An artificial intelligence-based discrete manufacturing production line scheduling optimization method

Abstract

The application discloses a discrete manufacturing production line scheduling optimization method based on artificial intelligence and belongs to the technical field of intelligent manufacturing, and comprises the following steps: acquiring first scheduling data, second scheduling data and a workshop execution state corresponding to the first scheduling data of a discrete manufacturing production line; generating a transferable commitment unit; generating a candidate transfer unit; introducing a reachable coupling evaluation function to generate an insertion task invasion evaluation sequence; inputting an improved Pointer Network model to determine a target transfer unit, generate a transfer task sequence and a backfill task sequence; performing commitment transfer reconstruction; and generating a scheduling optimization result. The application reduces the disturbance range of the insertion task on the original scheduling plan and improves the adaptability of the discrete manufacturing production line scheduling optimization result to the process undertaking relationship, the resource time window relationship and the delivery date commitment relationship.

Description

Technical Field

[0001] This invention relates to the field of intelligent manufacturing technology, and in particular to a method for optimizing the scheduling of discrete manufacturing production lines based on artificial intelligence. Background Technology

[0002] With the increasing application of digital workshops, manufacturing execution systems, and intelligent scheduling systems in discrete manufacturing enterprises, production line scheduling optimization technologies for multi-variety, small-batch orders, complex process routes, and multiple resource constraints have received widespread attention. Existing discrete manufacturing production line scheduling methods mainly rely on manual experience, fixed priority rules, heuristic scheduling algorithms, or conventional intelligent optimization models to sort production tasks and allocate resources. However, these methods commonly suffer from the following problems in practical applications: After a rush order enters the production line, the original production schedule's established production task sequence, resource occupancy windows, and delivery commitments are easily disrupted. Existing methods typically insert or rearrange tasks directly based on the urgency of the rush order, making it difficult to distinguish between frozen, adjustable, and transferable tasks. This leads to excessively large scheduling adjustments and decreased shop floor execution stability. Rush orders and original production tasks have process succession relationships, resource window relationships, and delivery constraints. Traditional scheduling methods often use equipment idle time or order priority as the basis for insertion, making it difficult to accurately evaluate the delivery buffer compression, increased resource switching, process waiting propagation, and task sequence disturbances caused by rush orders entering candidate scheduling positions. Regarding the compensation of original tasks after rush orders, existing technologies often only generate new scheduling results after rush orders, lacking backfilling arrangements and commitment restoration processing for transferred production tasks. This can easily lead to the completion of rush orders while original orders are delayed, affecting the overall delivery reliability of discrete manufacturing production lines.

[0003] Therefore, how to provide an artificial intelligence-based method for optimizing the scheduling of discrete manufacturing production lines is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0004] One objective of this invention is to propose a production scheduling optimization method for discrete manufacturing production lines based on artificial intelligence. This invention acquires first production scheduling data, second production scheduling data, and workshop execution status. It performs commitment freeze and commitment flexibility assessments on production tasks in the first production scheduling data to generate transferable commitment units. Based on the second production scheduling data and transferable commitment units, it generates candidate transferable units. A reachability coupling evaluation function is introduced to calculate the commitment intrusion strength of the candidate transferable units, forming an order insertion intrusion evaluation sequence. The candidate transferable units and the order insertion intrusion evaluation sequence are input into an improved PointerNetwork model to determine the target transferable units and generate a transfer task sequence and a backfill task sequence. Commitment transfer reconstruction is performed on the first and second production scheduling data. This addresses the problems in existing technologies, such as large disturbance ranges in the original production schedule after the insertion task enters, lack of commitment intrusion evaluation at candidate insertion positions, and lack of backfill arrangements for the transferred production tasks. It achieves low-disturbance production scheduling optimization between the insertion task and the original production task.

[0005] According to an embodiment of the present invention, a production scheduling optimization method for discrete manufacturing production lines based on artificial intelligence includes the following steps: Acquire the first production scheduling data, the second production scheduling data, and the workshop execution status corresponding to the first production scheduling data of the discrete manufacturing production line; Based on the first production scheduling data and the workshop execution status, the production tasks in the first production scheduling data are subject to commitment freeze determination and commitment flexibility determination, and a transferable commitment unit is generated. Based on the second production scheduling data and the transferable commitment unit, process acceptance matching and resource window matching are performed to generate candidate transfer units; An reachability coupling evaluation function is introduced to calculate the committed intrusion strength of candidate transfer units and generate an insertion intrusion evaluation sequence. The candidate transfer unit and the insertion intrusion evaluation sequence are input into the improved Pointer Network model to determine the target transfer unit and generate the transfer task sequence and backfill task sequence. The improved Pointer Network model includes an encoding module, a decoding module and a constraint verification module. A commitment state rolling mechanism is introduced in the encoding module and a transfer debt window repayment mechanism is introduced in the decoding module. Based on the target transfer unit, the transfer task sequence and the backfill task sequence, the commitment transfer reconstruction is performed on the first production scheduling data and the second production scheduling data to generate the target production scheduling plan. Update the task execution sequence and resource usage window of the discrete manufacturing production line based on the target production schedule, and generate production schedule optimization results.

[0006] Optionally, the step of obtaining the first production scheduling data, the second production scheduling data, and the workshop execution status corresponding to the first production scheduling data of the discrete manufacturing production line specifically includes: Retrieve the original production schedule record formed before the insertion task enters, and generate the first production schedule data by associating the production task identifier with the order identifier, process identifier, process sequence, production resource identifier, resource occupation time window, planned start time, planned completion time and delivery commitment; Retrieve the order insertion task record, associate the order insertion task identifier with the order insertion order identifier, product identifier, process route, process identifier, process sequence, resource requirement duration and delivery deadline, and generate the second production scheduling data; Obtain the workshop execution record corresponding to the first production scheduling data, and generate the workshop execution status by associating the task issuance status, task execution status, actual start time, actual completion time and production resource occupancy record with the production task identifier.

[0007] Optionally, based on the first production scheduling data and the workshop execution status, the step of performing commitment freeze determination and commitment flexibility determination on the production tasks in the first production scheduling data to generate transferable commitment units specifically involves: Based on the production task identifier in the first production scheduling data, the first production scheduling data is matched with the workshop execution status to generate a production task execution comparison relationship; Based on the aforementioned production task execution comparison relationship, if the task execution status of a production task is completed or in the processing state, then a commitment freeze flag is assigned to the production task. For production tasks without a commitment freeze flag, calculate the delivery time margin between the delivery commitment and the planned completion time, and calculate the time window adjustment margin between the resource occupancy window and the planned start and completion times; Production tasks whose delivery margin and time window adjustment margin both meet the requirements for inserting the second production scheduling data are identified as transferable production tasks. Generate a transferable commitment unit based on the production task identifier, process identifier, process sequence, production resource identifier, resource occupation window, and delivery commitment corresponding to the transferable production task.

[0008] Optionally, the step of generating candidate transfer units based on the second production scheduling data and the transferable commitment unit to perform process acceptance matching and resource time window matching specifically involves: Based on the order insertion task identifier, process route, and process sequence in the second production scheduling data, a second process chain is generated; based on the order insertion task identifier, production resource identifier, resource requirement duration, and delivery deadline in the second production scheduling data, a second resource requirement window is generated. Based on the process identifier and process sequence in the transferable commitment unit, the second process chain is matched with the transferable commitment unit to generate a process acceptance matching result. Based on the production resource identifier and resource occupation window in the transferable commitment unit, the second resource demand window is matched with the transferable commitment unit to generate a resource window matching result. The transferable commitment unit that meets the requirements for inserting the second production scheduling data in both the process acceptance matching result and the resource time window matching result is identified as the candidate transferable unit.

[0009] Optionally, the introduction of an reachability coupling evaluation function to calculate the committed intrusion strength of candidate transfer units and generate an insertion intrusion evaluation sequence specifically involves: For candidate transfer units, the delivery date commitment, planned completion time, process sequence, production resource identifier and resource occupation window are extracted from the first production scheduling data according to the production task identifier, and the second process chain, resource requirement duration and delivery deadline are extracted from the second production scheduling data according to the order insertion task identifier. Calculate the delivery buffer penetration based on the change in planned completion time and delivery commitment after the second process chain enters the candidate transfer unit; Calculate the resource switching increment based on the production resource identifiers, resource occupation time windows, and resource demand duration before and after the second process chain enters the candidate transfer unit; Based on the change in process sequence after the second process chain enters the candidate transfer unit and the change in the planned start time of the shifted production task, calculate the process waiting transmission amount and the task sequence offset. Based on the delivery commitments, resource occupancy windows, and resource demand duration of the shifted production tasks, calculate the debt repayment tension and backfilling period misalignment. Calculate the committed diffusion amount based on the process waiting transmission amount and the number of production tasks shifted after the second process chain enters the candidate transfer unit; The commitment intrusion vector is composed of delivery date buffer penetration amount, resource switching increment, process waiting transmission amount, task sequence offset amount, and debt repayment tension of the transferred debt. The commitment intrusion vector, backfill deadline misalignment amount, and commitment associated diffusion amount are input into the reachability coupling evaluation function to calculate the commitment intrusion intensity of the candidate transfer unit. Based on the correspondence between candidate transfer units and committed intrusion strength, candidate transfer units are sorted to generate an intrusion evaluation sequence.

[0010] Optionally, the improved Pointer Network model specifically includes an encoding module, a decoding module, and a constraint verification module; The encoding module receives candidate transfer units and insertion intrusion evaluation sequences, establishes candidate input items according to the candidate transfer units, generates an initial encoding representation, and introduces a commitment state rolling mechanism to continuously update the initial encoding representation, thereby generating a dynamic encoding representation of the candidate transfer units. The decoding module receives the dynamic encoded representation, introduces a debt window repayment mechanism, calculates the pointer score of the candidate transfer unit, determines the target transfer unit according to the pointer score, and generates a transfer task sequence and a backfill task sequence. The constraint verification module performs process acceptance constraint verification based on the process sequence in the target transfer unit and the second production schedule data, and generates process acceptance verification results; it performs resource time window constraint verification based on the production resource identifier of the target transfer unit, the resource occupancy time window, and the resource demand duration in the second production schedule data, and generates resource time window verification results; it performs delivery commitment constraint verification based on the transfer task sequence, the backfill task sequence, the delivery commitment in the first production schedule data, and the delivery deadline in the second production schedule data, and generates delivery commitment verification results; if the process acceptance verification results, resource time window verification results, and delivery commitment verification results are all passed, then the verified target transfer unit, transfer task sequence, and backfill task sequence are generated.

[0011] Optionally, the commitment state rolling mechanism specifically includes: After the coding module receives the candidate transfer unit and the order intrusion evaluation sequence, it establishes a commitment state vector according to the remaining transfer capacity, remaining delivery period margin, remaining backfill capacity, outstanding transfer debt and backfill period misalignment of the candidate transfer unit. The production task identifier, process identifier, process sequence, production resource identifier, resource occupation window, delivery commitment, commitment intrusion intensity, and order insertion task identifier corresponding to the candidate transfer unit are encoded to generate an initial encoded representation; the commitment state vector is mapped to the initial encoded representation to generate a dynamic encoded representation. After the decoding module determines the target transfer unit, it updates the remaining transfer capacity, remaining delivery time balance, remaining backfill capacity, outstanding transfer debt, and backfill period misalignment based on the correlation between the target transfer unit and the candidate transfer units in the process acceptance relationship, production resource identification matching relationship, and delivery commitment relationship. Based on the updated remaining transfer capacity, remaining delivery time, remaining backfill capacity, outstanding transfer debt, and backfill period misalignment, a dynamic encoded representation is regenerated and transmitted to the decoding module.

[0012] Optionally, the debt transfer window repayment mechanism specifically includes: The decoding module receives the dynamic encoded representation generated by the encoding module and calculates the basic pointer score of the candidate transfer unit based on the dynamic encoded representation and the decoding state. Based on the outstanding transfer debt, newly transferred debt and remaining backfill capacity of the candidate transfer unit, calculate the transfer debt pressure of the candidate transfer unit; Based on the basic index score, commitment intrusion intensity, remaining transfer capacity, transfer debt pressure, and remaining backfill capacity, a modified index score is generated for the candidate transfer unit. The pointer probabilities of candidate transfer units are generated based on the corrected pointer scores, and the candidate transfer unit with the highest pointer probability is determined as the target transfer unit; a transfer task sequence is generated according to the production task identifier, process identifier and process sequence corresponding to the target transfer unit; A backfilling task sequence is generated based on the production resource identifier, resource occupation window, delivery commitment, remaining backfilling capacity, and outstanding transferred debt corresponding to the transfer task sequence.

[0013] Optionally, the step of performing commitment transfer reconstruction on the first production scheduling data and the second production scheduling data based on the target transfer unit, transfer task sequence, and backfill task sequence to generate a target production scheduling plan specifically involves: Based on the production resource identifier and resource occupation time window corresponding to the target transfer unit, allocate the second production scheduling task time window to the insertion task in the second production scheduling data; Based on the transfer task sequence, extract the transferred production tasks from the first production schedule data, release the resource occupation window corresponding to the transferred production tasks, and generate a transfer window change record. Based on the backfill task sequence, the production resource identifier, resource occupation time window and delivery commitment in the first production schedule data, the backfill resource occupation time window is allocated to the transferred production task, and the first production schedule backfill record is generated. Merge the second production scheduling task window, the transfer window change record and the first production scheduling backfill record into the first production scheduling data and the second production scheduling data to generate a commitment transfer reconstruction record; The process acceptance constraint verification, resource time window constraint verification, and delivery date commitment constraint verification are performed on the commitment transfer reconstruction record. If the verification result is passed, the target production schedule plan is generated.

[0014] Optionally, the step of updating the task execution order and resource occupancy window of the discrete manufacturing production line according to the target production scheduling plan to generate production scheduling optimization results specifically includes: Update the production task order in the first and second production scheduling data according to the task execution order in the target production scheduling plan, and generate a task execution order update record; Update the production resource identifier, planned start time, and planned completion time corresponding to the production task according to the resource occupancy time window in the target production schedule, and generate a resource occupancy time window update record; write the task execution order update record and the resource occupancy time window update record into the production schedule execution record of the discrete manufacturing production line; Based on the production scheduling execution record, associated production task identifiers, insertion task identifiers, target transfer units, transfer task sequences, and backfill task sequences, the production scheduling optimization results are generated.

[0015] The beneficial effects of this invention are: This invention obtains first production scheduling data, second production scheduling data, and the workshop execution status of the first production scheduling data. It performs commitment freeze determination and commitment flexibility determination on the production tasks in the first production scheduling data, generates transferable commitment units, and generates candidate transfer units based on the second production scheduling data and transferable commitment units. This allows the insertion task to no longer directly enter the overall production scheduling rescheduling process, but to be selected from the production segments that have commitment transfer conditions. An reachability coupling evaluation function is introduced, which calculates the commitment intrusion intensity of candidate transfer units based on delivery date buffer penetration, resource switching increment, process waiting transmission, task sequence offset, debt repayment tension, backfilling deadline misalignment, and commitment-related diffusion. This generates an insertion intrusion evaluation sequence, enabling the evaluation of candidate transfer units to simultaneously reflect delivery date commitment, resource window, task sequence, and backfilling repayment pressure. The candidate transfer units and the intrusion evaluation sequence are input into the improved Pointer Network model. A commitment state rolling mechanism is introduced in the encoding module, and a transfer debt window repayment mechanism is introduced in the decoding module. The target transfer units are determined and transfer task sequences and backfilling task sequences are generated. This allows the model to simultaneously consider the commitment intrusion intensity, remaining transfer capacity, remaining backfilling capacity, and outstanding transfer debt when selecting target transfer units. Attached Figure Description

[0016] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a flowchart of a discrete manufacturing production line scheduling optimization method based on artificial intelligence proposed in this invention; Figure 2 This is a schematic diagram of the improved Pointer Network model proposed in this invention; Figure 3 This is a data flow diagram of a discrete manufacturing production line scheduling optimization method based on artificial intelligence proposed in this invention. Detailed Implementation

[0017] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0018] refer to Figures 1-3 An artificial intelligence-based method for optimizing the scheduling of discrete manufacturing production lines includes the following steps: Obtain the first production scheduling data, the second production scheduling data, and the workshop execution status corresponding to the first production scheduling data of the discrete manufacturing production line; Based on the first production schedule data and the workshop execution status, the production tasks in the first production schedule data are subject to commitment freeze and commitment flexibility determination to generate transferable commitment units. Based on the second production scheduling data and the transferable commitment units, the process acceptance matching and resource window matching are performed to generate candidate transfer units; An reachability coupling evaluation function is introduced to calculate the committed intrusion strength of candidate transfer units and generate an insertion intrusion evaluation sequence. The candidate transfer unit and the insertion intrusion evaluation sequence are input into the improved Pointer Network model to determine the target transfer unit and generate the transfer task sequence and backfill task sequence. The improved Pointer Network model includes an encoding module, a decoding module and a constraint verification module. The commitment state rolling mechanism is introduced in the encoding module and the transfer debt window repayment mechanism is introduced in the decoding module. Based on the target transfer unit, transfer task sequence and backfill task sequence, the commitment transfer reconstruction is performed on the first production scheduling data and the second production scheduling data to generate the target production scheduling plan. Update the task execution sequence and resource usage window of the discrete manufacturing production line based on the target production schedule, and generate production schedule optimization results.

[0019] In this embodiment, the first production scheduling data, the second production scheduling data, and the workshop execution status corresponding to the first production scheduling data of the discrete manufacturing production line are obtained, specifically as follows: Retrieve the original production schedule record generated before the insertion task enters, and generate the first production schedule data by associating the production task identifier with the order identifier, process identifier, process sequence, production resource identifier, resource occupation time window, planned start time, planned completion time, and delivery commitment; the first production schedule data is the original production schedule plan data generated before the insertion task enters. Retrieve the insertion task record, associate the insertion task identifier with the insertion order identifier, product identifier, process route, process identifier, process sequence, resource requirement duration, and delivery deadline, and generate the second production schedule data; the second production schedule data is the proposed insertion schedule data corresponding to the insertion task; Obtain the workshop execution record corresponding to the first production data, and generate the workshop execution status by associating the task issuance status, task execution status, actual start time, actual completion time and production resource usage record according to the production task identifier.

[0020] In this embodiment, based on the first production scheduling data and the workshop execution status, the production tasks in the first production scheduling data are subject to commitment freeze determination and commitment flexibility determination to generate transferable commitment units, specifically: Based on the production task identifiers in the first production schedule data, the order identifiers, process identifiers, process sequences, production resource identifiers, resource occupancy time windows, planned start times, planned completion times, and delivery commitments in the first production schedule data are matched with the task issuance status, task execution status, actual start time, actual completion time, and production resource occupancy records in the workshop execution status to generate a production task execution comparison relationship. Based on the production task execution comparison relationship, production tasks whose execution status is completed, processing, or whose production resource occupation record has been actually occupied are given a commitment freeze mark. Based on the production task execution comparison relationship, production tasks whose task issuance status is not issued and have not yet formed an actual start time are regarded as unfrozen production tasks. For production tasks that are not frozen, the delivery date margin is obtained by subtracting the planned completion date from the delivery date commitment date; the pre-time window margin is obtained by subtracting the resource occupation time window start date from the planned start date; the post-time window margin is obtained by subtracting the planned completion date from the resource occupation time window end date; and the pre-time window margin and the post-time window margin are added together to obtain the time window adjustment margin. The second production data insertion requirement is generated based on the resource demand duration and delivery deadline in the second production data. The second production data insertion requirement includes that the processing time required for the insertion task must not exceed the delivery deadline margin, the processing time required for the insertion task must not exceed the time window adjustment margin, and the completion time of the insertion task after insertion must not be later than the delivery deadline. Unfrozen production tasks whose delivery date margin and time window adjustment margin both meet the requirements for inserting the second production scheduling data are identified as transferable production tasks. Based on the production task identifier, process identifier, process sequence, production resource identifier, resource occupation window, and delivery commitment corresponding to the transferable production task, establish the correspondence between the transferable production task and the first production schedule data according to the production task identifier; combine the production task identifier, process identifier, process sequence, production resource identifier, resource occupation window, delivery commitment, delivery margin, and window adjustment margin corresponding to the transferable production task to generate a transferable commitment unit.

[0021] In this embodiment, based on the second production scheduling data and the transferable commitment units, process acceptance matching and resource time window matching are performed to generate candidate transferable units, specifically: Based on the insertion task identifier, process route, process identifier, and process sequence in the second production scheduling data, the insertion processes are arranged according to the sequential processing relationship corresponding to the process identifier in the process route to generate the second process chain; Based on the order insertion task identifier, production resource identifier, resource requirement duration and delivery deadline in the second production scheduling data, a second resource requirement time window is established according to the production resource identifier and resource requirement duration corresponding to the order insertion process. Based on the process identifier and process sequence in the transferable commitment unit, the process identifier and process sequence of the insert process in the second process chain are compared with the corresponding process of the transferable commitment unit to generate process acceptance matching results. The process acceptance matching results include process identifier comparison results and process sequence comparison results. The process identifier comparison results indicate the process category correspondence between the insert process in the second process chain and the corresponding process of the transferable commitment unit. The process sequence comparison results indicate the sequential acceptance relationship between the insert process and the corresponding process of the transferable commitment unit after the second process chain enters the transferable commitment unit. Based on the production resource identifier and resource occupation time window in the transferable commitment unit, the production resource identifier in the second resource demand time window is compared with the production resource identifier in the transferable commitment unit to generate a resource identifier comparison result. Based on the resource demand duration in the second resource demand window and the resource occupancy window in the transferable commitment unit, the length of the time window that can be ceded in the transferable commitment unit is calculated, and a resource time window comparison result is generated. The resource time window comparison result represents the matching relationship between the resource demand duration corresponding to the second resource demand window and the length of the time window that can be ceded in the transferable commitment unit. Based on the delivery deadline in the second production schedule data and the delivery commitment in the transferable commitment unit, calculate the completion time of the inserted order and the completion time of the transferable production task after the second process chain enters the transferable commitment unit, and generate the delivery time comparison result. The transferable commitment unit that meets the requirements for inserting the second production scheduling data in terms of process acceptance matching results, resource identifier comparison results, resource time window comparison results, and delivery time limit comparison results is identified as the candidate transferable unit.

[0022] In this embodiment, an reachability coupling evaluation function is introduced to calculate the committed intrusion strength of candidate transfer units and generate an insertion intrusion evaluation sequence, specifically: For candidate transfer units, the delivery date commitment, planned start time, planned completion time, process identifier, process sequence, production resource identifier and resource occupation window are extracted from the first production scheduling data according to the production task identifier. The second process chain, resource requirement duration and delivery deadline are extracted from the second production scheduling data according to the insertion task identifier. Based on the resource occupancy window corresponding to the candidate transfer unit and the resource demand duration corresponding to the second process chain, an insertion occupancy window is generated after the second process chain enters the candidate transfer unit. Based on the insertion occupancy window, resource occupancy window, and planned completion time, calculate the change in planned completion time after the second process chain enters the candidate transfer unit; The value obtained by subtracting the planned completion time from the delivery commitment is used as the original delivery time margin. The value obtained by subtracting the planned completion time after the second process chain enters the candidate transfer unit from the delivery commitment is used as the inserted delivery time margin. The value obtained by subtracting the inserted delivery time margin from the original delivery time margin is used as the delivery time buffer compression amount. The portion of the delivery date buffer compression that exceeds the original delivery date margin is considered as the delivery date buffer penetration amount. Based on the order of production resource identifiers before the second process chain enters the candidate transfer unit, the number of changes in adjacent production resource identifiers and the switching time of adjacent production resources are counted to generate the original resource switching quantity; based on the order of production resource identifiers after the second process chain enters the candidate transfer unit, the number of changes in adjacent production resource identifiers and the switching time of adjacent production resources are counted to generate the resource switching quantity after insertion. The value obtained by subtracting the original resource switching amount from the inserted resource switching amount is used as the resource switching increment. Based on the process sequence and planned start time before the second process chain enters the candidate transfer unit, the original process succession relationship is generated; based on the process sequence and planned start time after the second process chain enters the candidate transfer unit, the inserted process succession relationship is generated; the value obtained by subtracting the planned start time of the corresponding production task in the original process succession relationship from the planned start time of the shifted production task in the inserted process succession relationship is summed to generate the process waiting transmission quantity. Based on the process sequence before the second process chain enters the candidate transfer unit and the process sequence after the second process chain enters the candidate transfer unit, the number of task pairs whose relative order of production tasks changes is counted, and task sequence offset is generated. Based on the changes in the planned completion time and delivery commitments of the shifted production tasks, the amount of debt to be transferred is generated. Based on the resource occupancy window, production resource identifier, and resource demand duration corresponding to the shifted production task, the length of the resource window that can accommodate the shifted production task before the delivery commitment is calculated, and the backfill capacity is generated; the value obtained by dividing the amount of transferred debt by the backfill capacity is used as the debt repayment tension. Based on the backfill completion time and delivery commitment of the delayed production task, calculate the time difference between the backfill completion time and the delivery commitment, and generate the backfill deadline misalignment. Based on the process waiting transmission amount and the number of production tasks shifted to the later stage, calculate the waiting transmission amount corresponding to each production task shifted to the later stage, and generate the commitment-related diffusion amount. The commitment intrusion vector is composed of delivery date buffer penetration, resource switching increment, process waiting propagation, task sequence offset, and debt repayment tension. The commitment intrusion vector, backfilling deadline misalignment, and commitment-related diffusion are input into the reachability coupling evaluation function to calculate the commitment intrusion strength of candidate transfer units. The reachability coupling evaluation function is as follows: ; ; in, This indicates the commitment intrusion strength of the candidate transfer unit u. This represents the commitment intrusion vector of the candidate transfer unit u. Indicates the delivery buffer penetration amount. Indicates incremental resource switching. This indicates the amount of material the process is waiting to be transferred. Indicates the task sequence offset. This indicates a transfer of debt repayment tension. This indicates the amount of misalignment in the backfilling deadline. Indicates the commitment to the amount of spillover. M represents the dynamic weight vector, and M represents the coupling matrix. and Represents the evaluation coefficient. Represents a nonnegative mapping function; For each candidate transfer unit, calculate the corresponding committed intrusion strength and establish the correspondence between candidate transfer units and committed intrusion strength; Candidate transfer units are arranged in ascending order of committed intrusion intensity to generate an insertion intrusion evaluation sequence.

[0023] In this embodiment, the reachability coupling evaluation function is based on the technical understanding that the insertion impact during the order scheduling process of discrete manufacturing production lines not only occurs in the candidate transfer unit itself, but also propagates to the transferred production task along the process sequence, resource occupation time window, and delivery commitment in the first scheduling data. Specifically, the delivery buffer penetration amount, resource switching increment, process waiting transmission amount, task sequence offset, and transfer debt repayment tension generated after the second process chain enters the candidate transfer unit are used as commitment intrusion vectors. The backfilling deadline misalignment amount and commitment-related diffusion amount are used as independent evaluation quantities reflecting backfilling reachability and disturbance diffusion degree. The dynamic weight vector characterizes the different candidate transfer units in terms of delivery time, resources, and task sequence. The impact of order factors is represented by a coupling matrix, which characterizes the cross-influence relationships among delivery buffer penetration, resource switching increment, process waiting transmission, task sequence offset, and debt repayment tension. A non-negative mapping function is used to generate the commitment intrusion intensity of candidate transfer units. This approach eliminates reliance on a single delay time, single equipment idle time, or single priority indicator for evaluating candidate transfer units. Instead, it simultaneously reflects the impact of second-order data insertion on the delivery time compression, resource switching, process waiting, task sequence disturbance, and backfill repayment pressure of the first-order data. This allows the order insertion intrusion evaluation sequence to more accurately represent the commitment transfer cost of different candidate transfer units, providing stable input for improving the Pointer Network model's selection of target transfer units, generation of transfer task sequences, and backfill task sequences. This reduces the disturbance range of order insertion tasks on the original production schedule and improves the adaptability of the target production plan to process acceptance relationships, resource window relationships, and delivery commitment relationships.

[0024] In this embodiment, the improved Pointer Network model specifically includes an encoding module, a decoding module, and a constraint verification module; The encoding module receives candidate transfer units and insertion intrusion evaluation sequences, establishes candidate input items according to the candidate transfer units, generates initial encoding representations, and introduces a commitment state rolling mechanism to continuously update the initial encoding representations, thereby generating dynamic encoding representations of candidate transfer units. The decoding module receives the dynamically encoded representation, introduces a debt window repayment mechanism, calculates the pointer score of the candidate transfer unit, determines the target transfer unit according to the pointer score, and generates a transfer task sequence and a backfill task sequence. The constraint verification module receives the target transfer unit, transfer task sequence, backfill task sequence, first production scheduling data, and second production scheduling data. Based on the insertion task identifier corresponding to the target transfer unit, it extracts the insertion order identifier, process identifier, process sequence, resource requirement duration, and delivery deadline from the second production scheduling data. Based on the production task identifier corresponding to the target transfer unit, it extracts the order identifier, process identifier, process sequence, production resource identifier, resource occupation window, planned start time, planned completion time, and delivery commitment from the first production scheduling data. It then compares the process identifier and process sequence in the second production scheduling data with the process identifier and process sequence corresponding to the target transfer unit to determine the acceptance position and generates a process acceptance comparison record. If the order of the inserted process in the process acceptance comparison record has not changed, and the order of the process in the production task corresponding to the target transfer unit has not been reversed, a successful process acceptance verification result will be generated; if the order of the inserted process in the process acceptance comparison record has changed, or the order of the process in the production task corresponding to the target transfer unit has been reversed, a failed process acceptance verification result will be generated. The production resource identifier corresponding to the target transfer unit is compared with the production resource identifier in the second production scheduling data to generate a resource identifier comparison record; the resource occupancy time window length corresponding to the target transfer unit is compared with the resource demand duration in the second production scheduling data to generate a resource time window comparison record. If the resource identifier comparison records are consistent and the resource occupancy window length is not less than the resource demand duration, a successful resource time window verification result is generated; if the resource identifier comparison records are inconsistent, or the resource occupancy window length is less than the resource demand duration, a failed resource time window verification result is generated. Based on the production task identifier in the assigned task sequence, the delivery commitment and planned completion time are extracted from the first production schedule data; based on the resource occupation window and resource demand duration in the backfill task sequence, the backfill completion time of the assigned production task is generated; the backfill completion time of the assigned production task is compared with the delivery commitment in the first production schedule data to generate the first delivery comparison record. Based on the resource occupancy window in the target transfer unit and the resource demand duration in the second production schedule data, the completion time of the insertion task is generated; the completion time of the insertion task is compared with the delivery deadline in the second production schedule data to generate a second delivery date comparison record. If the completion time of backfilling in the first delivery date comparison record is not later than the delivery date commitment, and the completion time of the insertion task in the second delivery date comparison record is not later than the delivery deadline, a successful delivery date commitment verification result is generated. If the completion time of backfilling in the first delivery date comparison record is later than the delivery date commitment, or the completion time of the insertion task in the second delivery date comparison record is later than the delivery deadline, a failed delivery date commitment verification result is generated. If the process acceptance verification result, resource time window verification result, and delivery date commitment verification result are all successful, a successful target transfer unit, transfer task sequence, and backfilling task sequence are generated. If any of the process acceptance verification result, resource time window verification result, and delivery date commitment verification result are unsuccessful, the target transfer unit, transfer task sequence, and backfilling task sequence are returned to the decoding module for regeneration.

[0025] In its implementation, the improved Pointer Network model inherits the encoder-decoder-pointer selection framework of the traditional Pointer Network model, including an encoding module, a decoding module, and a constraint verification module. However, it introduces a commitment state rolling mechanism and a debt window compensation mechanism during the candidate transfer unit encoding process and the pointer selection process, respectively. This allows the model to no longer select sequences solely based on the static features of candidate input items, but to simultaneously perceive the commitment consumption, debt changes, and backfilling capacity changes of the first production data as the target transfer unit is gradually determined. Traditional Pointer Network models typically encode the input sequence into a fixed hidden state all at once, and the decoder directs the input sequence to the target position through attention distribution. This invention, targeting the order insertion scheduling scenario of discrete manufacturing production lines, uses candidate transfer units and order insertion intrusion evaluation sequences as model inputs, transforming the pointer selection object from ordinary sequence elements into candidate transfer units that can accept the insertion of second production data.

[0026] In the encoding module, the production task identifier, process identifier, process sequence, production resource identifier, resource occupation window, delivery commitment, order insertion task identifier, and commitment intrusion strength corresponding to the candidate transfer unit are established as candidate input items. The production task identifier, process identifier, production resource identifier, and order insertion task identifier are embedded to obtain the identifier embedding vector. The process sequence is positionally encoded to obtain the process sequence position vector. The start time, end time, and window length in the resource occupation window are numerically mapped to obtain the resource occupation window vector. The commitment intrusion strength is numerically mapped to the sorting position in the order insertion intrusion evaluation sequence to obtain the intrusion evaluation vector. The identifier embedding vector, process sequence position vector, resource occupation window vector, and intrusion evaluation vector are vector-connected according to a fixed dimension order, and the initial encoded representation is obtained through linear mapping. Unlike traditional Pointer Network models that only perform static encoding on the input sequence, this invention further introduces a commitment state rolling mechanism in the encoding module. The remaining transfer capacity, remaining delivery time margin, remaining backfill capacity, outstanding transfer debt, and backfill deadline misalignment of candidate transfer units are established as commitment state vectors. These commitment state vectors are then mapped to the initial encoding representation to generate a dynamic encoding representation. When the decoding module determines the target transfer unit, the encoding module updates the remaining transfer capacity, remaining delivery time margin, remaining backfill capacity, outstanding transfer debt, and backfill deadline misalignment based on the impact of the target transfer unit on the candidate transfer units in process acceptance, production resource identification matching, and delivery commitment relationships. The dynamic encoding representation is then regenerated based on the updated results, ensuring that the next pointer selection is based on the changed production commitment state.

[0027] In the decoding module, the dynamic encoded representation generated by the encoding module is input into the decoder, and the basic pointer score of the candidate transfer unit is calculated in combination with the decoding state. In the debt transfer window repayment mechanism, the debt transfer pressure is calculated based on the outstanding debt transfer, newly transferred debt transfer, and remaining backfilling capacity corresponding to the candidate transfer unit. The commitment intrusion intensity, remaining transfer capacity, debt transfer pressure, and remaining backfilling capacity are jointly introduced into the pointer score correction process to generate a corrected pointer score. The pointer probability of the candidate transfer unit is generated based on the corrected pointer score, and the candidate transfer unit with the highest pointer probability is determined as the target transfer unit. The transfer task sequence is generated according to the production task identifier, process identifier, and process sequence corresponding to the target transfer unit. The backfilling task sequence is generated according to the production resource identifier, resource occupation window, delivery commitment, remaining backfilling capacity, and outstanding debt transfer corresponding to the transfer task sequence. Unlike traditional Pointer Network models that directly select the position of the input sequence based on attention scores, the decoding module of this invention considers the commitment intrusion strength, transfer capability, transfer debt pressure, and backfilling repayment capability of the candidate transfer unit when selecting pointers. This ensures that the selection result of the target transfer unit is not limited to the optimal local pointer score, but simultaneously meets the scheduling requirements of the second scheduling data insertion and the first scheduling data backfilling compensation.

[0028] In the constraint verification module, the target transfer unit, transfer task sequence, backfill task sequence, first production scheduling data, and second production scheduling data are used as verification inputs. Based on the process sequence in the target transfer unit and second production scheduling data, process acceptance constraint verification is performed, generating process acceptance verification results. Based on the production resource identifier of the target transfer unit, resource occupancy time window, and resource demand duration in the second production scheduling data, resource time window constraint verification is performed, generating resource time window verification results. Based on the transfer task sequence, backfill task sequence, delivery date commitment in the first production scheduling data, and delivery deadline in the second production scheduling data, delivery date commitment constraint verification is performed, generating delivery date commitment verification results. When the process acceptance verification results, resource time window verification results, and delivery date commitment verification results are all passed, the verified target transfer unit, transfer task sequence, and backfill task sequence are generated. Through the above processing, the improved Pointer Network model transforms the input position selection capability of the traditional Pointer Network model into the commitment transfer selection capability for order insertion and production scheduling. Through dynamic encoding and debt-repayment decoding, the model can simultaneously generate the transfer task sequence and backfill task sequence of the first production data during the second production data insertion process, thereby reducing the disturbance range of the order insertion task on the original production schedule.

[0029] In this embodiment, the commitment state rolling mechanism is specifically as follows: The remaining transfer capacity, remaining delivery time margin, remaining backfill capacity, outstanding transfer debt, and backfill maturity misalignment of candidate transfer unit u before the t-th decoding are used to form a commitment state vector: ; in, This represents the commitment state vector of candidate transfer unit u before the t-th decoding. Indicates remaining transferable capacity. This indicates the remaining delivery time margin. Indicates the remaining backfill capacity. This indicates that the outstanding debt transferred has not been repaid. This indicates the amount of misalignment in the backfill deadline; Initial encoding is performed on the production task identifier, process identifier, process sequence, production resource identifier, resource occupancy window, delivery commitment, commitment intrusion intensity, and order insertion task identifier corresponding to the candidate transfer unit u, generating an initial encoding representation. ; Based on the commitment state vector and initial encoding representation Computation of dynamic coding representation: ; in, This represents the dynamic encoded representation of the candidate transfer unit u before the t-th decoding iteration. Represents the state mapping matrix; If the decoding module determines the target transfer unit in the t-th decoding iteration Then, based on the candidate transfer unit u and the target transfer unit... Calculate the correlation impact coefficient based on the process succession relationship, production resource identification matching relationship, and delivery commitment relationship between them. ; Update the commitment state vector: ; ; ; ; ; in, Indicates the target transfer unit The duration of resource requirements occupied by the second row of production data. Indicates the target transfer unit Changes in the planned completion time of the corresponding production task. Indicates the target transfer unit Corresponding backfilling capacity Indicates the target transfer unit The resulting debt transfer Indicates the target transfer unit The resulting misalignment in the backfilling period; The dynamic encoded representation before the (t+1)th decoding is calculated based on the updated commitment state vector.

[0030] In this implementation, the debt transfer window repayment mechanism is specifically as follows: The dynamic encoding representation of candidate transfer unit u before the t-th decoding. and decoding state Input pointer scoring function, get base pointer score: ; in, This represents the base pointer score of the candidate transfer unit u in the t-th decoding iteration. , , and Indicates model parameters; The debt pressure is calculated based on the outstanding debt transferred, the debt transferred by the target transfer unit, and the remaining backfilling capacity. ; in, This represents the debt pressure that candidate transfer unit u receives during the t-th decoding iteration. This represents the debt transferred after candidate transfer unit u is selected; The revised indicator score is calculated based on the base indicator score, the intensity of committed encroachment, the remaining transfer capacity, the debt transfer pressure, and the remaining backfill capacity. ; in, This indicates the score for correcting the pointer. This represents the normalized commitment intrusion strength. This represents the normalized residual transferability. This indicates the transfer of debt pressure after normalization. This represents the normalized remaining backfill capacity. , , and Indicates the decoding coefficients; Calculate the pointer probability based on the corrected pointer score: ; in, This represents the pointer probability of the candidate transfer unit u in the t-th decoding iteration. This represents the set of candidate transfer units to be selected during the t-th decoding iteration; The candidate transfer unit corresponding to the maximum pointer probability is determined as the target transfer unit; Generate a transfer task sequence according to the production task identifier, process identifier, and process sequence corresponding to the target transfer unit; A backfilling task sequence is generated based on the production resource identifier, resource occupation window, delivery commitment, remaining backfilling capacity, and outstanding transferred debt corresponding to the transfer task sequence.

[0031] In this embodiment, based on the target transfer unit, the transfer task sequence, and the backfill task sequence, a commitment transfer reconstruction is performed on the first production scheduling data and the second production scheduling data to generate a target production scheduling plan, specifically as follows: Based on the production resource identifier and resource occupation window corresponding to the target transfer unit, allocate the second production scheduling task window to the insertion task in the second production scheduling data; based on the transfer task sequence, extract the transferred production task from the first production scheduling data, release the resource occupation window corresponding to the transferred production task, and generate a transfer window change record; based on the backfill task sequence, the production resource identifier, resource occupation window and delivery commitment in the first production scheduling data, allocate a backfill resource occupation window to the transferred production task, and generate a first production scheduling backfill record; The second production scheduling task time window, the transfer time window change record, and the first production scheduling backfill record are merged into the first production scheduling data and the second production scheduling data to generate a commitment transfer reconstruction record. The commitment transfer reconstruction record is then subjected to process acceptance constraint verification, resource time window constraint verification, and delivery date commitment constraint verification. If the verification result is passed, the target production scheduling plan is generated.

[0032] In this embodiment, the task execution order and resource usage time window of the discrete manufacturing production line are updated according to the target production scheduling plan to generate the production scheduling optimization result, specifically: Update the production task order in the first and second production scheduling data according to the task execution order in the target production scheduling plan, and generate a task execution order update record; Update the production resource identifier, planned start time, and planned completion time corresponding to the production task according to the resource occupancy time window in the target production schedule, and generate a resource occupancy time window update record; write the task execution order update record and the resource occupancy time window update record into the production schedule execution record of the discrete manufacturing production line; Based on the production scheduling execution record, associated production task identifiers, insertion task identifiers, target transfer units, transfer task sequences, and backfill task sequences, the production scheduling optimization results are generated.

[0033] Example 1: To verify the feasibility of this invention in practice, it was applied to a discrete manufacturing production line for automotive parts. This production line includes CNC machining, drilling and tapping, heat treatment, precision grinding, cleaning, and final inspection processes, mainly producing parts such as steering knuckle brackets, shock absorber connecting seats, and motor end covers. Before the order was inserted, the first production scheduling data contained 26 incomplete production tasks, involving 6 types of production resources and 38 resource occupancy windows. Among them, 9 production tasks had been issued, and 5 were in the processing stage. A customer temporarily added 120 steering knuckle brackets, with a delivery deadline of 18 hours after the order was inserted, forming the second production scheduling data. Traditionally, the order is usually inserted directly into the CNC machining and precision grinding processes according to the priority of urgent orders, resulting in the overall postponement of many production tasks in the original production schedule. After acquiring the first production scheduling data, the second production scheduling data, and the workshop execution status, this invention assigns a commitment freeze flag to production tasks in the completed and processing states, calculates delivery time margins and time window adjustment margins for unfrozen production tasks, and generates 7 transferable commitment units; obtains 4 candidate transfer units through process acceptance matching and resource time window matching; calculates the commitment intrusion intensity through the reachability coupling evaluation function, and generates an order insertion intrusion evaluation sequence; inputs the candidate transfer units and the order insertion intrusion evaluation sequence into the improved Pointer Network model, determines 2 target transfer units, generates 3 transfer task sequences and 3 backfill task sequences, and generates a target production scheduling scheme through commitment transfer reconstruction.

[0034] Table 1 Comparison of Data on Order Scheduling Disturbance Control

[0035] As shown in Table 1, while the traditional priority-based order insertion scheme can free up processing space for 120 insertion tasks, it requires changes to 18 production tasks, of which 14 tasks have altered execution order, indicating a significant spread of the insertion's impact. This invention limits the insertion's acceptance scope through transferable commitment units and candidate transferable units, refactoring only 6 production tasks and reducing the actual number of tasks whose execution order is changed to 4. The average commitment intrusion intensity decreased from 0.74 to 0.38, indicating that the reachability coupling evaluation function can screen out target transferable units with lower delivery time buffer compression, resource switching increments, and debt repayment tension. The maximum process waiting time decreased from 4.6 hours to 1.7 hours, and the maximum postponement time of the original production schedule decreased from 5.2 hours to 1.9 hours, demonstrating that the transferable task sequence and backfill task sequence can effectively suppress the continued spread of insertion disturbances along the process sequence.

[0036] Table 2 Comparison of Delivery and Resource Utilization

[0037] As shown in Table 2, the target scheduling scheme of this invention not only meets the delivery deadline of the inserted tasks but also provides more significant protection for the original scheduling tasks. The traditional priority insertion scheme caused delays in six original scheduling tasks, with an average delay of 2.3 hours. This was mainly because the inserted tasks directly occupied the resource occupancy windows of key CNC equipment and grinding equipment, resulting in a lack of backfilling arrangements for the delegated production tasks. This invention, by improving the Pointer Network model to determine the target transfer unit and generating a transfer task sequence and a backfilling task sequence, reduces the completion time of the inserted tasks to 16.8 hours, reduces the number of delayed original scheduling tasks to one, and increases the delivery commitment fulfillment rate to 96.2%. Simultaneously, the number of resource switching operations decreased from 17 to 9, and the idle fragmentation time of the grinding equipment decreased by 2.3 hours. This indicates that the commitment transfer reconstruction can balance the insertion of inserted tasks and the backfilling of original scheduling tasks, improving the adaptability of the scheduling optimization results to the resource occupancy windows.

[0038] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A production scheduling optimization method for discrete manufacturing production lines based on artificial intelligence, characterized in that, Includes the following steps: Acquire the first production scheduling data, the second production scheduling data, and the workshop execution status corresponding to the first production scheduling data of the discrete manufacturing production line; Based on the first production scheduling data and the workshop execution status, the production tasks in the first production scheduling data are subject to commitment freeze determination and commitment flexibility determination, and a transferable commitment unit is generated. Based on the second production scheduling data and the transferable commitment unit, process acceptance matching and resource window matching are performed to generate candidate transfer units; An reachability coupling evaluation function is introduced to calculate the committed intrusion strength of candidate transfer units and generate an insertion intrusion evaluation sequence. The candidate transfer unit and the insertion intrusion evaluation sequence are input into the improved Pointer Network model to determine the target transfer unit and generate the transfer task sequence and backfill task sequence. The improved Pointer Network model includes an encoding module, a decoding module and a constraint verification module. A commitment state rolling mechanism is introduced in the encoding module and a transfer debt window repayment mechanism is introduced in the decoding module. Based on the target transfer unit, the transfer task sequence and the backfill task sequence, the commitment transfer reconstruction is performed on the first production scheduling data and the second production scheduling data to generate the target production scheduling plan. Update the task execution sequence and resource usage window of the discrete manufacturing production line based on the target production schedule, and generate production schedule optimization results.

2. The method for optimizing production scheduling of discrete manufacturing lines based on artificial intelligence according to claim 1, characterized in that, The acquisition of the first production scheduling data, the second production scheduling data, and the workshop execution status corresponding to the first production scheduling data of the discrete manufacturing production line specifically includes: Retrieve the original production schedule record formed before the insertion task enters, and generate the first production schedule data by associating the production task identifier with the order identifier, process identifier, process sequence, production resource identifier, resource occupation time window, planned start time, planned completion time and delivery commitment; Retrieve the order insertion task record, associate the order insertion task identifier with the order insertion order identifier, product identifier, process route, process identifier, process sequence, resource requirement duration and delivery deadline, and generate the second production scheduling data; Obtain the workshop execution record corresponding to the first production scheduling data, and generate the workshop execution status by associating the task issuance status, task execution status, actual start time, actual completion time and production resource occupancy record with the production task identifier.

3. The method for optimizing the scheduling of discrete manufacturing production lines based on artificial intelligence according to claim 1, characterized in that, Based on the first production scheduling data and the workshop execution status, the production tasks in the first production scheduling data are subject to commitment freeze and commitment flexibility determination to generate transferable commitment units, specifically as follows: Based on the production task identifier in the first production scheduling data, the first production scheduling data is matched with the workshop execution status to generate a production task execution comparison relationship; Based on the aforementioned production task execution comparison relationship, if the task execution status of a production task is completed or in the processing state, then a commitment freeze flag is assigned to the production task. For production tasks without a commitment freeze flag, calculate the delivery time margin between the delivery commitment and the planned completion time, and calculate the time window adjustment margin between the resource occupancy window and the planned start and completion times; Production tasks whose delivery date margin and time window adjustment margin both meet the requirements for inserting the second production scheduling data are identified as transferable production tasks. Generate a transferable commitment unit based on the production task identifier, process identifier, process sequence, production resource identifier, resource occupation window, and delivery commitment corresponding to the transferable production task.

4. The method for optimizing the scheduling of discrete manufacturing production lines based on artificial intelligence according to claim 1, characterized in that, The step of generating candidate transfer units based on the second production scheduling data and the transferable commitment unit's process acceptance matching and resource time window matching is as follows: Based on the order insertion task identifier, process route, and process sequence in the second production scheduling data, a second process chain is generated; based on the order insertion task identifier, production resource identifier, resource requirement duration, and delivery deadline in the second production scheduling data, a second resource requirement window is generated. Based on the process identifier and process sequence in the transferable commitment unit, the second process chain is matched with the transferable commitment unit to generate a process acceptance matching result. Based on the production resource identifier and resource occupation window in the transferable commitment unit, the second resource demand window is matched with the transferable commitment unit to generate a resource window matching result. The transferable commitment unit that meets the requirements for inserting the second production scheduling data in both the process acceptance matching result and the resource time window matching result is identified as the candidate transferable unit.

5. The method for optimizing the scheduling of discrete manufacturing production lines based on artificial intelligence according to claim 1, characterized in that, The introduction of the reachability coupling evaluation function calculates the committed intrusion strength of candidate transfer units and generates an insertion intrusion evaluation sequence, specifically as follows: For candidate transfer units, the delivery date commitment, planned completion time, process sequence, production resource identifier and resource occupation window are extracted from the first production scheduling data according to the production task identifier, and the second process chain, resource requirement duration and delivery deadline are extracted from the second production scheduling data according to the order insertion task identifier. Calculate the delivery buffer penetration based on the change in planned completion time and delivery commitment after the second process chain enters the candidate transfer unit; Calculate the resource switching increment based on the production resource identifiers, resource occupation time windows, and resource demand duration before and after the second process chain enters the candidate transfer unit; Based on the change in process sequence after the second process chain enters the candidate transfer unit and the change in the planned start time of the shifted production task, calculate the process waiting transmission amount and the task sequence offset. Based on the delivery commitments, resource occupancy windows, and resource demand duration of the shifted production tasks, calculate the debt repayment tension and backfilling period misalignment. Calculate the committed diffusion amount based on the process waiting transmission amount and the number of production tasks shifted after the second process chain enters the candidate transfer unit; The commitment intrusion vector is composed of delivery date buffer penetration amount, resource switching increment, process waiting transmission amount, task sequence offset amount, and debt repayment tension of the transferred debt. The commitment intrusion vector, backfill deadline misalignment amount, and commitment associated diffusion amount are input into the reachability coupling evaluation function to calculate the commitment intrusion intensity of the candidate transfer unit. Based on the correspondence between candidate transfer units and committed intrusion strength, candidate transfer units are sorted to generate an intrusion evaluation sequence.

6. The method for optimizing the scheduling of discrete manufacturing production lines based on artificial intelligence according to claim 1, characterized in that, The improved Pointer Network model specifically includes an encoding module, a decoding module, and a constraint verification module; The encoding module receives candidate transfer units and insertion intrusion evaluation sequences, establishes candidate input items according to the candidate transfer units, generates an initial encoding representation, and introduces a commitment state rolling mechanism to continuously update the initial encoding representation, thereby generating a dynamic encoding representation of the candidate transfer units. The decoding module receives the dynamic encoded representation, introduces a debt window repayment mechanism, calculates the pointer score of the candidate transfer unit, determines the target transfer unit according to the pointer score, and generates a transfer task sequence and a backfill task sequence. The constraint verification module performs process acceptance constraint verification based on the process sequence in the target transfer unit and the second production data, and generates process acceptance verification results. Based on the production resource identifier of the target transfer unit, the resource occupation time window, and the resource demand duration in the second production data, perform resource time window constraint verification and generate resource time window verification results. Based on the transfer task sequence, backfill task sequence, delivery date commitment in the first production schedule data, and delivery deadline in the second production schedule data, the delivery date commitment constraint verification is performed, and the delivery date commitment verification result is generated. If the process acceptance verification result, resource window verification result, and delivery date commitment verification result are all passed, the target transfer unit, transfer task sequence, and backfill task sequence that have passed the verification are generated.

7. The method for optimizing the scheduling of discrete manufacturing production lines based on artificial intelligence according to claim 6, characterized in that, The commitment state rolling mechanism is as follows: After the coding module receives the candidate transfer unit and the order intrusion evaluation sequence, it establishes a commitment state vector according to the remaining transfer capacity, remaining delivery period margin, remaining backfill capacity, outstanding transfer debt and backfill period misalignment of the candidate transfer unit. The production task identifier, process identifier, process sequence, production resource identifier, resource occupation window, delivery commitment, commitment intrusion intensity, and order insertion task identifier corresponding to the candidate transfer unit are encoded to generate an initial encoded representation; the commitment state vector is mapped to the initial encoded representation to generate a dynamic encoded representation. After the decoding module determines the target transfer unit, it updates the remaining transfer capacity, remaining delivery time balance, remaining backfill capacity, outstanding transfer debt, and backfill period misalignment based on the correlation between the target transfer unit and the candidate transfer units in the process acceptance relationship, production resource identification matching relationship, and delivery commitment relationship. Based on the updated remaining transfer capacity, remaining delivery time margin, remaining backfill capacity, outstanding transfer debt, and backfill period misalignment, a dynamic encoded representation is regenerated and transmitted to the decoding module.

8. The method for optimizing the scheduling of discrete manufacturing production lines based on artificial intelligence according to claim 6, characterized in that, The aforementioned debt transfer window repayment mechanism is as follows: The decoding module receives the dynamic encoded representation generated by the encoding module and calculates the basic pointer score of the candidate transfer unit based on the dynamic encoded representation and the decoding state. Based on the outstanding transfer debt, newly transferred debt and remaining backfilling capacity of the candidate transfer unit, calculate the transfer debt pressure of the candidate transfer unit; Based on the basic index score, commitment intrusion intensity, remaining transfer capacity, transfer debt pressure, and remaining backfill capacity, a modified index score is generated for the candidate transfer unit. The pointer probabilities of candidate transfer units are generated based on the corrected pointer scores, and the candidate transfer unit with the highest pointer probability is determined as the target transfer unit. Generate a transfer task sequence according to the production task identifier, process identifier, and process sequence corresponding to the target transfer unit; A backfilling task sequence is generated based on the production resource identifier, resource occupation window, delivery commitment, remaining backfilling capacity, and outstanding transferred debt corresponding to the transfer task sequence.

9. The method for optimizing the scheduling of discrete manufacturing production lines based on artificial intelligence according to claim 1, characterized in that, The step of reconstructing the first and second production scheduling data based on the target transfer unit, transfer task sequence, and backfill task sequence to generate a target production scheduling plan is as follows: Based on the production resource identifier and resource occupation time window corresponding to the target transfer unit, allocate the second production scheduling task time window to the insertion task in the second production scheduling data; Based on the transfer task sequence, extract the transferred production tasks from the first production schedule data, release the resource occupation window corresponding to the transferred production tasks, and generate a transfer window change record. Based on the backfill task sequence, the production resource identifier, resource occupation time window and delivery commitment in the first production schedule data, the backfill resource occupation time window is allocated to the transferred production task, and the first production schedule backfill record is generated. Merge the second production scheduling task window, the transfer window change record and the first production scheduling backfill record into the first production scheduling data and the second production scheduling data to generate a commitment transfer reconstruction record; The process acceptance constraint verification, resource time window constraint verification, and delivery date commitment constraint verification are performed on the commitment transfer and reconstruction record. If the verification result is passed, the target production schedule is generated.

10. The method for optimizing the scheduling of discrete manufacturing production lines based on artificial intelligence according to claim 1, characterized in that, The process of updating the task execution order and resource usage window of the discrete manufacturing production line based on the target production scheduling plan, and generating production scheduling optimization results, specifically includes: Update the production task order in the first and second production scheduling data according to the task execution order in the target production scheduling plan, and generate a task execution order update record; Update the production resource identifier, planned start time, and planned completion time corresponding to the production task according to the resource occupancy time window in the target production scheduling plan, and generate a resource occupancy time window update record; Write the task execution order update record and resource usage window update record into the production scheduling execution record of the discrete manufacturing production line; Based on the production scheduling execution record, associated production task identifiers, insertion task identifiers, target transfer units, transfer task sequences, and backfill task sequences, the production scheduling optimization results are generated.

Images