-person tournament algorithm is used to select the mating pool from the first parent population; then the individual in the first parent population is used to randomly select the parent from the mating pool, and the processing sequence is exchanged by using the precedence operation crossover (POX), and the machine selection information and the factory information are exchanged by using the universal crossover (UX), so as to generate the first offspring population of the first parent population.

[0064] Optionally, in an embodiment of the present application, a preset data-driven neighborhood search strategy includes: finding a key job with the largest delay in the cross-organizational machine shop scheduling model, and searching a first target job meeting a preset delivery time requirement, exchanging the positions of the key job and the first target job, or inserting the key job into a position in front of the first target job; or randomly selecting any job in the cross-organizational machine shop scheduling model, and searching a second target job meeting a preset delivery time requirement, and exchanging the positions of the second target job and any job; or finding a key factory and a key path corresponding to the key factory according to the maximum completion time of the cross-organizational machine shop scheduling model, and randomly selecting a key process, and scanning the completion times of all machines to obtain a target machine meeting a preset completion time requirement, so as to insert the key process into the target machine.

[0065] It should be noted that the data-driven neighborhood search strategy of the embodiment of the present application provides four neighborhood structures N1-N4, and specifically, the four neighborhood structures are as follows:

[0066] N1: finding a key job with the largest delay, searching a job with a later delivery time (i.e. a first target job), and exchanging the positions of the key job and the selected job (i.e. the first target job);

[0067] N2: finding a key job with the largest delay, finding a job with a later delivery time in front, and inserting the key job in front of the selected job (i.e. the first target job);

[0068] N3: As shown in Figure 3 The embodiment of the application randomly selects a workpiece, searches for a workpiece with a later delivery date (i.e., a second target workpiece), and exchanges the positions of the two workpieces;

[0069] N4: As shown in Figure 4 According to the maximum completion time, a critical factory is found, a critical process is randomly selected, the completion times of all machines are scanned, and the machine that does not increase the maximum completion time is selected for insertion, thereby finally reducing the maximum completion time.

[0070] Therefore, the embodiment of the application designs four knowledge-driven neighborhood structures through a knowledge-driven local search strategy, thereby effectively improving the resource optimization efficiency of the Gantt chart and accelerating the convergence of the evolutionary computing population.

[0071] In step S103, a new first parent population is generated based on the first parent population and the second child population, and a new second parent population is obtained based on the second parent population and the first child population. A multi-operator collaborative neighborhood search operation is performed on the new first parent population to obtain second search knowledge corresponding to the new first parent population.

[0072] In step S104, a one-dimensional tensor is obtained by horizontally splicing the factory vector, the process vector, and the machine vector according to the first search knowledge and the second search knowledge, and the one-dimensional tensor is input into a pre-trained neighborhood action recommendation neural network model to obtain a scheduling scheme corresponding to the machining workshop production scene that satisfies the preset requirements.

[0073] Further, the embodiment of the application can horizontally splice the three vectors (i.e., the factory vector, the process vector, and the machine vector) into a one-dimensional tensor based on a Markov decision strategy, and input the one-dimensional tensor into Figure 5 the feedforward neural network shown in the figure to extract scheduling scheme features and output the selection probability of different neighborhood search operators according to the scheduling scheme features.

[0074] Specifically, the Markov decision is designed as follows:

[0075] 1. State: In the embodiment of the application, each solution represents a state, which can be recorded by the parameters of the feedforward neural network.

[0076] 2. Action: The knowledge-driven neighborhood structure is used as the action selected by the agent.

[0077] 3. Reward: When the action A t optimizes the current state S t finds a better solution S t+1 , the reward Rt=10, otherwise Rt=0.

[0078] 4. Transfer: the state of the current solution is S t , the scheduling information such as processing sequence, machine selection, factory selection changes, and the state of the solution also changes to S t t+1

[0079] Therefore, the embodiment of the application solves the problem that the traditional method cannot remember the historical search experience by using the data-driven multi-operation collaborative search, recording the search information by using the neural network, and training the network parameters by using the reinforcement learning, thereby providing the decision for the evolution calculation population.

[0080] Optionally, in an embodiment of the application, a one-dimensional tensor is input into a pre-trained neighborhood action recommendation neural network model to obtain a scheduling scheme corresponding to the machining workshop production scene and satisfying the preset requirement, including: extracting the current state of the one-dimensional tensor, and inputting the current state into the evaluation network of the neighborhood action recommendation neural network model to obtain the Q value of the current state, and simultaneously determining the Q value of the current strategy according to the selected neighborhood action; extracting the next state of the one-dimensional tensor, and inputting the next state into the target network of the neighborhood action recommendation neural network model to obtain the Q value of all actions of the next state; based on the Q value of all actions of the next state, obtaining the maximum Q value in all actions of the next state, and calculating a target value according to the maximum Q value, a preset discount factor and reward information of the one-dimensional tensor; calculating the root mean square error of the neighborhood action recommendation neural network model according to the Q value of the current action and the target value, and optimizing the evaluation network parameters by using a preset back propagation algorithm, and simultaneously updating the target network parameters by using the optimized evaluation network parameters in the case that the target network parameters meet the preset update requirement; adjusting the root mean square error by using the updated target network parameters, and obtaining the scheduling scheme corresponding to the machining workshop production scene and satisfying the preset requirement in the case that the root mean square error meets the preset error condition.

[0081] As an implementable manner, the neighborhood action recommendation neural network in the embodiment of the application can adopt a deep Q network algorithm, which includes an evaluation network Q_val and a target network Q_tar, wherein the evaluation network is used to evaluate the Q value of the current state, and the target network is used to fit Q_val and update the parameters by back propagation. The specific training process is as follows:

[0082] 1. Take data with a size of bs from the experience pool, and the data format is (S t , A t , R t , S t+1 );

[0083] 2. Extract the state S t , input into the evaluation network Q_val to obtain the Q value of the current state, such as​Figure 6 As shown, select the Q value Q_val(S) for the current action. t A t );

[0084] 3. Extract the next state S t+1 Input the target network Q_tar to obtain the Q-values ​​of all actions in the next state, Q_tar(S t+1 ,:);

[0085] 4. Select the maximum Q value of the next state, multiply it by the discount factor γ, and add the reward R. t The target value q is obtained. t .

[0086] 5. Calculate the root mean square error MSE(Q_val(S)). t A t ), q t The backpropagation algorithm is used to update the parameters of Q_val to minimize the MSE.

[0087] 6. If iter_max has not been updated in Q_tar, then copy the value of Q_val to Q_tar.

[0088] Therefore, the embodiments of this application establish a cross-organizational machining workshop scheduling model based on the actual cross-organizational machining workshop production characteristics of complex aerospace products, and form a framework for collaborative solution of genetic evolutionary computation and reinforcement learning through a data and knowledge-driven scheduling algorithm, thereby achieving rapid convergence and efficient solution of the resource grouping complex machining workshop scheduling problem.

[0089] The execution logic of the cross-organizational machining workshop dual-drive intelligent scheduling method of this application is described below through a specific embodiment and in conjunction with the accompanying drawings.

[0090] Figure 7 This is a schematic diagram illustrating the execution logic of the cross-organizational analog-digital dual-drive intelligent scheduling method for machining workshops according to this application. Figure 7 As shown, the execution process of the cross-organizational machining workshop dual-drive intelligent scheduling method of this application is as follows:

[0091] S701: Initialize evolutionary computation and reinforcement learning populations P1 and P2, and neural network parameters;

[0092] S702: Decode populations P1 and P2, and calculate the objective function value;

[0093] S703: Genetic evolutionary search population P1 yields offspring C1;

[0094] S704: Use a neural network to perform a neighborhood search on P2 to obtain the offspring C2, and store the search knowledge (P2, A tR t ,C2);

[0095] S705: combine P1 and C2 as shown in Figure 8 , and select a new population P1' using fast non-dominated sorting;

[0096] S706: combine P2 and C1, and select a new population P2' using fast non-dominated sorting;

[0097] S707: perform multi-operator cooperative neighborhood search on population P1', store search knowledge (P1', A t ,R t ,C1') ;

[0098] S708: determine whether the end condition is met, if yes, end, otherwise go to S703.

[0099] In addition, the present application can illustrate and analyze the performance of the digital-analog dual-drive intelligent scheduling method of the cross-organizational machine shop of the present application through a specific embodiment.

[0100] Specifically, in order to test the performance of the present application in the worker-constrained complex welding shop scheduling, the present application can randomly generate 20 test problems according to the actual production data for testing. The number of workpieces N = {20, 40, 60, 80, 100}, the number of parallel machines M f = {4, 6}, the number of workpieces F = {2, 3}, and the processing time is randomly selected in U ~ [5, 95].

[0101] As an implementable way, the specific embodiment of the present application can use python for related programming operations, the simulation environment is Intel(R) core(TM) i9-13900k CPU @ 3.4GHz / 16GB RAM, GPU NVIDIA 4070 12GB, and the operating system is windows 10; the termination criterion of the algorithm DRLMCCA corresponding to the digital-analog dual-drive intelligent scheduling method of the cross-organizational machine shop of the present application is that the algorithm stops when the maximum evaluation number MaxNFEs reaches 400*N; in the experiment, the hypervolume (Hypervolume, HV) and the generation distance (Generation Distance, GD) can be used to represent the comprehensive performance and convergence of the algorithm to verify the present application.

[0102] In order to verify the effectiveness of the present application, the following four variant algorithms are designed in the experiment:

[0103] 1. MA-L represents that the evolutionary calculation deletes local search;

[0104] 2. MA-I represents that the evolutionary calculation deletes multi-strategy cooperative initialization;

[0105] 3. MA means using only evolutionary computing population to solve;

[0106] 4. DRL means using only reinforcement learning population to solve.

[0107] The specific embodiments of the present application can compare the average HV value of DRLMCCA with the variant algorithm, as shown in Table 1, and compare the average GD value of DRLMCCA with the variant algorithm, as shown in Table 2.

[0108] Table 1

[0109]

[0110]

[0111] Table 2

[0112]

[0113]

[0114] From Table 1 and Table 2, it can be seen that DRLMCCA achieves the best HV and GD indicators, indicating that after deleting the improvement strategy and each functional module, the algorithm performance has significantly decreased, effectively verifying the effectiveness of each algorithm unit of the present application.

[0115] After that, the present application can sort the Frideman rank sum test results of DRLMCCA and the variant algorithm, as shown in Table 3.

[0116] Table 3

[0117]

[0118] From Table 3, it can be seen that DRLMCCA has the highest HV and GD indicator ranking, and compared with other algorithms, the p value is less than 0.05, and the algorithm performance significantly decreases after deleting the improvement strategy, verifying the significance of the different improvement strategies designed in the present application.

[0119] In order to further verify the solving performance of the algorithm, the present application selects six kinds of the latest and best algorithms on the cross-organizational machine shop scheduling problem, which are NSGA-II, MOEA / D, DABC, KTPO, VICA and KPMA. Each algorithm is independently run 20 times, and the average HV and GD indicators of the algorithm are calculated, wherein the statistical results of the average HV value of DRLMCCA and the comparison algorithm are shown in Table 4, and the statistical results of the average HV value of DRLMCCA and the comparison algorithm are shown in Table 5.

[0120] Table 4

[0121]

[0122]

[0123] Table 5

[0124]

[0125]

[0126] From Table 4 and Table 5, it can be seen that the DRLMCCA algorithm of the present application has achieved significant advantages in solving the complex problem, because the data-driven strategy of the present application can record historical search experience, and also provide effective decision-making for the evolutionary computing process.

[0127] Further, the Frideman rank sum test results of the DRLMCCA and the comparative algorithm can be ranked by the present application, as shown in Table 6.

[0128] Table 6

[0129]

[0130] From Table 6, it can be seen that the DRLMCCA has the highest HV and GD ranking, and the p value is less than 0.05, thus powerfully indicating that the DRLMCCA has significant advantages in comprehensive solving performance and convergence.

[0131] According to the cross-organizational machine workshop numerical model double-drive intelligent scheduling method provided in the embodiment of the application, the actual machine workshop production scene is modeled, a cross-organizational machine workshop scheduling model corresponding to the machine workshop production scene is established, and vector coding is performed on the cross-organizational machine workshop scheduling model to obtain a factory vector, a process vector and a machine vector corresponding to the cross-organizational machine workshop scheduling model; a first parent population and a second parent population corresponding to the cross-organizational machine workshop scheduling model are generated according to a preset collaborative initialization strategy, a genetic evolution search operation is performed on the first parent population to obtain a first child population of the first parent population, and based on the second parent population and a preset data-driven neighborhood search strategy, a second child population is obtained and a first search knowledge corresponding to the second child population is recorded; based on the first parent population and the second child population, a new first parent population is generated, and based on the second parent population and the first child population, a new second parent population is obtained, and a multi-operator collaborative neighborhood search operation is performed on the new first parent population to obtain a second search knowledge corresponding to the new first parent population; a one-dimensional tensor is obtained by transversely splicing the factory vector, the process vector and the machine vector according to the first search knowledge and the second search knowledge, and the one-dimensional tensor is input into a pre-trained neighborhood action recommendation neural network model to obtain a scheduling scheme corresponding to the machine workshop production scene and satisfying a preset requirement. The application considers various resource configurations and constraint conditions, models the production scene of the actual machine workshop as a cross-organizational machine workshop scheduling problem, thereby improving the manufacturing capacity of the aircraft product in the machine workshop from the perspective of resource optimization, improving the solving efficiency of the resource grouping complex machine workshop scheduling problem, and reducing the manufacturing cost of the aircraft product.

[0132] Secondly, the cross-organizational machine workshop numerical model double-drive intelligent scheduling device provided in the embodiment of the application is described with reference to the accompanying drawings.

[0133] Figure 9 FIG. 1 is a block schematic diagram of the cross-organizational machine workshop numerical model double-drive intelligent scheduling device according to the embodiment of the application.

[0134] As shown in FIG. 1, the cross-organizational machine workshop numerical model double-drive intelligent scheduling device 10 includes a modeling module 100, a data-driven module 200, a knowledge-driven module 300 and a splicing module 400. Figure 9 The modeling module 100 is configured to model the actual machine workshop production scene, establish a cross-organizational machine workshop scheduling model corresponding to the machine workshop production scene, and perform vector coding on the cross-organizational machine workshop scheduling model to obtain a factory vector, a process vector and a machine vector corresponding to the cross-organizational machine workshop scheduling model.

[0135]

[0136] ​The data driving module 200 is configured to generate a first parent population and a second parent population corresponding to a cross-organization machine shop scheduling model according to a preset collaborative initialization strategy, perform a genetic evolution search operation on the first parent population to obtain a first offspring population of the first parent population, and obtain a second offspring population based on the second parent population and a preset data-driven neighborhood search strategy, and record first search knowledge corresponding to the second offspring population.

[0137] The knowledge driving module 300 is configured to generate a new first parent population based on the first parent population and the second offspring population, obtain a new second parent population according to the second parent population and the first offspring population, and perform a multi-operator collaborative neighborhood search operation on the new first parent population to obtain second search knowledge corresponding to the new first parent population.

[0138] The splicing module 400 is configured to horizontally splice a factory vector, a process vector and a machine vector to obtain a one-dimensional tensor according to the first search knowledge and the second search knowledge, and input the one-dimensional tensor into a pre-trained neighborhood action recommendation neural network model to obtain a scheduling scheme corresponding to a machine shop production scene and satisfying a preset requirement.

[0139] Optionally, in an embodiment of the present application, the data driving module 200 comprises a first initialization strategy unit or a second initialization strategy unit or a third initialization strategy unit or a fourth initialization strategy unit.

[0140] The first initialization strategy unit is configured to select a parallel machine with the minimum processing time for each workpiece in the cross-organization machine shop scheduling model.

[0141] The second initialization strategy unit is configured to accumulate the processing time of each workpiece in different factories to obtain a cumulative value corresponding to each workpiece, arrange the cumulative values of all workpieces in descending order to generate a cumulative value sequence, and distribute each workpiece to a corresponding factory according to the cumulative value sequence.

[0142] The third initialization strategy unit is configured to select a parallel machine with the minimum completion time in all factories for each workpiece.

[0143] The fourth initialization strategy unit is configured to arrange all processes in ascending order according to the delivery period of each process in the cross-organization machine shop scheduling model to obtain a process sequence, and distribute each workpiece to a corresponding factory based on the process sequence and a preset factory code.

[0144] Optionally, in an embodiment of the present application, the data driving module 200 further comprises a calculation unit, an adjustment unit and a merging unit.

[0145] The calculation unit is configured to calculate total tardiness and maximum completion time corresponding to the machine shop production scene according to the factory vector, the process vector and the machine vector.

[0146] The adjustment unit is configured to randomly initialize a plurality of target sub-populations, and adjust the factory vector, the process vector and the machine vector corresponding to each target sub-population in the plurality of target sub-populations by any initialization strategy in a plurality of preset initialization strategies.

[0147] The merging unit is configured to merge the plurality of target sub-populations according to the adjusted factory vector, the process vector and the machine vector, to obtain a first parent population and a second parent population.

[0148] Optionally, in an embodiment of the present application, the data driving module 200 further comprises a selection unit and a exchange unit.

[0149] The selection unit is configured to obtain a mating pool corresponding to the first parent population based on a two-person tournament algorithm, and randomly select a parent from the mating pool by each individual in the first parent population.

[0150] The exchange unit is configured to exchange the process processing sequence of the cross-organizational machine shop scheduling model according to a process-based crossover strategy, and exchange the machine selection information and the factory information of the cross-organizational machine shop scheduling model by using a general crossover strategy, to generate a first offspring population of the first parent population.

[0151] Optionally, in an embodiment of the present application, the data driving module 200 further comprises a first neighborhood searching unit or a second neighborhood searching unit or a third neighborhood searching unit.

[0152] The first neighborhood searching unit is configured to find a key job with the largest tardiness in the cross-organizational machine shop scheduling model, search a first target job meeting a preset delivery time requirement, exchange the positions of the key job and the first target job, or insert the key job into a position in front of the first target job.

[0153] The second neighborhood searching unit is configured to randomly select any job in the cross-organizational machine shop scheduling model, search a second target job meeting a preset delivery time requirement, and exchange the positions of the second target job and any job.

[0154] The third neighborhood searching unit is configured to find a key factory and a key path corresponding to the key factory of the cross-organizational machine shop scheduling model according to the maximum completion time, randomly select a key process, and scan the completion times of all machines to obtain a target machine meeting a preset completion time requirement, so as to insert the key process into the target machine.

[0155] Optionally, in an embodiment of the present application, the splicing module 400 comprises a first extraction unit, a second extraction unit, an acquisition unit, an optimization unit and an adjustment unit.

[0156] The first extraction unit is configured to extract a current state of the one-dimensional tensor and input the current state into an evaluation network of the neighborhood action recommendation neural network model to obtain a Q value of the current state, and determine a Q value of a current policy according to a selected neighborhood action.

[0157] The second extraction unit is configured to extract a next state of the one-dimensional tensor and input the next state into a target network of the neighborhood action recommendation neural network model to obtain Q values of all actions of the next state.

[0158] The acquisition unit is configured to acquire a maximum Q value of all actions of the next state based on the Q values of all actions of the next state, and calculate a target value according to the maximum Q value, a preset discount factor and reward information of the one-dimensional tensor.

[0159] The optimization unit is configured to calculate a root mean square error of the neighborhood action recommendation neural network model according to the Q value of the current action and the target value, and optimize parameters of the evaluation network through a preset back propagation algorithm, and update parameters of the target network by using the optimized parameters of the evaluation network when the parameters of the target network meet a preset update requirement.

[0160] The adjustment unit is configured to adjust the root mean square error by using the updated parameters of the target network, and obtain a scheduling scheme that meets a preset requirement for a production scene of the machine shop when the root mean square error meets a preset error condition.

[0161] It should be noted that the above description of the embodiment of the method for intelligent scheduling of the cross-organization machine shop driven by digital and analog modes also applies to the embodiment of the device for intelligent scheduling of the cross-organization machine shop driven by digital and analog modes, which will not be described here.

[0162] The digital-analog dual-drive intelligent scheduling device across the cross-organizational machine workshop provided by the embodiment of the application comprises a modeling module, which is used for modeling an actual machine workshop production scene, establishing a cross-organizational machine workshop scheduling model corresponding to the machine workshop production scene, and performing vector coding on the cross-organizational machine workshop scheduling model to obtain a factory vector, a process vector and a machine vector corresponding to the cross-organizational machine workshop scheduling model; a data driving module, which is used for generating a first parent population and a second parent population corresponding to the cross-organizational machine workshop scheduling model according to a preset collaborative initialization strategy, performing a genetic evolution search operation on the first parent population to obtain a first child population of the first parent population, and obtaining a second child population based on the second parent population and a preset data-driven neighborhood search strategy and recording first search knowledge corresponding to the second child population; a knowledge driving module, which is used for generating a new first parent population based on the first parent population and the second child population, obtaining a new second parent population according to the second parent population and the first child population, and performing a multi-operator collaborative neighborhood search operation on the new first parent population to obtain second search knowledge corresponding to the new first parent population; and a splicing module, which is used for horizontally splicing the factory vector, the process vector and the machine vector according to the first search knowledge and the second search knowledge to obtain a one-dimensional tensor, and inputting the one-dimensional tensor into a pre-trained neighborhood action recommendation neural network model to obtain a scheduling scheme corresponding to the machine workshop production scene and satisfying a preset requirement. The application models the production scene of the actual machine workshop as a cross-organizational machine workshop scheduling problem by considering various resource configurations and constraint conditions, thereby improving the manufacturing capacity of the aircraft product in the machine workshop from the perspective of resource optimization, improving the solving efficiency of the resource grouping complex machine workshop scheduling problem, and reducing the manufacturing cost of the aircraft product.

[0163] Figure 10 The structure schematic diagram of the electronic device provided by the embodiment of the application is provided. The electronic device can comprise:

[0164] The memory 1001, the processor 1002 and the computer program stored in the memory 1001 and executable on the processor 1002.

[0165] The processor 1002 implements the cross-organizational machine workshop digital-analog dual-drive intelligent scheduling method provided in the above embodiment when executing the program.

[0166] Further, the electronic device further comprises:

[0167] The communication interface 1003 is used for communication between the memory 1001 and the processor 1002.

[0168] The memory 1001 is used for storing the computer program executable on the processor 1002.

[0169] The memory 1001 can include a high-speed RAM memory, and can also include a non-volatile memory, for example, at least one disk memory.

[0170] If the memory 1001, the processor 1002 and the communication interface 1003 are implemented independently, the communication interface 1003, the memory 1001 and the processor 1002 can be connected to each other through a bus and complete communication between each other. The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. The bus can be divided into an address bus, a data bus, a control bus, etc. For convenience of representation, Figure 10 Only one thick line is used in the figure to represent the bus, but it does not mean that there is only one bus or only one type of bus.

[0171] Optionally, in a specific implementation, if the memory 1001, the processor 1002 and the communication interface 1003 are integrated on a chip, the memory 1001, the processor 1002 and the communication interface 1003 can complete communication between each other through an internal interface.

[0172] The processor 1002 can be a Central Processing Unit (CPU), or an Application Specific Integrated Circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of the present application.

[0173] The embodiments of the present application also provide a computer readable storage medium, which stores a computer program, and the program is executed by a processor to implement the above cross-organizational machine shop digital-analog dual-drive intelligent scheduling method.

[0174] In the description of the application, reference to "one embodiment", "some embodiments", "an example", "a specific example", or "some examples" means that a particular feature, structure, material, or characteristic being described is included in at least one embodiment or example of the application. The appearances of the phrase in various places in the specification are not necessarily all referring to the same embodiment or example. Furthermore, the described specific features, structures, materials, or characteristics can be combined in any suitable manner in one or more embodiments or examples. In addition, the usage of "N" means at least two, for example, two, three or the like, unless explicitly stated otherwise.

[0175] Furthermore, the terms "first", "second", or the like, are used merely as a designation of certain elements or features, and do not imply or connote relative importance or a specific order of categorization thereof. Accordingly, features described as "first" or "second" can be explicitly or implicitly included in at least one of the features. In the description of the application, the meaning of "N" is at least two, for example, two, three, etc., unless explicitly specified otherwise.

[0176] Any process or method descriptions or blocks in flow charts or otherwise described herein represent embodiments which can be managed as one or more modules, segments, or portions of code which include one or more executable instructions for implementing specific logic functions or steps, and alternate implementations are possible. In some embodiments, the processes and methods described can be executed by one or more apparatuses or devices, either directly or after conversion to another language. Alternate implementations are possible.

[0177] The logic and / or steps represented in the flowcharts and / or described herein, for example, can be considered as a sequence of instructions to implement logical functions, and can be embodied in any computer-readable medium for use by an instruction execution system, apparatus, or device, such as a computer-based system, processor- based system, or other system that can fetch the instructions from the instruction execution system, apparatus, or device and execute the instructions. For purposes of this specification, a "computer-readable medium" can be any apparatus that can contain, store, communicate, propagate, or transport the program for use by or in connection with the instruction execution system, apparatus, or device. The computer-readable medium can be a computer- readable storage medium or a computer-readable signal medium. The computer- readable storage medium can be, for example, but not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any suitable combination of the foregoing. More specific examples (a non-exhaustive list) of the computer-readable storage medium include the following: an electrical connection having one or more wires (electrical connections), a portable computer diskette (a magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), an optical fiber, and a portable compact disc read-only memory (CDROM). In addition, the computer-readable medium can even be paper or another suitable medium upon which the program is printed, as the program can be electronically captured, via, for example, optical scanning of the paper or other medium, then compiled, interpreted, or otherwise processed in a suitable manner, if necessary, and then stored in a computer memory.

[0178] It should be understood that aspects of the application can be implemented in hardware, software, firmware or combinations thereof. In the above embodiments, the N steps or methods can be implemented in software or firmware stored in a memory and executed by a suitable instruction execution system. If implemented in hardware and in another embodiment, any of the following technologies, known in the art, or their combinations can be used: discrete logic circuitry having logic gates for implementing logic functions on data signals, application specific integrated circuits having appropriate combinational logic gates, programmable gate arrays (PGA), field programmable gate arrays (FPGA), etc.

[0179] Those skilled in the art can understand that all or part of the steps carried out by the above-mentioned embodiments can be completed by programs instructing related hardware, and the programs can be stored in a computer-readable storage medium. When the programs are executed, one or a combination of the steps of the method embodiments is included.

[0180] In addition, each of the functional units in the various embodiments of the present application can be integrated in one processing module, or each of the units can be physically present separately, or two or more units can be integrated in one module. The integrated module can be realized in the form of hardware or in the form of a software functional module. When the integrated module is realized in the form of a software functional module and sold or used as an independent product, it can also be stored in a computer readable storage medium.

[0181] The storage medium mentioned above can be a read-only memory, a magnetic disk or an optical disk, etc. Although the embodiments of the present application have been shown and described above, it should be understood that the above embodiments are exemplary and should not be construed as limiting the present application, and those skilled in the art can make changes, modifications, replacements and variations to the above embodiments within the scope of the present application.

Claims

1. A cross-organizational intelligent scheduling method for dual-drive digital-analog machining workshops, characterized in that, Includes the following steps: A model is created for the actual machining workshop production scenario, and a cross-organizational machining workshop scheduling model is established corresponding to the machining workshop production scenario. The cross-organizational machining workshop scheduling model is then vector-encoded to obtain the factory vector, process vector, and machine vector corresponding to the cross-organizational machining workshop scheduling model. The first parent population and the second parent population corresponding to the cross-organizational machining workshop scheduling model are generated according to the preset collaborative initialization strategy. Genetic evolution search operation is performed on the first parent population to obtain the first offspring population of the first parent population. At the same time, based on the second parent population and the preset data-driven neighborhood search strategy, the second offspring population is obtained, and the first search knowledge corresponding to the second offspring population is recorded. Based on the first parent population and the second offspring population, a new first parent population is generated. At the same time, based on the second parent population and the first offspring population, a new second parent population is obtained. A multi-operator cooperative neighborhood search operation is performed on the new first parent population to obtain the second search knowledge corresponding to the new first parent population. A one-dimensional tensor is obtained by horizontally concatenating the factory vector, the process vector, and the machine vector based on the first search knowledge and the second search knowledge. The one-dimensional tensor is then input into a pre-trained neighborhood action recommendation neural network model to obtain a scheduling scheme that meets the preset requirements for the machining workshop production scenario.

2. The method according to claim 1, characterized in that, The preset collaborative initialization strategy includes: In the cross-organizational machining workshop scheduling model, the parallel machine with the shortest processing time is selected for each workpiece; Alternatively, the processing time of each workpiece in different factories can be accumulated to obtain the cumulative value corresponding to each workpiece, and the cumulative values ​​of all workpieces can be sorted in descending order to generate a cumulative value sequence, so as to allocate each workpiece to the corresponding factory according to the cumulative value sequence; Alternatively, each workpiece may be selected from all factories using a parallel machine with the shortest completion time; Alternatively, based on the delivery date of each process in the cross-organizational machining workshop scheduling model, each process is sorted in ascending order to obtain a process sequence, and each workpiece is assigned to the corresponding factory based on the process sequence and a preset factory code.

3. The method according to claim 1, characterized in that, The step of generating the first and second parent populations corresponding to the cross-organizational machining workshop scheduling model according to the preset collaborative initialization strategy includes: Calculate the total delay and maximum completion time corresponding to the machining workshop production scenario based on the factory vector, the process vector, and the machine vector; Multiple target subpopulations are randomly initialized, and the factory vector, process vector, and machine vector corresponding to each target subpopulation are adjusted by any of the preset initialization strategies. The multiple target subpopulations are merged based on the adjusted factory vector, process vector, and machine vector to obtain the first parent population and the second parent population.

4. The method according to claim 3, characterized in that, The step of performing a genetic evolutionary search operation on the first parent population to obtain the first offspring population of the first parent population includes: Based on the two-person tournament algorithm, the mating pool corresponding to the first parent population is obtained, and each individual in the first parent population randomly selects a parent from the mating pool; The first offspring population of the first parent population is generated by exchanging the process sequence of the cross-organizational machining shop scheduling model according to the process-based crossover strategy and the parent population, and by exchanging the machine selection information and factory information of the cross-organizational machining shop scheduling model using a general crossover strategy.

5. The method according to claim 4, characterized in that, The preset data-driven neighborhood search strategy includes: Find the critical workpiece with the longest delay in the cross-organizational machining workshop scheduling model, and search forward to find the first target workpiece that meets the preset delivery date requirement. Swap the positions of the critical workpiece and the first target workpiece, or insert the critical workpiece forward to the position in front of the first target workpiece. Alternatively, randomly select any workpiece in the cross-organizational machining workshop scheduling model, search for a second target workpiece that meets the preset delivery date requirement, and swap the positions of the second target workpiece and the any workpiece. Alternatively, based on the maximum completion time, the critical factory and the critical path corresponding to the critical factory in the cross-organizational machining workshop scheduling model are found. At the same time, a critical process is randomly selected, and the completion time of all machines is scanned to obtain the target machine that meets the preset completion time requirement, so as to insert the critical process into the target machine.

6. The method according to claim 5, characterized in that, The step of inputting the one-dimensional tensor into a pre-trained neighborhood action recommendation neural network model to obtain a scheduling scheme that meets preset requirements for the machining workshop production scenario includes: Extract the current state of the one-dimensional tensor and input the current state into the evaluation network of the neighborhood action recommendation neural network model to obtain the Q value of the current state. At the same time, determine the Q value of the current policy based on the selected neighborhood action. Extract the next state of the one-dimensional tensor and input the next state into the target network of the neighborhood action recommendation neural network model to obtain the Q value of all actions in the next state; Based on the Q-values ​​of all actions in the next state, obtain the maximum Q-value among all actions in the next state, and calculate the target value based on the maximum Q-value, the preset discount factor, and the reward information of the one-dimensional tensor. The root mean square error of the neighborhood action recommendation neural network model is calculated based on the Q value of the current action and the target value. The evaluation network parameters are optimized by a preset backpropagation algorithm. Meanwhile, if the target network parameters meet the preset update requirements, the optimized evaluation network parameters are used to update the target network parameters. The root mean square error is adjusted by updating the target network parameters, and a scheduling scheme that meets the preset requirements corresponding to the machining workshop production scenario is obtained when the root mean square error meets the preset error conditions.

7. A dual-drive intelligent scheduling device for digital and analog circuits across machining workshops, characterized in that, include: The modeling module is used to model the actual machining workshop production scenario, establish a cross-organization machining workshop scheduling model corresponding to the machining workshop production scenario, and perform vector encoding on the cross-organization machining workshop scheduling model to obtain the factory vector, process vector and machine vector corresponding to the cross-organization machining workshop scheduling model; The data-driven module is used to generate a first parent population and a second parent population corresponding to the cross-organizational machining workshop scheduling model according to a preset collaborative initialization strategy, and to perform a genetic evolution search operation on the first parent population to obtain the first offspring population of the first parent population. At the same time, based on the second parent population and the preset data-driven neighborhood search strategy, the second offspring population is obtained, and the first search knowledge corresponding to the second offspring population is recorded. The knowledge-driven module is used to generate a new first parent population based on the first parent population and the second child population, and to obtain a new second parent population based on the second parent population and the first child population. It also performs a multi-operator cooperative neighborhood search operation on the new first parent population to obtain the second search knowledge corresponding to the new first parent population. The splicing module is used to horizontally splice the factory vector, the process vector, and the machine vector according to the first search knowledge and the second search knowledge to obtain a one-dimensional tensor, and input the one-dimensional tensor into a pre-trained neighborhood action recommendation neural network model to obtain a scheduling scheme that meets the preset requirements corresponding to the machining workshop production scenario.

8. The apparatus according to claim 7, characterized in that, The data-driven module includes: The first initialization strategy unit is used to select the parallel machine with the shortest processing time for each workpiece in the cross-organizational machining shop scheduling model; or... The second initialization strategy unit is used to accumulate the processing time of each workpiece in different factories to obtain a cumulative value for each workpiece, and to sort the cumulative values ​​of all workpieces in descending order to generate a cumulative value sequence, so as to allocate each workpiece to the corresponding factory according to the cumulative value sequence; or, The third initialization strategy unit is used to select a parallel machine with the minimum completion time from all factories for each workpiece; or, The fourth initialization strategy unit is used to sort each process in ascending order according to the delivery date of each process in the cross-organization machining workshop scheduling model to obtain a process sequence, and to allocate each workpiece to the corresponding factory based on the process sequence and a preset factory code.

9. An electronic device, characterized in that, include: The system includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the cross-organizational machining workshop dual-drive intelligent scheduling method as described in any one of claims 1-6.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, The program is executed by the processor to implement the cross-organizational analog-digital dual-drive intelligent scheduling method for machining workshops as described in any one of claims 1-6.

Images