Production plan scheduling method and system of MES system

Abstract

The invention belongs to the technical field of production plan scheduling, and discloses a production plan scheduling method and system of an MES system. The method comprises the following steps: preprocessing production order data to generate a comprehensive scheduling value priority sequence; based on the comprehensive scheduling value priority sequence, compatibility combination groups are sequentially obtained, a scheduling scheme population is constructed, multi-objective evolution fitness is calculated based on the delivery time satisfaction rate, the switching cost, the work-in-process turnover time and the repair risk index, non-dominated sorting, congestion distance evaluation, selection, intersection and mutation operation are executed, and an optimal scheduling scheme is generated; verifying the optimal scheduling scheme, and generating a verification passing scheme set and a verification failing scheme set; if the verification is not passed, returning to re-optimize, and if the verification is not passed, generating a scheduling instruction and issuing and executing the scheduling instruction; the production scheduling accuracy is improved, the order response period is shortened, and the process switching burden is reduced.

Description

Technical Field

[0001] This invention relates to the field of production planning and scheduling technology, and more specifically, to a production planning and scheduling method and system for a MES system. Background Technology

[0002] Manufacturing enterprises operating under a multi-variety, small-batch production model generally face challenges such as tight order deadlines, complex and diverse process routes, limited resource allocation, and high rework risks. Existing production planning and scheduling methods typically employ rule-based scheduling strategies or single-indicator priority ranking methods, combined with limited process compatibility assessments and experience-based adjustments to arrange production tasks. While some scheduling systems possess order merging and batch sorting capabilities, they often rely on fixed-weight linear evaluation models, lacking the ability to collaboratively optimize multiple objectives such as delivery time, changeover costs, turnaround time, and quality reliability. Consequently, they struggle to achieve a global balance between on-time delivery and production efficiency in complex order environments.

[0003] In existing technologies, due to the failure to integrate multi-objective evolutionary scheduling algorithms with process parameter similarity grouping methods, scheduling schemes often suffer from deficiencies in batch process compatibility, on-time delivery rates, and rework risks. This leads to increased complexity in process changeovers during production, extended order response cycles, and decreased production line resource utilization. Furthermore, existing scheduling schemes lack automatic closed-loop linkage with the production verification process. Scheduling results are executed directly without verification, which can easily cause production bottlenecks and plan failures, making it difficult to meet the comprehensive requirements of scientific production planning and scheduling and execution stability in highly flexible production environments.

[0004] Therefore, there is an urgent need for a production planning and scheduling method based on multi-objective evolutionary optimization, batch compatibility merging, and verification closed-loop to improve scheduling accuracy, shorten order response cycles, reduce process changeover burden, and minimize rework risks. In view of this, this invention proposes a production planning and scheduling method and system for a MES system to solve the above problems. Summary of the Invention

[0005] To overcome the aforementioned deficiencies of the prior art and achieve the above objectives, the present invention provides the following technical solution: a production planning and scheduling method for a MES system, comprising:

[0006] Step 1: Analyze, identify and classify the received N production order data to generate a comprehensive scheduling value priority sequence;

[0007] Step 2: Based on the comprehensive scheduling value priority sequence, obtain the compatibility merging group in sequence, and construct a population of scheduling schemes. Calculate the multi-objective evolutionary fitness based on delivery satisfaction rate, switching cost, work-in-process turnover time, and rework risk index. Perform non-dominated sorting, congestion distance assessment, selection, crossover, and mutation operations to generate the optimal scheduling scheme.

[0008] Step 3: Verify the optimal scheduling scheme, and generate a set of schemes that pass verification and a set of schemes that fail verification based on feasibility, stability and performance indicators;

[0009] Step 4: If the set of failed verification schemes is not empty, then the set of failed verification schemes will be used as the priority sequence of comprehensive scheduling value, and the process will return to Step 2 for execution; if the set of failed verification schemes is empty, then a scheduling instruction will be generated based on the set of successful verification schemes and executed.

[0010] Furthermore, the methods for generating the set of valid solutions and the set of invalid solutions include:

[0011] For each optimal scheduling scheme, obtain the corresponding production order data and construct a production order data set; initialize the feasibility flag, stability flag, and performance indicator flag to no; input the optimal scheduling scheme and the production order data set into the scheduling scheme evaluation model to obtain an evaluation result set including feasibility score, stability score, and performance indicator score;

[0012] The feasibility score, stability score, and performance index score are compared with their respective preset thresholds. If the score is greater than the threshold, the corresponding flag is set to "yes". The feasibility flag, stability flag, and performance index flag are all "yes". If all three are "yes", the optimal scheduling scheme is added to the set of verified schemes. If any flag is "no", the optimal scheduling scheme is added to the set of verified schemes that fail.

[0013] Furthermore, the method for generating the optimal scheduling scheme includes:

[0014] Obtain each compatibility merging group from the comprehensive scheduling value priority sequence. For each compatibility merging group, initialize a population of scheduling schemes. Each scheduling scheme includes batch production sequence, process changeover plan, production launch time plan, and resource allocation strategy. Based on the production order data corresponding to the compatibility merging group, construct the objective function of delivery fulfillment rate, objective function of changeover cost, objective function of work-in-process turnover time, and objective function of rework risk index. Calculate the multi-objective evolutionary fitness of each scheduling scheme.

[0015] A non-dominated sorting process is performed on the scheduling scheme population to obtain a Q-level Pareto sorting structure. The comprehensive congestion distance index of the scheduling scheme corresponding to the Pareto optimal front at each level is calculated. Based on the comprehensive congestion distance index, a binary tournament selection is performed, and mapping crossover and random mutation are performed to generate offspring scheduling schemes. The parent and offspring scheduling schemes are merged to form a new scheduling scheme population. After iterating to the preset maximum number of iterations, the scheduling scheme with the largest comprehensive congestion distance index in the first-level Pareto optimal front is selected as the optimal scheduling scheme.

[0016] Furthermore, the method for calculating the multi-objective evolutionary fitness includes:

[0017] The delivery time satisfaction rate is calculated based on the constructed objective function for delivery time satisfaction rate;

[0018] The switching cost is calculated based on the constructed switching cost objective function;

[0019] The work-in-process turnover time is calculated based on the constructed objective function for work-in-process turnover time.

[0020] The rework risk index is calculated based on the constructed objective function of the rework risk index.

[0021] Multi-objective evolutionary fitness is calculated based on delivery fulfillment rate, changeover cost, work-in-process turnover time, and rework risk index.

[0022] Furthermore, the method for obtaining the Q-layer Pareto sort structure includes:

[0023] S300: Let the initial value of q be 1, and the range of q is from 1 to Q;

[0024] S301: For all scheduling schemes in the scheduling scheme population, for any pair of scheduling schemes, let them be scheduling scheme one and scheduling scheme two;

[0025] If scheduling scheme 1 has better objective function values ​​than scheduling scheme 2 in all four dimensions of delivery date fulfillment rate, switching cost, work-in-process turnover time and rework risk index, and the multi-objective evolutionary fitness of scheduling scheme 1 is greater than that of scheduling scheme 2, then scheduling scheme 1 is determined to dominate scheduling scheme 2.

[0026] The number of times each scheduling scheme is dominated by other scheduling schemes is recorded as the number of dominated schemes. The scheduling scheme with a number of dominated schemes of 0 is recorded as the Pareto optimal frontier of the qth level.

[0027] The qth Pareto optimal front is taken as the qth layer of the Q-layer Pareto sorting structure;

[0028] Remove the scheduling scheme corresponding to the Pareto optimal front of the q-th layer from the scheduling scheme population;

[0029] S302: Let q = q + 1. If q is less than Q, return to S301 to continue execution. If q is greater than Q, end the current process.

[0030] Furthermore, the calculation method for the comprehensive congestion distance index of the scheduling scheme corresponding to each Pareto optimal frontier includes:

[0031] The Pareto optimal fronts of each layer are obtained sequentially from the Pareto sorting structure. The number of scheduling schemes in each Pareto optimal front is determined, and the scheduling schemes are sorted in ascending order according to their multi-objective evolutionary fitness to form a Pareto optimal front sequence. For each scheduling scheme in the Pareto optimal front sequence, the congestion distance increment is calculated based on the difference in objective function values ​​between adjacent scheduling schemes. The congestion distance increments in each objective function dimension are accumulated to obtain the corresponding comprehensive congestion distance index.

[0032] Furthermore, the method for obtaining the comprehensive scheduling value priority sequence includes:

[0033] Perform data preprocessing on the received N production order data;

[0034] Extract the process route identifier corresponding to each production order data, and merge production orders with the same process route identifier into the same process route group to form Y process route groups;

[0035] For each process route group, execute predefined process compatibility judgment rules on the production orders to form M compatibility merging groups;

[0036] Based on the production order data corresponding to the M compatibility merging groups, a comprehensive scheduling value assessment is performed to obtain the comprehensive scheduling value score corresponding to the M compatibility merging groups;

[0037] The M compatibility merging groups are sorted in descending order according to their corresponding comprehensive scheduling value scores to form a comprehensive scheduling value priority sequence.

[0038] Furthermore, the method for forming M compatibility merge groups includes:

[0039] Obtain the set of product process parameters for each production order in each process route group, summarize the set of product process parameters for the same group, and construct a grouped process parameter set;

[0040] For each set of process parameters in a group, the process parameters of any two production orders are compared item by item to calculate the process parameter similarity. If the process parameter similarity is higher than the preset process parameter similarity threshold, they are marked as compatible order pairs and a compatibility tag is generated. It is then determined whether the compatibility tag matches the compatibility tag of an existing compatibility merging group. If they match, the compatible order pair is added to the corresponding compatibility merging group. If they do not match, a new compatibility merging group is created and a compatibility tag is assigned, ultimately forming M compatibility merging groups.

[0041] Furthermore, the method for generating and executing scheduling instructions based on a set of schemes, according to verification, includes:

[0042] For each optimal scheduling scheme in the verified scheme set, the corresponding batch production sequence, process changeover plan, production launch time plan, and resource allocation strategy are extracted, and structured encapsulation is performed based on a predefined production scheduling instruction template to form a scheduling execution instruction file. The scheduling execution instruction file is automatically sent to the manufacturing execution system and execution units for execution.

[0043] A production planning and scheduling system for a MES system, used to implement the production planning and scheduling method of the MES system, includes:

[0044] The order preprocessing module is used to systematically parse, identify compatibility, automatically classify and merge the received N production order data to form a standardized comprehensive scheduling value priority sequence.

[0045] The scheduling optimization module, based on the comprehensive scheduling value priority sequence, sequentially obtains each compatibility merging group and constructs a population of scheduling schemes that includes batch production sequence, process changeover plan, production time plan and resource allocation strategy. It uses delivery satisfaction rate, changeover cost, work-in-process turnover time and rework risk index to perform multi-objective evolutionary fitness calculation, non-dominated sorting and congestion distance evaluation. It iteratively generates the optimal scheduling scheme through binary tournament selection, crossover and mutation operations.

[0046] The scheduling scheme verification module is used to verify the optimal scheduling scheme output by the scheduling optimization module in real time, evaluate the feasibility, stability and performance indicators of the optimal scheduling scheme under actual production conditions, and generate a set of schemes that pass verification and a set of schemes that fail verification.

[0047] If the set of failed verification schemes is not empty, the instruction generation and issuance module will use the set of failed verification schemes as the priority sequence of comprehensive scheduling value and return it to the scheduling optimization module for execution; if the set of failed verification schemes is empty, it will generate scheduling instructions based on the set of successful verification schemes and execute them.

[0048] Compared with existing technologies, the technical effects and advantages of the production planning and scheduling method and system of the MES system proposed in this invention are as follows:

[0049] This application integrates order preprocessing, process compatibility identification, comprehensive scheduling value priority assessment, multi-objective evolutionary scheduling optimization, scheduling scheme verification, and instruction generation and issuance into an integrated production planning and scheduling process, which can significantly improve the scientific nature, flexibility, and execution reliability of production scheduling in a multi-variety, small-batch production environment.

[0050] Compared to existing scheduling methods that rely solely on single-indicator ranking or manual experience adjustments, this application first utilizes process route identifiers and process parameter similarity to group and merge orders, reducing the complexity of process changeovers and improving the adaptability of batch merging. Through a multi-objective evolutionary scheduling optimization model constructed based on four dimensions—delivery fulfillment rate, changeover cost, work-in-process turnover time, and rework risk index—a comprehensive quantitative evaluation and global equilibrium optimization of scheduling schemes is achieved, taking into account delivery timeliness, production costs, workflow efficiency, and quality reliability.

[0051] Furthermore, this application verifies and hierarchically filters the optimal scheduling scheme through a scheduling scheme verification module, ensuring the feasibility and stability of the output scheme under actual production conditions. The instruction generation and issuance module automatically converts the verified scheduling scheme into scheduling instructions conforming to the Manufacturing Execution System (MES) interface specification, supporting the efficient issuance and execution of production plans. Overall, this application demonstrates significant benefits in improving production scheduling accuracy, shortening order response cycles, reducing process changeover burden and rework risks, enhancing production resource utilization, and increasing customer satisfaction. Attached Figure Description

[0052] Figure 1 This is a schematic diagram of a production planning and scheduling system of an MES system according to Embodiment 1 of the present invention;

[0053] Figure 2 This is a flowchart of a production planning and scheduling method for an MES system according to Embodiment 2 of the present invention;

[0054] Figure 3 A flowchart illustrating the method for generating the set of solutions that pass verification and the set of solutions that fail verification;

[0055] Figure 4 Flowchart of the method for generating the optimal scheduling scheme;

[0056] Figure 5 A flowchart illustrating the method for obtaining the overall scheduling value priority sequence;

[0057] Figure 6 This is a flowchart illustrating the method for calculating multi-objective evolutionary fitness. Detailed Implementation

[0058] The technical solutions of the embodiments of the present invention will be described in detail, clearly, and completely below with reference to the accompanying drawings. It should be particularly noted that the specific embodiments described below are only for better illustrating and explaining the technical solutions of the present invention, and are intended to enable those skilled in the art to better understand and implement the present invention, and should not be construed as limiting the scope of protection of the present invention. Without departing from the spirit and substance of the present invention, those skilled in the art can modify, adjust, or make equivalent substitutions based on the content disclosed in the present invention, and these should all be considered within the scope of protection of the present invention.

[0059] Example 1:

[0060] Please see Figure 1 As shown in the figure, this embodiment discloses a production planning and scheduling system for an MES system, including an order preprocessing module, a scheduling optimization module, a scheduling scheme verification module, and an instruction generation and issuance module. Each module is connected by wired and / or wireless means to realize data transmission.

[0061] The order preprocessing module is used to systematically parse, identify compatibility, automatically classify and merge the received N production order data to form a standardized comprehensive scheduling value priority sequence, thereby providing an accurate and unified data foundation for subsequent multi-objective evolutionary scheduling.

[0062] The production order data includes order number, product model, production quantity, delivery date, process route identifier, bill of materials, and product process parameter set; the product process parameter set includes equipment type, number of process sections, sequence of processes, set of line changeover requirements, and set of product production conditions.

[0063] It should be noted that the process route identifier is used to uniquely identify the code or ID corresponding to the product's process route.

[0064] In this embodiment, the changeover requirement refers to the preparation, adjustment, and calibration conditions required when switching production equipment between different product or process batches. Changeover requirements include, but are not limited to, equipment cleaning time, tooling and fixture replacement steps, parameter adjustment procedures, and production environment reset. For example, when the products produced earlier differ from those produced later in terms of material, size, or process, adaptive adjustments to the production line are necessary to ensure the production process meets quality standards. When performing process compatibility identification, the system determines whether there is a significant changeover operation load between two production orders based on the changeover requirements. If additional changeover steps are required or the changeover time exceeds a predefined threshold, the changeover is considered incompatible and will not be merged into the same batch. If the changeover requirements are within the set acceptable range, the order is considered compatible and can be included in the same candidate production batch.

[0065] In this application embodiment, product manufacturing conditions refer to the production environment, process parameters, or operational constraints that must be met for the characteristics of the process or product when performing each process step. Product manufacturing conditions include, but are not limited to, the ambient temperature, humidity, cleanliness level, specific atmosphere conditions required by the process, such as nitrogen protection, anti-static requirements, tooling preheating temperature, characteristics of the processing liquid, and related operational safety measures.

[0066] like Figure 5 As shown, the method for obtaining the comprehensive scheduling value priority sequence includes:

[0067] The received N production order data are preprocessed; the data preprocessing includes data cleaning, legality verification and data normalization. The data preprocessing process can be implemented based on existing data processing algorithms, which are common technical means known to those skilled in the art, and therefore will not be described in detail in this specification.

[0068] Extract the process route identifier corresponding to each production order data, and group production orders with the same process route identifier into the same process route group, forming Y process route groups. That is, production orders in each process route group share the same process route identifier and have a consistent process arrangement basis. This pre-grouping step significantly reduces the complexity of subsequent process compatibility determination and improves processing efficiency.

[0069] For each process route group, execute predefined process compatibility judgment rules on the production orders to form M compatibility merging groups;

[0070] Based on the production order data corresponding to the M compatibility merging groups, a comprehensive scheduling value assessment is performed to obtain the comprehensive scheduling value score corresponding to the M compatibility merging groups;

[0071] The M compatibility merging groups are sorted in descending order according to their corresponding comprehensive scheduling value scores to form a comprehensive scheduling value priority sequence.

[0072] The methods for forming M compatibility merge groups include:

[0073] S100: Let the initial value of y be 1, and the range of y is from 1 to Y;

[0074] S101: Obtain the set of product process parameters corresponding to each production order in the y-th process route group, and summarize the set of product process parameters to construct the y-th group process parameter set;

[0075] For all possible combinations of production orders in the y-th group of process parameters, compare each process parameter index and calculate the process parameter similarity; if the process parameter similarity is higher than the preset similarity threshold, mark the corresponding production order combination as a compatible order pair and generate a compatibility tag.

[0076] Determine whether the compatibility tag matches the compatibility tag corresponding to an existing compatibility merge group; if the determination result is a match, add the compatibility order pair to the corresponding compatibility merge group; if the determination result is a mismatch, create a new compatibility merge group based on the compatibility order pair, and assign a compatibility tag to the new compatibility merge group.

[0077] S102: Let y = y + 1. If y is less than or equal to y, return to S101 to continue execution. If y is greater than y, record the number of compatibility merge groups as M, obtain M compatibility merge groups, and end the current process.

[0078] The process parameter similarity is used to characterize the degree of closeness between two production orders in terms of the product process parameter set dimension. The product process parameter set contains several process parameter items, and the total number of process parameter items is J. For any production order to be compared, the values ​​of the two production orders on the same process parameter item are read respectively, and the process parameter similarity is obtained by scoring item by item and weighted summarization.

[0079] During the calculation process, two parameters are pre-configured for each process parameter: one is the maximum permissible deviation of the process parameter, which is used to limit the maximum acceptable range of difference for the process parameter in business operations, and the maximum permissible deviation is a positive value; the other is the weight coefficient of the process parameter, which is used to represent the contribution ratio of the process parameter to the overall similarity. The sum of the weight coefficients of all process parameter items is limited to 1. The weight coefficients are flexibly configured by those skilled in the art based on factors such as production business scenarios, line change sensitivity, and sensitivity to key quality characteristics.

[0080] For the j-th process parameter, the difference in value between the two production orders for this parameter is first calculated, represented by the absolute value of the difference between the two values. Then, this absolute difference is normalized and compared with the maximum permissible deviation of the j-th process parameter to obtain the deviation percentage, which reflects the proportion of the actual difference relative to the permissible range. When the deviation percentage is less than 1, the single-item similarity score for the j-th process parameter is set to "1 minus the deviation percentage," thus achieving a linear evaluation rule where the single-item score is 1 when the difference is 0 and close to 0 when the difference is close to the maximum permissible deviation. When the deviation percentage is greater than or equal to 1, the single-item similarity score is fixed at 0, indicating that the difference of this process parameter has exceeded the maximum permissible deviation range and no longer contributes to the similarity score. Subsequently, the single-item similarity score for each process parameter is multiplied by its corresponding weight coefficient to obtain a weighted score. The weighted scores of all J process parameter items are accumulated, and the accumulated result is used as the process parameter similarity between the two production orders. Since the individual similarity score is limited to the range of 0 to 1 and the sum of the weight coefficients is limited to 1, the process parameter similarity is also limited to the range of 0 to 1. The closer the process parameter similarity is to 1, the more similar the process parameters of the two production orders are; the closer the process parameter similarity is to 0, the more significant the difference in process parameters between the two production orders is.

[0081] In one embodiment, a process parameter similarity threshold is used to determine whether two production orders meet the business condition of similar process parameters. The process parameter similarity threshold can be set by those skilled in the art based on the statistical level of the actual business scenario: first, in the historical order pair set or the current order pair set to be scheduled, the process parameter similarity of each order pair is calculated using the method described above, and the average value of the process parameter similarity is obtained; then, the process parameter similarity threshold is set as the average value multiplied by a preset proportional coefficient, where the preset proportional coefficient is between 0 and 1 and configured according to business requirements, thereby adaptively adjusting the threshold according to the overall similarity level. For example, when the average value of the process parameter similarity is 1 and the preset proportional coefficient is 90%, the process parameter similarity threshold is set to 0.9. When the process parameter similarity of two production orders is not less than the process parameter similarity threshold, the two production orders are determined to meet the process parameter similarity condition and are used for subsequent compatibility merging group construction or scheduling optimization processing.

[0082] For example, in one embodiment of the present invention, the product process parameter set includes equipment type, number of process segments, sequence of operations, set of line changeover requirements, and set of product production conditions. To reflect the different degrees of influence of different parameter items on compatibility judgment, weight coefficients are pre-set for each of the above parameter items, and the sum of all weight coefficients is limited to 1; for example, the weight coefficient for equipment type can be set to 0.25, the weight coefficient for number of process segments to 0.15, the weight coefficient for sequence of operations to 0.20, the weight coefficient for set of line changeover requirements to 0.20, and the weight coefficient for set of product production conditions to 0.20. Meanwhile, the maximum allowable difference range is pre-configured for each parameter item to unify the evaluation criteria for differences in different dimensions: the equipment type must be completely consistent, and if the equipment types of two orders are inconsistent, the similarity score for that item is directly determined to be 0; the maximum allowable deviation for the number of process segments is set to 2 segments; the sequence of operations is measured by the "number of process positions with inconsistent sequence" and the maximum allowable difference is set to 3 inconsistent positions; the set of line change requirements is measured by the "number of inconsistent items in the two orders' line change requirement sets" and the maximum allowable difference is set to 3 items, where the number of inconsistent items is the number of items that appear only in one set and not in the other set; the set of product production conditions can be expanded into several production condition items and allowable deviations are set for each, and the "number of production condition items exceeding the allowable deviation" is used as the difference measure and the maximum allowable difference is set to 2 items.

[0083] In this example, two production orders are selected for comparison: The first production order uses a Class A injection molding machine, has 4 process segments, and the sequence of processes is preheating, injection molding, holding pressure, cooling and demolding. The set of line change requirements is {mold change, barrel cleaning, color masterbatch replacement}, and the set of product production conditions is {mold temperature 80℃, material temperature 220℃, holding pressure 5.0MPa, cooling time 60s}. The second production order uses a Class A injection molding machine, has 5 process segments, and the sequence of processes is preheating, injection molding, cooling, holding pressure and demolding. The set of line change requirements is {mold change, barrel cleaning}, and the set of product production conditions is {mold temperature 83℃, material temperature 228℃, holding pressure 5.2MPa, cooling time 70s}.

[0084] Similarity calculations are performed based on a calculation method that includes item-by-item difference calculation, difference normalization conversion, determination of individual similarity scores, and weighted summation: Since the equipment types are the same, the equipment type similarity score is 1 and contributes 0.25 to the weighted score; the difference in the number of process segments is 1 segment, compared to the maximum allowable deviation of 2 segments, resulting in an individual similarity score of 0.5 and a weighted contribution of 0.075; in the comparison of the sequence of processes, the 3rd and 4th positions are interchanged, resulting in 2 position inconsistencies, compared to the maximum allowable difference of 3 position inconsistencies, resulting in an individual similarity score of approximately 0.33 and a weighted contribution of approximately 0.066; in the comparison of the set of line change requirements, the first production order has one more item, "replace color masterbatch," compared to the second production order. The number of inconsistent items is 1. Compared to the maximum allowable difference of 3 items, the single-item similarity score obtained according to the decreasing rule is approximately 0.67, and the weighted contribution is approximately 0.134. In the comparison of the product production conditions set, the mold temperature difference of 3℃ does not exceed the corresponding allowable deviation, and the example of the corresponding allowable deviation is 5℃; the material temperature difference of 8℃ exceeds the corresponding allowable deviation, and the example of the corresponding allowable deviation is 5℃; the holding pressure difference of 0.2MPa does not exceed the corresponding allowable deviation, and the example of the corresponding allowable deviation is 0.5MPa; the cooling time difference of 10s does not exceed the corresponding allowable deviation, and the example of the corresponding allowable deviation is 10s. Therefore, the number of production condition items that exceed the allowable deviation is 1. Compared to the maximum allowable difference of 2 items, the single-item similarity score obtained according to the decreasing rule is 0.5, and the weighted contribution is 0.10.

[0085] By summing up the contributions of each item according to their weights, the similarity of the process parameters of the two production orders is approximately 0.625. The closer the similarity value is to 1, the more similar the process parameters of the two orders are. The results of this example show that the two orders have a high level of overall similarity when the equipment type is the same, the process sequence is partially interchangeable, the line change requirements are basically the same, and there are a few out-of-tolerance items in the production conditions. This can be used for subsequent compatibility merging group construction or scheduling optimization.

[0086] The methods for obtaining the comprehensive scheduling value score corresponding to the M compatibility merging groups include:

[0087] For each compatibility merging group, the production order data contained in the compatibility merging group are obtained sequentially to construct a compatibility merging group order data set. The compatibility merging group order data set is then input into the comprehensive scheduling value assessment model to obtain the corresponding comprehensive scheduling value score.

[0088] The training method for the comprehensive scheduling value assessment model includes:

[0089] The comprehensive scheduling value assessment model is used in MES rolling scheduling scenarios to output a comprehensive scheduling value score for candidate production orders or compatibility merging groups, and generate a comprehensive scheduling value priority sequence based on this score to drive the search direction for subsequent compatibility merging group construction, scheduling scheme population initialization, and multi-objective evolutionary solution. To ensure that the comprehensive scheduling value assessment model is trainable, reproducible, and intrinsically consistent with manufacturing site data, the training process of the comprehensive scheduling value assessment model adopts a hierarchical structure of an offline training subsystem and an online inference subsystem. The offline training subsystem is sequentially connected to the data extraction module, sample construction module, feature construction module, label generation module, model training module, model evaluation module, and model release module. The online inference subsystem is sequentially connected to the real-time data access module, feature synchronization module, model inference module, and ranking output module. The model version and feature configuration released by the offline training subsystem are synchronized to the online inference subsystem to ensure consistency between the training and deployment standards.

[0090] The offline training subsystem's data extraction module extracts multi-source data required for training from the MES database and related manufacturing systems and performs primary key association. The training data includes: order master data, process route data, equipment and resource ledgers, line changeover and switching records, production execution records, quality and rework records, and historical scheduling results and actual performance indicators. The order master data includes order number, product number, quantity, delivery date, priority, customer level, and work order release time. The process route data includes the number of process segments, process sequence, key process identifiers, cycle time, and standards. The data extraction module cleans and aligns missing values, outliers, and cross-system coding inconsistencies, and outputs a standardized order sample master table, resource status table, and execution result table, providing a consistent data foundation for subsequent sample construction. The equipment and resource ledger includes equipment type, processable product families, production line capacity, available shifts, and maintenance windows. The line changeover and switching records include line changeover requirements such as mold change, cleaning, and first-piece confirmation, historical line changeover duration, and line changeover failures or rework events. The production execution records include start and finish times, work-in-process dwell time, waiting reasons, and abnormal downtime. The quality and rework records include inspection results, defect types, number of reworks, and scrap ratio. The actual achievement indicators include delivery date achievement, switching costs, work-in-process turnover, and rework risk results.

[0091] The sample construction module builds training samples based on rolling scheduling snapshots at multiple time points to ensure that the comprehensive scheduling value assessment model learns the decision-making patterns of scheduling scenarios evolving over time. Specifically, each scheduling trigger moment is taken as a sample time point. At each sample time point, the candidate order set and the current production line resource status are extracted to form a data package at the sample time point level. For each candidate order within the sample time point, an order-level training sample is generated, and for each compatibility merging group within the sample time point, a merging group-level training sample is generated. The order-level training samples and merging group-level training samples share the same feature caliber but have different aggregation granularities. The sample construction module divides the samples into training, validation, and test sets according to time, preferably using the most recent time window as the test set to conform to the objective law of the manufacturing site distribution drifting over time.

[0092] The feature construction module constructs structured features strongly relevant to the field of this invention for each training sample. These features include order-side features, process-side features, resource-side features, changeover-side features, and quality risk-side features, and clearly defines the inherent relationships between inputs and outputs. Order-side features include delivery urgency features, order size features, and business priority features. Delivery urgency features include, for example, remaining delivery time and the ratio of remaining time to standard working hours; order size features include, for example, quantity and estimated processing time; business priority features include, for example, customer level and expedited marking. Process-side features include equipment type requirement features, number of process segments, process sequence structure features, and product production condition set features. Resource-side features include available capacity features of the target equipment, current work-in-process load features, and bottleneck resource occupancy features. Available capacity features include, for example, available shifts and maintenance window distance; current work-in-process load features include, for example, queue length and estimated queue time. Changeover-side features include, for example, the size of the changeover requirement set, the number of differences between the changeover requirement set and the current equipment status, historical average changeover time, and estimated first-piece confirmation time. Quality risk features include historical defect rate, rework rate, number of times key condition items exceed tolerances, and recent quality fluctuation index for the same product family. The feature construction module outputs feature vectors and simultaneously outputs a feature dictionary and feature normalization configuration. The feature dictionary is used to ensure consistency between online and offline fields, and the feature normalization configuration is used to ensure consistency in numerical standards between the training and inference periods.

[0093] The label generation module provides calculable target values ​​for supervised learning training and maintains consistency with the multi-objective scheduling evaluation dimensions of this invention. Based on historical scheduling and actual execution results, the label generation module generates a comprehensive scheduling value label for each sample. This comprehensive scheduling value label consists of four sub-labels: delivery date contribution, changeover cost contribution, turnover contribution, and rework risk penalty, all using a standardized scoring caliber with unified dimensions. Delivery date contribution is mapped to a score based on the order's historical delivery date achievement and delay level; lower delays result in higher scores. Changeover cost contribution is mapped to a score based on the number of line changes, changeover duration, or changeover cost at the sample time point; lower changeover costs result in higher scores. Turnover contribution is mapped to a score based on work-in-process dwell time and waiting time; faster turnover results in higher scores. Rework risk penalty is mapped to a deduction score based on the probability of rework, number of reworks, or defect rate; higher rework risk results in larger deductions. Each sub-label undergoes interval truncation and normalization before mapping to avoid unstable impacts on training caused by abnormal samples. The comprehensive scheduling value label is obtained by fusing the four types of sub-labels according to preset weights. The preset weights are configured by the business strategy and are consistent with the scheduling objectives, so that the value ranking output by the comprehensive scheduling value assessment model can directly serve the scheduling optimization objectives of this invention. To enhance trainability, the label generation module can also output relative preference labels within the sample time point. That is, based on historical scheduling results or simulation evaluation results, orders within the same time point are paired to form superior-inferior pairs for training the ranking model, enabling the model to learn the relative relationships of which orders should be prioritized for scheduling under the same on-site conditions.

[0094] The model training module performs training based on the aforementioned sample features and labels. The preferred model type is a gradient boosting tree ranking model or regression model, which can handle mixed discrete and continuous features and is easy to interpret. Alternatively, a neural network model such as a multilayer perceptron can also be used. The model type is replaceable within this field, but the training approach remains consistent. Taking the ranking model as an example, the model training module uses pairs of orders with varying quality within the same scheduling trigger point as training samples, and optimizes using pairwise ranking loss. The model output is a comprehensive scheduling value score for the orders, which is used for ranking. Taking the regression model as an example, the model training module uses the comprehensive scheduling value label as the regression target, and optimizes using mean squared error or robust regression loss. The model output is a predicted value score. During training, the learning rate, maximum number of iterations, model complexity, and regularization parameters are set. The validation set is used for early stopping and hyperparameter selection. Hyperparameters can be determined using grid search or Bayesian optimization. Simultaneously, to address the class imbalance problem in manufacturing site data, the model training module reweights or stratifies delayed samples, abnormal downtime samples, or rework samples to improve the model's ability to identify key risk scenarios. After training, the model training module outputs the model file, model parameters, feature dictionary version, and training data time range, and writes them to the model repository to support version management and backtracking.

[0095] The model evaluation module performs offline evaluation of the model on the test set. Evaluation metrics include ranking consistency metrics and business replay metrics. Ranking consistency metrics measure the degree of consistency between the model's output ranking and historical priorities or the best simulation ranking. Business replay metrics input the model's ranking results into the scheduling simulation or historical replay engine to compare the improvement in key indicators such as delivery date achievement rate, changeover costs, work-in-process turnover, and rework risk. The model evaluation module outputs an evaluation report and threshold suggestions. For example, release is allowed when ranking consistency reaches a preset threshold and business indicator improvement meets preset requirements; otherwise, it reverts to the previous version or triggers adjustments to feature and label definitions.

[0096] The model release module publishes the evaluated model version along with the feature dictionary, normalization configuration, and weighting strategy to the online inference subsystem. At each scheduling trigger, the online inference subsystem accesses the real-time order set and real-time resource status, generates feature vectors using the same feature construction criteria as offline, and calls the model inference module to output a comprehensive scheduling value score for each order or each compatibility merging group. The ranking output module generates a comprehensive scheduling value priority sequence based on the scores from high to low, and outputs this priority sequence to the compatibility merging group construction and multi-objective evolutionary scheduling solution stages, thus achieving an inherent closed-loop correlation between the model output and the scheduling process of this invention. To ensure interpretability and maintainability, the online inference subsystem can simultaneously output key feature contribution information affecting the score, used to display the ranking reasons in the MES interface and assist schedulers in auditing.

[0097] In one optional implementation, to adapt to the differences between different factories, production lines, and product families, the training method supports domain-specific training and deployment based on factory, production line, equipment type, or product family dimensions. This means constructing samples and training models within each domain separately, while retaining the shared feature dictionary backbone and domain-specific feature extensions. This achieves strong binding between the model and the application scenario of this invention, and allows for transferable deployment. Simultaneously, the training method supports periodic incremental training, with the incremental training cycle configurable weekly or monthly. Incremental training updates labels and feature statistics with the most recently added execution data, and performs canary releases after offline evaluation gating. This continuously improves the accuracy and stability of the comprehensive scheduling value assessment without altering the main workflow of this invention.

[0098] For example, to facilitate those skilled in the art to reproduce the training process of the comprehensive scheduling value assessment model, a set of practical training parameter settings can be provided. This example uses the set of candidate orders within the rolling schedule trigger point as the sorting training unit, employs a gradient boosting tree sorting model for training, and selects the most recent 90 days of MES historical data for the training data time window, with the first 70 days as the training set, the last 10 days as the validation set, and the last 10 days as the test set. Samples are grouped according to the scheduling trigger point, with each scheduling trigger point constituting a group, containing all candidate order samples at that point. To ensure consistency between the labels and the multi-objective scheduling evaluation dimensions of this invention, the comprehensive scheduling value label is obtained by fusing four parts: delivery date contribution, switching cost contribution, work-in-process turnover contribution, and rework risk penalty. The fusion weights can be set as follows: delivery date contribution 0.40, switching cost contribution 0.25, work-in-process turnover contribution 0.20, and rework risk penalty 0.15. Before fusion, each sub-label is standardized from 0 to 100 and subjected to upper and lower bound truncation. For example, samples with delivery delays exceeding 72 hours are uniformly treated as having the lowest score, samples with line change times exceeding 120 minutes are uniformly treated as having the lowest score, and samples with work-in-process stays exceeding 48 hours are uniformly treated as having the lowest score, in order to avoid the impact of extreme values ​​on training instability.

[0099] In this example, the training parameters for the ranking model can be set as follows: 800 trees, maximum depth of 8, learning rate of 0.05, maximum number of leaves per tree of 64, minimum number of leaf samples of 50, feature sampling ratio of 0.80, sample sampling ratio of 0.80, and L2 regularization coefficient of 1.0. An early stopping strategy is adopted, with the number of early stopping rounds set to 50. That is, when the ranking metric on the validation set shows no improvement within 50 consecutive iterations, training is stopped and rolled back to the optimal round. The ranking optimization objective can be set to NDCG@10, which evaluates the ranking quality of the top 10 in each group at each scheduling trigger point, and uses the improvement of NDCG@10 on the validation set as a gating condition for hyperparameter selection and model release. For example, when the NDCG@10 on the validation set is not lower than 0.72 and the business replay metric meets the condition of "delivery achievement rate improvement of not less than 2 percentage points or total line change time reduction of not less than 3%", the model version is deemed to meet the release conditions.

[0100] In this example, to address the training bias caused by the relative scarcity of delayed or reworked samples, a sample weighting strategy can be set: samples with delivery delays are assigned a weight of 2.0, samples with rework are assigned a weight of 1.5, and normal samples maintain a weight of 1.0. An upper limit is set on the number of samples within each group; for example, a maximum of 200 candidate order samples are extracted for training at each scheduling trigger point. When this limit is exceeded, stratified sampling is performed according to business priority to ensure training efficiency and stable group distribution. The above parameter example enables reproducible training under typical data scales in MES production planning and scheduling, and ensures that the comprehensive scheduling value score output by the model maintains consistent data association and process flow with the order priority generation, compatibility merging group construction, and subsequent multi-objective evolutionary scheduling solution stages of this invention.

[0101] It should be noted that by employing the aforementioned order preprocessing module and the comprehensive scheduling value priority sequence generation method, the problems of scattered production orders, unscientific batch allocation, unclear scheduling priorities, and lack of quantitative basis for scheduling decisions in multi-variety, small-batch production scenarios in existing technologies can be effectively solved. This embodiment first performs data cleaning, legality verification, and normalization on the received N production order data to ensure the accuracy and consistency of the data used, eliminating the impact of information noise or outliers on subsequent analysis from the source.

[0102] Furthermore, by extracting the process route identifiers corresponding to production orders and grouping orders with the same process route identifiers into the same process route group, the interference of process differences on batch merging can be reduced while ensuring the consistency of the process flow, thus achieving integrated configuration of production processes. Compared to relying solely on order delivery dates or manual experience for batch allocation, grouping by process route identifiers in advance has higher process adaptability and production stability, which helps to facilitate the smooth execution of subsequent automated scheduling.

[0103] Based on the process route grouping, predefined process compatibility judgment rules are applied to the production orders within each group. Orders with high similarity are further divided into M compatibility merging groups to ensure that each compatibility merging group has a high degree of consistency in terms of line change requirements, specific production conditions and key process parameters, which significantly reduces the risk of frequent line change, production fluctuations or quality inconsistencies caused by process deviations.

[0104] By conducting a comprehensive scheduling value assessment based on the production order data corresponding to the compatibility merging groups, this method can comprehensively evaluate each merging group across multiple dimensions, including delivery urgency, customer value, production feasibility, and potential quality risks, thereby generating a scientific and objective comprehensive scheduling value score. Compared to traditional methods that prioritize orders based solely on delivery date or a single dimension, this approach more comprehensively reflects the overall value and urgency of production orders, improving scheduling rationality.

[0105] Finally, by sorting the M compatibility merging groups in descending order according to their comprehensive scheduling value scores, a comprehensive scheduling value priority sequence is generated. This enables the production management system to quickly select the highest priority production batch based on the sorting results during subsequent production batch generation and production scheduling. This ensures that production resources are prioritized for high-value and high-urgency orders, improves production response speed and customer satisfaction, and reduces delivery delays or resource waste caused by unreasonable scheduling order.

[0106] The scheduling optimization module, based on the comprehensive scheduling value priority sequence, sequentially obtains each compatibility merging group and constructs a population of scheduling schemes that includes batch production sequence, process changeover plan, production start-up time plan and resource allocation strategy. It uses delivery fulfillment rate, changeover cost, work-in-process turnover time and rework risk index to perform multi-objective evolutionary fitness calculation, non-dominated sorting and congestion distance evaluation. It iteratively generates the optimal scheduling scheme through binary tournament selection, crossover and mutation operations.

[0107] like Figure 4 As shown, the method for generating the optimal scheduling scheme includes:

[0108] S200: Let the initial value of m be 1, and the range of m is from 1 to M;

[0109] S201: Obtain the m-th compatibility merging group from the comprehensive scheduling value priority sequence;

[0110] S202: Initialize the scheduling scheme population, where each individual in the scheduling scheme population represents a scheduling scheme corresponding to the m-th compatibility merging group. The scheduling scheme includes batch production sequence, process changeover plan, production start-up time plan, and resource allocation strategy. The size of the scheduling scheme population can be flexibly set according to the number of batches. For example, in this application, the size of the scheduling scheme population can be set to 20 to 100 individuals.

[0111] S203: Based on the production order data corresponding to the m-th compatibility merging group, construct the objective functions of delivery fulfillment rate, switching cost, work-in-process turnover time, and rework risk index, and calculate the multi-objective evolutionary fitness of each scheduling scheme in the scheduling scheme population;

[0112] Based on multi-objective evolutionary fitness, a non-dominated sorting is performed on the population of the scheduling scheme to obtain a Q-layer Pareto sorting structure.

[0113] Calculate the comprehensive congestion distance index for the scheduling scheme corresponding to the Pareto optimal frontier of each layer in the Q-layer Pareto sorting structure.

[0114] A binary tournament selection mechanism prioritizing comprehensive congestion distance is adopted to select the parent scheduling scheme. The offspring scheduling scheme is generated through mapping crossover and random mutation operations, expanding the solution space coverage. The crossover and mutation probabilities can be dynamically configured according to business needs; for example, the crossover probability can be set to 0.9 and the mutation probability to 0.2.

[0115] The parent scheduling scheme and the child scheduling scheme are merged into a new scheduling scheme population; if the current iteration count is less than the preset maximum number of iterations, then S203 is continued; if the current iteration count is not less than the preset maximum number of iterations, then the scheduling scheme with the largest comprehensive congestion distance index is selected as the optimal scheduling scheme in the first-level Pareto optimal frontier.

[0116] S204: Let m = m + 1. If m is less than or equal to M, then return to S201 for execution. If m is greater than M, then end the current process.

[0117] like Figure 6 As shown, the method for calculating the multi-objective evolutionary fitness includes:

[0118] The delivery time satisfaction rate is calculated based on the constructed objective function for delivery time satisfaction rate;

[0119] The switching cost is calculated based on the constructed switching cost objective function;

[0120] The work-in-process turnover time is calculated based on the constructed objective function for work-in-process turnover time.

[0121] The rework risk index is calculated based on the constructed objective function of the rework risk index.

[0122] Multi-objective evolutionary fitness is calculated based on delivery fulfillment rate, changeover cost, work-in-process turnover time, and rework risk index.

[0123] The objective function for the delivery date satisfaction rate is defined as follows:

[0124] ;

[0125] in, To achieve on-time delivery rate, This indicates the number of orders whose expected completion time is earlier than or equal to the target delivery date under the scheduling plan. This represents the total number of orders involved in the scheduling plan. The delivery date fulfillment rate objective function aims to maximize the proportion of orders whose delivery dates are met by the scheduling plan. The delivery date fulfillment rate objective function measures the overall effectiveness of the generated scheduling plan in meeting the delivery date requirements of production orders, aiming to maximize on-time delivery. Specifically, the system compares the estimated completion time of the orders included in each candidate production batch with the target delivery date of the orders, counts the number of orders that meet the delivery date requirements, and calculates the delivery date fulfillment rate. When performing multi-objective optimization analysis and generating scheduling plans, the delivery date fulfillment rate should be maximized.

[0126] The switching cost objective function is defined as follows:

[0127] ;

[0128] in, To reduce switching costs, This represents the switching cost between batch pl and batch pl+1. The switching cost objective function is used to evaluate the manpower, time, and material costs required for production switchover, line changeover, tooling replacement, and equipment adjustment in the scheduling scheme, aiming to minimize the overall burden of process switchover. This represents the total batch size. The switching cost between different batches can be obtained by a person skilled in the art through comprehensive evaluation of switching multiple product process parameter sets, and a switching cost database can be pre-built for retrieval. When performing multi-objective optimization analysis and generating scheduling schemes, the switching cost should be minimized. The batch size refers to the total number of candidate production batches scheduled for execution in a production scheduling scheme, i.e., the number of batch sets formed by the system based on order preprocessing, process compatibility merging, capacity allocation, and priority sorting, which will be put into production sequentially within a specified scheduling period.

[0129] The objective function for work-in-process turnover time is defined as follows:

[0130] ;

[0131] in, Work-in-process turnover time This is the estimated start time for batch pl. The estimated end time of batch p1 is used to quantify the production cycle and work-in-process (WIP) dwell time of all batches under the scheduling scheme. The objective function for WIP turnaround time aims to reduce production cycle time and inventory holding. When performing multi-objective optimization analysis and generating scheduling schemes, WIP turnaround time should be minimized.

[0132] The rework risk index objective function is used to evaluate the potential rework risk of the scheduling scheme in terms of quality reliability. The aim is to prioritize batches with lower overall quality risk in the scheduling scheme, and to optimize the rework risk index as much as possible during multi-objective optimization analysis and scheduling scheme generation. The rework risk index uses multiple production batches included in the scheduling scheme as statistical objects, weights the rework risk level of each batch according to the corresponding production quantity, and normalizes it using the total production quantity. This ensures comparability between scheduling schemes with different batch sizes, and the rework risk index value remains within the range of 0 to 1.

[0133] When calculating the rework risk index, the number of batches included in the scheduling plan is first determined, and a rework risk weighting coefficient and the production quantity of each production batch are assigned. The rework risk weighting coefficient characterizes the relative risk level of rework for that batch, and its value is limited to 0 to 1, with a higher value indicating a higher rework risk. The rework risk weighting coefficient can be set by those skilled in the art based on information such as quality inspection records, rework records, product type, equipment type, fulfillment of changeover requirements, and deviations from the product production condition set. The production quantity characterizes the output scale of that batch within the scheduling plan, reflecting the business principle that larger output contributes more to the overall rework risk.

[0134] In the derivation and calculation of the rework risk index, a batch-weighted risk value is calculated for each production batch. This batch-weighted risk value is the product of the rework risk weight coefficient for that batch and the production quantity of that batch. Then, the batch-weighted risk values ​​of all batches within the scheduling plan are summed to obtain the total weighted risk. Simultaneously, the production quantities of all batches within the scheduling plan are summed to obtain the total production quantity. Finally, the rework risk index is obtained by dividing the total weighted risk by the total production quantity. This calculation method, which weights the risk by production quantity and normalizes it to the total production quantity, avoids the bias caused by simply averaging based on batch quantity or risk weight alone. It allows the rework risk index to simultaneously reflect both the risk level and the scale of risk exposure, thus providing a consistent evaluation caliber for scheduling plans with different batch combinations and production scales.

[0135] In an optional implementation, when the total production quantity of the scheduling scheme is 0, the rework risk index can be set to a preset default value to avoid division by zero errors. The preset default value can be set to 0 or configured by business rules. When the rework risk weight coefficient of a certain batch included in the scheduling scheme is missing, it can be filled by historical statistical values ​​within the same product type, equipment type, or production condition set range of the same product, or the missing batch can be marked as high risk and assigned a larger rework risk weight coefficient to ensure the robustness of the evaluation process. Through the definition and calculation method of the above-mentioned rework risk index objective function, those skilled in the art can conduct an achievable, reproducible, and comparable quantitative assessment of the quality rework risk of candidate scheduling schemes in the MES production planning and scheduling scenario, and minimize the rework risk index as one of the quality reliability optimization objectives in the multi-objective optimization solution.

[0136] It should be noted that, in the example of this invention, the rework risk weighting coefficient is used to characterize the rework risk level of a production batch in terms of quality reliability. The value range is limited to 0 to 1, with a higher value indicating a higher rework risk. To facilitate implementation by those skilled in the art, the rework risk weighting coefficient can be set using a combination of rule mapping and data support: First, extract criteria items strongly correlated with rework risk from quality and execution data, including the rework rate or defect rate of the corresponding product in the most recent statistical window, the stability of the equipment type involved in the batch, the fulfillment of the changeover requirement set, and the deviation of the product production condition set. Equipment type stability includes, for example, recent fluctuations in the start-up yield of the same equipment type; fulfillment of the changeover requirement set includes, for example, whether key items such as mold change, cleaning, and first article confirmation are complete; deviation of the product production condition set includes, for example, whether there are items exceeding the allowable deviation for key conditions such as temperature, pressure, and time.

[0137] Then, the above criteria are mapped to tiered or ranged values ​​for the rework risk weight coefficient. For example: when the recent statistical window rework rate is no higher than 0.5% and all line change requirements are met and there are no out-of-tolerance critical production conditions, the rework risk weight coefficient can be set to 0.10; when the recent statistical window rework rate is between 0.5% and 2% or there is no more than one out-of-tolerance critical production condition but the line change requirements are met, the rework risk weight coefficient can be set to 0.30; when the recent statistical window rework rate is between 2% and 5% or there are two out-of-tolerance critical production conditions or there are missing non-critical line change items, the rework risk weight coefficient can be set to 0.60; when the recent statistical window rework rate is higher than 5% or there are missing critical line change items or three or more out-of-tolerance critical production conditions, the rework risk weight coefficient can be set to 0.85 or higher to reflect the penalty intensity of high-risk batches in the scheduling plan quality objectives.

[0138] To illustrate how the rework risk weighting coefficient is set, example data from four production batches are provided.

[0139] Batch 1 corresponds to product type 1 and equipment type A, with a rework rate of 0.3% in the last 30 days. The set of line change requirements for batch 1 includes mold change requirements, cleaning requirements, and first article confirmation requirements, and all of the line change requirements are met. There are no records of the key conditions in the product production condition set of batch 1 exceeding the allowable deviation. Therefore, the rework risk weight coefficient for batch 1 is set to 0.10.

[0140] Batch 2 corresponds to product type 1 and equipment type A, with a rework rate of 1.2% in the last 30 days. The set of changeover requirements for batch 2 includes mold change requirements, cleaning requirements, and first article confirmation requirements, and all of the changeover requirements are met. There is one record of a critical condition item in the product production condition set of batch 2 exceeding the allowable deviation, such as a temperature-related condition item exceeding the tolerance once. Therefore, the rework risk weight coefficient for batch 2 is set to 0.30.

[0141] Batch 3 corresponds to product type 2 and equipment type B, with a rework rate of 3.8% in the last 30 days. The set of changeover requirements for batch 3 includes mold change requirements, cleaning requirements, first article confirmation requirements, and optional secondary cleaning requirements. The set of changeover requirements is satisfied only if there are missing items, specifically the secondary cleaning requirement. The set of production conditions for batch 3 has two records of exceeding the allowable deviation for key conditions, such as one temperature condition and one time condition. Therefore, the rework risk weighting coefficient for batch 3 is set to 0.60.

[0142] Batch 4 corresponds to product type 3 and equipment type B, with a rework rate of 6.5% in the last 30 days. The set of changeover requirements for batch 4 includes mold change requirements, cleaning requirements, and first article confirmation requirements. The set of changeover requirements is satisfied only if there are missing items, and the missing item is the first article confirmation requirement. The set of production conditions for batch 4 has records of 3 or more key conditions exceeding the allowable deviation. Therefore, the rework risk weight coefficient for batch 4 is set to 0.90.

[0143] As can be seen from the above exemplary data, the rework risk weight coefficient can comprehensively reflect the impact of historical rework performance, the fulfillment of the set of line change requirements, and the deviation of the set of product production conditions on rework risk. It can also be directly used for batch-level weighted calculation of the rework risk index objective function, thereby supporting the optimization and evaluation of the scheduling scheme in the dimension of quality reliability.

[0144] Methods for calculating multi-objective evolutionary fitness based on delivery fulfillment rate, changeover cost, work-in-process turnover time, and rework risk index include:

[0145] ;

[0146] in, For multi-objective evolutionary fitness, it is used to comprehensively evaluate candidate scheduling schemes and as input for non-dominated sorting, crowding distance calculation and selection operations; , , , These represent the delivery time satisfaction rate, respectively. Switching costs Work-in-process turnover time and repair risk index The evaluation values, after normalization and homogenization, are all limited to the range of 0 to 1, with higher values ​​indicating better scheduling schemes. The evaluation value of delivery fulfillment rate is obtained by forward normalization of the original value; the evaluation values ​​of changeover cost, work-in-process turnaround time, and rework risk index are obtained by inverse normalization or decreasing mapping of the corresponding original values, so that the lower the changeover cost, the shorter the turnaround time, and the lower the rework risk, the higher the evaluation value. The weighting coefficients are all non-negative and sum to 1, reflecting the relative importance of each evaluation indicator in the current production scenario.

[0147] , , and For the corresponding weighting coefficients, , , and The sum is 1. Weighting coefficients reflect the importance and priority of different optimization objectives in production scheduling decisions. These coefficients can be configured by those skilled in the art based on business strategies, order structures, production line bottlenecks, and quality risk conditions, and can be dynamically adjusted according to the triggering point of rolling scheduling. For example, when there is an emergency delivery, a concentration of customized customer orders, or an increased risk of delivery default, the weighting coefficient corresponding to the delivery date fulfillment rate indicator is increased; when there are frequent line changes, high equipment switching costs, or sensitive energy consumption, the weighting coefficient corresponding to the switching cost indicator is increased; when work-in-process inventory backlog, workshop congestion, or production line turnover efficiency become the main constraints, the weighting coefficient corresponding to the work-in-process turnover time indicator is increased; when quality requirements are strict, rework costs are high, or recent quality fluctuations are significant, the weighting coefficient corresponding to the rework risk index is increased. Through the above weighting constraint and configuration mechanism, without changing the multi-objective evolutionary solution framework, the fitness function can form an interpretable strategy preference for different production scenarios, and guide the population to converge in a direction that conforms to the current business objectives during evolutionary iteration.

[0148] For example, three typical scenarios are given for weighting coefficient configuration. In the delivery-priority mode, suitable for urgent delivery scenarios or scenarios with concentrated customer-customized orders, the weighting coefficient for the delivery fulfillment rate indicator can be set to 0.50, the weighting coefficient for the switching cost indicator to 0.20, the weighting coefficient for the work-in-process turnover time indicator to 0.15, and the weighting coefficient for the rework risk index to 0.15, to emphasize prioritizing on-time delivery. In the cost control mode, suitable for scenarios with high line changeover costs, frequent equipment changesovers, or the need to reduce line changeover expenses, the weighting coefficient for the delivery fulfillment rate indicator can be set to 0.20, the weighting coefficient for the switching cost indicator to 0.50, the weighting coefficient for the work-in-process turnover time indicator to 0.15, and the weighting coefficient for the rework risk index to 0.15, to emphasize suppressing switching costs. In the quality reliability mode, which is suitable for precision manufacturing scenarios with high quality requirements, high rework costs, or recent increases in quality risks, the weight coefficient of the delivery date fulfillment rate index can be set to 0.20, the weight coefficient of the switching cost index can be set to 0.20, the weight coefficient of the work-in-process turnover time index can be set to 0.10, and the weight coefficient of the rework risk index can be set to 0.50 to highlight the control of potential rework risks.

[0149] The above weighting coefficients are merely examples. Those skilled in the art can adjust the weighting coefficients according to the optimization needs of actual production scenarios, and ensure that the weighting coefficients meet the non-negativity constraint and the constraint that the sum is 1, so as to obtain a multi-objective evolutionary fitness evaluation result consistent with the business objective.

[0150] Methods for obtaining the Q-layer Pareto sort structure include:

[0151] S300: Let the initial value of q be 1, and the range of q is from 1 to Q;

[0152] S301: For all scheduling schemes in the scheduling scheme population, for any pair of scheduling schemes, let them be scheduling scheme one and scheduling scheme two;

[0153] If scheduling scheme 1 has better objective function values ​​than scheduling scheme 2 in all four dimensions of delivery date fulfillment rate, switching cost, work-in-process turnover time and rework risk index, and the multi-objective evolutionary fitness of scheduling scheme 1 is greater than that of scheduling scheme 2, then scheduling scheme 1 is determined to dominate scheduling scheme 2.

[0154] The number of times each scheduling scheme is dominated by other scheduling schemes is recorded as the number of dominated schemes. The scheduling scheme with a number of dominated schemes of 0 is recorded as the Pareto optimal frontier of the qth level.

[0155] The qth Pareto optimal front is taken as the qth layer of the Q-layer Pareto sorting structure;

[0156] Remove the scheduling scheme corresponding to the Pareto optimal front of the q-th layer from the scheduling scheme population;

[0157] S302: Let q = q + 1. If q is less than Q, return to S301 to continue execution. If q is greater than Q, end the current process.

[0158] Methods for calculating the comprehensive congestion distance index for the scheduling scheme corresponding to the Pareto optimal front in each layer of the Q-level Pareto ranking structure include:

[0159] S400: Let the initial value of q be 1, and the range of q is from 1 to Q;

[0160] S401: Obtain the q-th Pareto optimal frontier from the Q-th Pareto sorting structure; denot the number of scheduling schemes corresponding to the q-th Pareto optimal frontier as... ;

[0161] according to The multi-objective evolutionary fitness corresponding to each scheduling scheme is sorted in ascending order to obtain the Pareto optimal front sequence;

[0162] For the Pareto optimal frontier sequence, the first... For each scheduling scheme, the corresponding congestion distance increment is calculated based on the numerical difference of the objective function of adjacent scheduling schemes. The comprehensive congestion distance index is obtained by summing the congestion distance increments of all adjacent scheduling schemes.

[0163] S402: Let q = q + 1. If q is less than or equal to Q, return to S401 to continue execution. If q is greater than Q, end the current process.

[0164] The calculation method for the comprehensive congestion distance index includes:

[0165] The congestion distance increment is used to measure the relative interval between adjacent scheduling schemes along a certain objective function dimension. It is used to evaluate the sparsity of scheduling schemes within the same Pareto layer, thereby prioritizing the retention of more dispersed and comprehensive scheduling schemes during the evolutionary selection phase. The objective function dimensions include four dimensions: delivery time fulfillment rate objective function, switching cost objective function, work-in-process turnover time objective function, and rework risk index objective function.

[0166] When calculating the congestion distance increment, for the set of scheduling schemes within the same Pareto level, one objective function dimension is first selected. Then, all scheduling schemes within the Pareto level are sorted according to their objective function values ​​under this objective function dimension, resulting in an ordered sequence of scheduling schemes. The k-th and (k+1)-th adjacent scheduling schemes form a pair of adjacent scheduling schemes. Subsequently, the maximum and minimum values ​​of the objective function within the Pareto level are calculated under this objective function dimension to obtain the range of values ​​for this objective function dimension.

[0167] For adjacent scheduling scheme pairs, the congestion distance increment under the objective function dimension is determined as follows: the objective function value of the (k+1)th scheduling scheme under this objective function dimension is subtracted from the objective function value of the kth scheduling scheme under this objective function dimension, and the difference is taken as the adjacent difference; then, the adjacent difference is divided by the range of values ​​to achieve normalization, thereby obtaining the congestion distance increment under this objective function dimension. This normalization process avoids the impact of differences in the dimensions of different objective functions on the congestion distance calculation and makes the increments of different objective function dimensions additive and comparable. The index k of the adjacent scheduling scheme pair ranges from 1 to "the number of scheduling schemes in this Pareto layer minus 1".

[0168] When calculating the comprehensive congestion distance index for scheduling schemes corresponding to each Pareto optimal front, the Pareto optimal front sequence is first obtained by sorting the Pareto optimal fronts in ascending order of multi-objective evolutionary fitness. The k-th and (k+1)-th adjacent scheduling schemes in the sequence are identified as a pair of adjacent scheduling schemes. For the four objective function dimensions—delivery fulfillment rate, switching cost, work-in-process turnover time, and rework risk index—the function value difference of adjacent scheduling schemes in the corresponding objective function dimension is calculated. The difference between the maximum and minimum function values ​​of this objective function dimension within the current Pareto optimal front is then normalized to obtain the congestion distance increment under the objective function dimension. For each scheduling scheme in the Pareto optimal front sequence, the congestion distance increments corresponding to the adjacent scheduling scheme pairs formed by the adjacent scheduling schemes are accumulated under each objective function dimension. The accumulated results of the four objective function dimensions are then summed to obtain the comprehensive congestion distance index corresponding to the scheduling scheme.

[0169] It should be noted that the comprehensive crowding distance index reflects the relative distance between the scheduling scheme and its neighboring solutions in the multi-objective space. The larger the comprehensive crowding distance index, the sparser the distribution of solutions around the corresponding scheduling scheme, the stronger the coverage, and the higher the necessity of retaining it. When performing selection operations within the same Pareto layer, scheduling schemes with larger comprehensive crowding distances can be given priority to improve population diversity and avoid excessive concentration of solutions in local areas.

[0170] It should be further explained that, in this application, the scheduling optimization module is mainly used in the production order scheduling stage to perform group optimization scheduling on compatibility merging groups that have passed the comprehensive scheduling value priority screening. The core function of the scheduling optimization module is to achieve global equilibrium evaluation and optimal solution search of the scheduling scheme based on four multi-dimensional objective functions: delivery fulfillment rate, changeover cost, work-in-process turnover time, and rework risk index, thereby providing a multi-objective collaborative optimal scheduling decision basis for production scheduling.

[0171] Specifically, the scheduling optimization module acquires each compatibility merging group in the comprehensive scheduling value priority sequence in turn, constructs a population of scheduling schemes that includes batch production sequence, process changeover plan, production time plan and resource allocation strategy, and performs multi-objective evolutionary fitness quantification calculation on each scheduling scheme based on four objective functions. It uses non-dominated sorting to determine the multi-dimensional superiority and inferiority of the schemes, and combines the crowding distance index to evaluate the sparsity of the solution set distribution. Then, it continuously generates new solutions through binary tournament selection, mapping crossover and random mutation operations, thereby achieving multi-generation evolutionary optimization of the scheduling scheme population.

[0172] The scheduling optimization module plays a specific role in the technical solution of this application. Specifically, through a multi-objective evolutionary process, it can meet delivery constraints while reducing changeover costs, shortening work-in-process turnover cycles, and reducing rework risks, thereby achieving global balance and flexibility in production scheduling. In addition, by performing optimization separately within each compatibility merging group, it can fully utilize the ranking basis of the comprehensive scheduling value priority sequence to ensure that the scheduling schemes of each batch achieve a reasonable balance between global ranking and local optima.

[0173] The scheduling scheme verification module is used to verify the optimal scheduling scheme output by the scheduling optimization module in real time, evaluate the feasibility, stability and performance indicators of the optimal scheduling scheme under actual production conditions, and generate a set of schemes that pass verification and a set of schemes that fail verification.

[0174] like Figure 3 As shown, the methods for generating the set of valid solutions and the set of invalid solutions include:

[0175] S500: Let the initial value of m be 1, and the range of m is from 1 to M;

[0176] S501: Obtain the m-th optimal scheduling scheme, obtain the production order data corresponding to the compatibility merging group of the m-th optimal scheduling scheme, and construct a production order data set.

[0177] Initialize the feasibility flag, stability flag, and performance indicator flag to No;

[0178] The optimal scheduling scheme and the production order data set are input into the scheduling scheme evaluation model to obtain the scheduling scheme evaluation result set; the scheduling scheme evaluation result set includes feasibility score, stability score and performance index score.

[0179] If the feasibility score is greater than the preset feasibility score threshold, the feasibility flag will be set to yes.

[0180] If the stability score is greater than the preset stability score threshold, the stability flag will be set to "yes".

[0181] If the performance indicator score is greater than the preset performance indicator score threshold, then the performance indicator flag is set to "yes".

[0182] Determine whether the feasibility flag, stability flag, and performance indicator flag are all "yes". If the result is "yes", add the optimal scheduling scheme to the set of verified schemes. If the result is "no", add the optimal scheduling scheme to the set of verified schemes.

[0183] S502: Let m = m + 1. If m is less than or equal to M, then return to S501 for execution. If m is greater than M, then end the current process.

[0184] It should be noted that, in the embodiments of this application, the feasibility score, stability score, and performance index score used in the scheduling scheme verification module are all used to verify the execution capability of the optimal scheduling scheme under actual production conditions from multiple dimensions. The feasibility score, stability score, and performance index score are all expressed using a standardized percentage system, and their values ​​range from [missing value]. A higher score indicates a higher degree of suitability or achievement of the corresponding indicator.

[0185] Specifically, the feasibility score mainly reflects the feasibility of the selected scheduling scheme under given production resources, process conditions and production windows. The stability score is used to evaluate the ability of the scheduling scheme to maintain consistency under dynamic conditions such as production load fluctuations and equipment utilization fluctuations. The performance index score is used to quantify the overall performance of comprehensive production efficiency, delivery timeliness and production line resource utilization.

[0186] During the scoring process, the system uses preset threshold standards to ensure consistency and objectivity in the evaluation criteria when comparing and selecting multiple solutions. Examples of threshold settings include: setting the feasibility score to 80 points (a feasibility score greater than 80 indicates feasibility is met); setting the stability score to 75 points (a stability score greater than 75 indicates stability is met); and setting the performance index score to 85 points (a performance index score greater than 85 indicates performance is met). These threshold standards can be flexibly adjusted based on different production line scales, process complexity, and enterprise production scheduling strategies to ensure adaptability and relevance in various production environments.

[0187] The training method for the scheduling scheme evaluation model includes:

[0188] The scheduling scheme evaluation model is used to quickly evaluate and screen candidate scheduling schemes generated by multi-objective evolution. The evaluation model and the scheduling scheme verification process are interconnected: after the multi-objective evolutionary algorithm outputs candidate scheduling schemes, these schemes are input into the evaluation model to obtain evaluation results. Based on the evaluation results, a candidate scheduling scheme is determined to be included in either the verified scheme set or the verified scheme set. The verified scheme set is used to generate scheduling instructions, while the verified scheme set is fed back to trigger subsequent re-optimization iterations. The output of the scheduling scheme evaluation model includes three types of scores: feasibility score, stability score, and performance index score. These three scores use a unified scoring dimension and are compared with preset thresholds. These scores also provide interpretive evidence for schedulers to support rolling scheduling decisions on the manufacturing floor.

[0189] In the embodiments of this invention, the scheduling scheme evaluation model adopts a hierarchical sub-model structure to ensure feasibility and maintainability. The scheduling scheme evaluation model includes a feasibility evaluation sub-model, a stability evaluation sub-model, and a performance evaluation sub-model. Each sub-model shares the same set of input features and outputs a corresponding score value. The feasibility evaluation sub-model is used to characterize the degree to which candidate scheduling schemes meet hard constraints, including constraints on equipment type matching, number of process segments and sequence of operations, set of set of changeover requirements, set of set of product production conditions, and resource calendar constraints. The stability evaluation sub-model is used to characterize the robustness of candidate scheduling schemes in the presence of disturbances, including order insertion, equipment stoppage, changeover time fluctuations, and processing time fluctuations. The performance evaluation sub-model is used to characterize the comprehensive performance of candidate scheduling schemes in terms of delivery time fulfillment, changeover cost, work-in-process turnover, and quality reliability, and maintains an intrinsic correlation with the delivery time fulfillment rate, changeover cost, work-in-process turnover time, and rework risk index used in the multi-objective evolution of this invention.

[0190] The input data for the scheduling scheme evaluation model consists of structured scheduling scheme data and production site status data at the scheduling trigger time, ensuring a strong binding between the evaluation results and the MES production planning and scheduling scenario. The structured scheduling scheme data includes batch production sequence, process changeover plan, production launch time plan, and resource allocation strategy. The batch production sequence describes the processing order of compatible merge groups or order batches on the production line; the process changeover plan describes the set of line changeover requirements and the sequence of changeover actions between adjacent batches; the production launch time plan describes the planned start and finish times for each batch; and the resource allocation strategy describes the allocation results of equipment types and specific equipment, workstations, shifts, and other resources. Production site status data includes the available time window for candidate equipment, maintenance window, current work-in-process load, current line changeover status, critical material arrival status, and quality status information. The output data of the scheduling scheme evaluation model includes feasibility score, stability score, performance index score, and an optional comprehensive judgment indicator. The comprehensive judgment indicator indicates whether the candidate scheduling scheme meets the verification pass conditions and is determined by comparing the three types of score values ​​with a preset threshold.

[0191] Training samples are organized according to the rolling scheduling trigger points to ensure consistency between the training samples and the online inference scenario. The data extraction module extracts historical scheduling trigger points, the candidate order set at that time, the production site status at that time, the actual scheduling schemes used, and execution result data from the MES and related manufacturing systems, and aligns the cross-system codes to form time-point level sample packages. The sample generation module constructs a scheduling scheme sample set within each time-point level sample package. The scheduling scheme sample set includes the actual issued and executed scheduling scheme samples, and further includes candidate scheduling scheme samples that have been replaced, rolled back, or deemed unusable at the same time point. When the system does not explicitly record the replaced scheme, the actual scheme is locally perturbed using a method consistent with multi-objective evolutionary coding to generate control scheme samples. The control scheme samples are used to supplement negative samples and improve the model's ability to identify infeasible and unstable schemes. The training set, validation set, and test set are constructed using a time-segmented approach. The training set uses an earlier time window, the validation set uses an intermediate time window, and the test set uses the most recent time window to adapt to the objective law of manufacturing site data drifting over time.

[0192] A unified feature vector is constructed based on structured data of the scheduling scheme and production site status data. This unified feature vector includes constraint satisfaction features, disturbance sensitivity features, and performance contribution features. Constraint satisfaction features support the training of the feasibility assessment sub-model and include equipment type matching markers, process segment consistency markers, process sequence consistency markers, changeover requirement set satisfaction markers, product production condition set deviation counts, resource time window conflict counts, and critical material shortage risk markers. Disturbance sensitivity features support the training of the stability assessment sub-model and include time margin statistics for each batch, bottleneck equipment load rate, critical changeover duration percentage, constraint batch number, and planned path dependency strength index. The time margin statistics are jointly calculated from the remaining time between the planned completion time and delivery date, and the remaining time relative to the resource availability window at the planned start time. Performance contribution features are used to support the training of the performance evaluation sub-model. These features include estimates of planned delivery date achievement rate, planned changeover cost, planned work-in-process turnover time, and planned rework risk index. These estimates are consistent with the objective function in the multi-objective evolution of this invention, ensuring that the output of the performance evaluation sub-model is semantically aligned with the multi-objective optimization indicators. The feature dictionary and normalization configuration are then constructed to ensure consistency between online and offline fields, and the normalization configuration ensures consistency in numerical processing methods.

[0193] Training labels are generated for the three types of sub-models and uniformly mapped to a scoring space of 0 to 100 for direct comparison with the validation threshold. Feasibility score labels are generated based on the degree of hard constraint violation. When the scheduling scheme has no hard constraint violations, the feasibility score label is 100. When hard constraint violations occur, the feasibility score label is generated according to a deduction rule. The deduction rule calculates the violation ratio for each violation, such as equipment type mismatch, resource time window conflict, unmet set of line switching requirements, and out-of-tolerance key conditions of the product production condition set, and accumulates the deductions according to a preset constraint violation deduction weight coefficient. The constraint violation deduction weight coefficient describes the relative influence of different hard constraint violation types in the feasibility score deduction calculation. The constraint violation deduction weight coefficient is only used for the deduction summary in the feasibility score label generation stage. The violation ratio is the ratio of the violation amount to the allowable amount, truncated to 0 to 1. The deduction result is subtracted from 100 to obtain the feasibility score label, thus the higher the degree of violation, the lower the score.

[0194] The stability score label is generated from the deviation between the plan and the actual situation and the frequency of reordering. The stability score label is based on the historical execution records to statistically analyze the deviation between the planned start and the actual start, the deviation between the planned completion and the actual completion, the number of reorderings and the magnitude of reorderings caused by order insertion or anomalies, and maps the deviation and the magnitude of reordering to deduction values. The deduction value is deducted from 100 to obtain the stability score label, so that the greater the deviation and the more frequent the reordering, the lower the score.

[0195] The performance index scoring label is jointly generated by the actual delivery date fulfillment rate, changeover cost, work-in-process turnover time, and rework risk index. The performance index scoring label first normalizes the four types of results into a common evaluation value according to a predetermined standard and limits it to the range of 0 to 1. Then, it is fused into a comprehensive performance evaluation value according to the preset performance target fusion weight coefficient. The comprehensive performance evaluation value is then mapped to 0 to 100 to obtain the performance index scoring label, so that the solution with better delivery date, lower changeover time, faster turnover, and lower rework risk can obtain a higher performance index scoring label.

[0196] It should be noted that, for example, the feasibility score is out of 100. Weighting coefficients for hard constraint violations are set for each type, and the sum of all such weighting coefficients is limited to 1. This is used to map the violation proportions of different violation types to a uniform penalty value and perform a weighted summation. For example, hard constraint violation types include equipment type mismatch, resource time window conflict, unmet set of line change requirements, and out-of-tolerance key conditions in the product production condition set. Weighting coefficients for these violations are set to 0.35, 0.30, 0.20, and 0.15 respectively, with the sum of these coefficients being 1. If a candidate scheduling scheme has the following violations: equipment type mismatch (0.20%), resource time window conflict (0.10%), failure to meet line switching requirements (0.30%), and exceeding the tolerance of key conditions in the product production condition set (0.00%), then the overall violation level is 0.35×0.20+0.30×0.10+0.20×0.30+0.15×0.00=0.16. The corresponding deduction can be set as the overall violation level multiplied by 100, i.e., a deduction of 16 points. The feasibility score label is 100 minus 16, which equals 84 points.

[0197] For example, performance indicator scoring labels are generated based on delivery time fulfillment rate, changeover cost, work-in-process turnover time, and rework risk index. To ensure consistency in the fusion, changeover cost, work-in-process turnover time, and rework risk index are first mapped in descending order to be converted into unidirectional evaluation values, ensuring that all four evaluation values ​​are limited to the range of 0 to 1, with higher values ​​indicating better performance. Then, a performance target fusion weight coefficient is set and its sum is limited to 1. The four unidirectional evaluation values ​​are then weighted and fused to obtain a comprehensive performance evaluation value. For example, in a delivery-priority scenario, the performance target fusion weight coefficient can be set to: delivery time fulfillment rate 0.50, changeover cost 0.20, work-in-process turnover time 0.15, and rework risk index 0.15, with the sum of these weight coefficients being 1. If a candidate scheduling scheme has four evaluation values ​​after homogenization: delivery fulfillment rate 0.90, switching cost 0.70, work-in-process turnaround time 0.60, and rework risk index 0.80, then the comprehensive performance evaluation value is 0.50×0.90+0.20×0.70+0.15×0.60+0.15×0.80=0.80. The comprehensive performance evaluation value is then converted into a performance index scoring label from 0 to 100 according to a preset mapping. For example, multiplying it by 100 gives 80 points, which are used to compare with the performance index scoring threshold and participate in the verification and judgment.

[0198] Supervised learning is employed to train the feasibility assessment, stability assessment, and performance assessment sub-models, with a unified training configuration ensuring model reproducibility. Each sub-model utilizes a gradient boosting tree regression model, employing robust regression loss to mitigate the impact of outliers. Early stopping is used during training, and the optimal number of iterations is selected based on the validation set scoring error. Example training hyperparameters include a learning rate of 0.05, a maximum number of iterations of 800, a maximum tree depth of 8, a minimum number of leaf samples of 50, a feature sampling ratio of 0.80, a sample sampling ratio of 0.80, a regularization coefficient of 1.0, and 50 early stopping iterations. The training data time window is the most recent 180 days, divided into a 120-day training set, a 30-day validation set, and a 30-day test set. An example sample weighting strategy assigns higher weights to samples experiencing delivery delays or rework events to enhance the model's ability to identify high-risk scenarios. Upon completion of training, the model version number, feature dictionary version number, and normalization configuration version number are output and stored in the model repository to ensure consistency during deployment.

[0199] The three types of sub-models, feature dictionaries, and normalized configurations are synchronized to the online evaluation service. At each scheduling trigger, the online evaluation service accesses the structured data of candidate scheduling schemes and the current production site status data, generates feature vectors according to offline consistency standards, and outputs three types of score values. The verification and judgment module performs verification and judgment based on the three types of score values ​​and preset thresholds. Examples of thresholds are: feasibility score not less than 80, stability score not less than 75, and performance index score not less than 85. When all three types of score values ​​simultaneously meet the threshold conditions, the candidate scheduling scheme is added to the set of verified successful schemes; otherwise, it is added to the set of failed successful schemes and subsequent optimization iterations are triggered. Through the above training method and input / output settings, training samples can be constructed based on real data from MES production planning and scheduling, the scheduling scheme evaluation model can be trained, and online inference can be completed, thereby realizing a feasible closed loop in the scheduling scheme verification and selection process of this invention.

[0200] It should be noted that, in this application, the scheduling scheme verification module is used to systematically and quantitatively verify and screen the optimal scheduling scheme output by the scheduling optimization module. The scheduling scheme verification module inputs the scheduling scheme and its corresponding production order data into the scheduling scheme evaluation model, and judges the scheme based on three core evaluation dimensions: feasibility score, stability score, and performance index score, according to preset thresholds. This achieves a comprehensive evaluation of the scheduling scheme's executability, capacity matching, and quality risks under actual production conditions.

[0201] By performing multi-dimensional verification on each optimal scheduling scheme, the system ensures that only scheduling schemes that meet the requirements in terms of delivery date, process switching, resource utilization, and production cycle time are identified as verified schemes, effectively reducing the probability of potential production anomalies or execution failures. In addition, by automatically classifying scheduling schemes that do not meet the threshold conditions into the set of verified unverified schemes, the system can automatically isolate low-quality scheduling schemes, providing a clear basis for subsequent re-optimization or callback mechanisms.

[0202] Regarding the integration with subsequent operations, the verified solution set will be directly used to generate production scheduling execution instructions and resource allocation instructions, serving as the basis for production plan issuance and execution monitoring. The verified solution set will trigger an automatic callback mechanism, allowing the scheduling optimization module to perform a new round of evolutionary optimization and scheduling recalculation, achieving a closed-loop improvement in scheduling quality. This significantly enhances the reliability, adaptability, and stability of the production scheduling solution, providing technical support for the integration of multi-objective evolutionary optimization and full-process verification in this application.

[0203] If the set of failed verification schemes is not empty, the instruction generation and issuance module will use the set of failed verification schemes as the priority sequence of comprehensive scheduling value and return it to the scheduling optimization module for execution; if the set of failed verification schemes is empty, it will generate scheduling instructions based on the set of successful verification schemes and execute them.

[0204] Methods based on verification that generate and execute scheduling instructions from a set of schemes include:

[0205] For each optimal scheduling scheme in the verified scheme set, the corresponding batch production sequence, process changeover plan, production launch time plan, and resource allocation strategy are extracted and structured based on a predefined production scheduling instruction template to form a scheduling execution instruction file that can be recognized by the Manufacturing Execution System (MES).

[0206] The scheduling execution instruction file is automatically sent to the manufacturing execution system and related execution units for execution.

[0207] It should be noted that in this application, the instruction generation and issuance module is used to collect, judge, and distribute the verification results output by the scheduling scheme verification module, and convert the verified optimal scheduling scheme into structured scheduling execution instructions to realize the implementation of the production plan. The instruction generation and issuance module first determines whether the set of schemes that fail verification is empty. If there are scheduling schemes that fail verification, the set of schemes that fail verification is used as a new comprehensive scheduling value priority sequence, and automatically calls back to the multi-objective evolutionary scheduling optimization module to re-execute the optimization calculation, realizing closed-loop iteration and dynamic correction of scheduling quality.

[0208] If the set of failed verification solutions is empty, it means that all optimal scheduling solutions have met the feasibility, stability, and performance constraints. The Manufacturing Execution System (MES) then automatically enters the scheduling instruction generation process. For each optimal scheduling solution in the verified solution set, the instruction generation and issuance module extracts the corresponding batch production sequence, process changeover plan, production launch time plan, and resource allocation strategy. These are then standardized and encapsulated using a predefined production scheduling instruction template to form a scheduling execution instruction file that conforms to the MES interface specification.

[0209] Through the instruction generation and distribution module, the production scheduling optimization results are efficiently and accurately transformed into control instructions for the production execution layer, achieving a closed-loop connection from optimized scheduling to actual production. The role of the instruction generation and distribution module in this solution is reflected in ensuring the accuracy and structural consistency of scheduling instructions, while also automatically triggering re-optimization processes for solutions that fail verification, thereby improving the adaptability of production scheduling. The output of the instruction generation and distribution module provides clear and executable instructions for production execution, ensuring a seamless connection between multi-objective evolutionary optimization, feasibility verification, and production plan distribution, supporting this application in achieving an efficient, intelligent, and closed-loop production scheduling optimization technology solution.

[0210] Example 2:

[0211] Please see Figure 2As shown, this embodiment provides a production planning and scheduling method for an MES system, including:

[0212] Step 1: Analyze, identify and classify the received N production order data to generate a comprehensive scheduling value priority sequence;

[0213] Step 2: Based on the comprehensive scheduling value priority sequence, obtain the compatibility merging group in sequence, and construct a population of scheduling schemes. Calculate the multi-objective evolutionary fitness based on delivery satisfaction rate, switching cost, work-in-process turnover time, and rework risk index. Perform non-dominated sorting, congestion distance assessment, selection, crossover, and mutation operations to generate the optimal scheduling scheme.

[0214] Step 3: Verify the optimal scheduling scheme, and generate a set of schemes that pass verification and a set of schemes that fail verification based on feasibility, stability and performance indicators;

[0215] Step 4: If the set of failed verification schemes is not empty, then the set of failed verification schemes will be used as the priority sequence of comprehensive scheduling value, and the process will return to Step 2 for execution; if the set of failed verification schemes is empty, then a scheduling instruction will be generated based on the set of successful verification schemes and executed.

[0216] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

[0217] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A production planning and scheduling method for a MES system, characterized in that, include: Step 1: Analyze, identify and classify the received N production order data to generate a comprehensive scheduling value priority sequence; Step 2: Based on the comprehensive scheduling value priority sequence, obtain the compatibility merging group in sequence, and construct a population of scheduling schemes. Calculate the multi-objective evolutionary fitness based on delivery satisfaction rate, switching cost, work-in-process turnover time, and rework risk index. Perform non-dominated sorting, congestion distance assessment, selection, crossover, and mutation operations to generate the optimal scheduling scheme. Step 3: Verify the optimal scheduling scheme, and generate a set of schemes that pass verification and a set of schemes that fail verification based on feasibility, stability and performance indicators; Step 4: If the set of failed verification schemes is not empty, then the set of failed verification schemes will be used as the priority sequence of comprehensive scheduling value, and the process will return to Step 2 for execution; if the set of failed verification schemes is empty, then a scheduling instruction will be generated based on the set of successful verification schemes and executed.

2. The production planning and scheduling method for a MES system according to claim 1, characterized in that, The methods for generating the set of valid solutions and the set of invalid solutions include: For each optimal scheduling scheme, obtain the corresponding production order data and construct a production order data set; initialize the feasibility flag, stability flag, and performance indicator flag to no; input the optimal scheduling scheme and the production order data set into the scheduling scheme evaluation model to obtain an evaluation result set including feasibility score, stability score, and performance indicator score; The feasibility score, stability score, and performance index score are compared with their respective preset thresholds. If the score is greater than the threshold, the corresponding flag is set to "yes". The feasibility flag, stability flag, and performance index flag are all "yes". If all three are "yes", the optimal scheduling scheme is added to the set of verified schemes. If any flag is "no", the optimal scheduling scheme is added to the set of verified schemes that fail.

3. The production planning and scheduling method for a MES system according to claim 1, characterized in that, The method for generating the optimal scheduling scheme includes: Obtain each compatibility merging group from the comprehensive scheduling value priority sequence. For each compatibility merging group, initialize a population of scheduling schemes. Each scheduling scheme includes batch production sequence, process changeover plan, production launch time plan, and resource allocation strategy. Based on the production order data corresponding to the compatibility merging group, construct the objective function of delivery fulfillment rate, objective function of changeover cost, objective function of work-in-process turnover time, and objective function of rework risk index. Calculate the multi-objective evolutionary fitness of each scheduling scheme. A non-dominated sorting process is performed on the scheduling scheme population to obtain a Q-level Pareto sorting structure. The comprehensive congestion distance index of the scheduling scheme corresponding to the Pareto optimal front at each level is calculated. Based on the comprehensive congestion distance index, a binary tournament selection is performed, and mapping crossover and random mutation are performed to generate offspring scheduling schemes. The parent and offspring scheduling schemes are merged to form a new scheduling scheme population. After iterating to the preset maximum number of iterations, the scheduling scheme with the largest comprehensive congestion distance index in the first-level Pareto optimal front is selected as the optimal scheduling scheme.

4. The production planning and scheduling method for a MES system according to claim 1, characterized in that, The method for calculating the multi-objective evolutionary fitness includes: The delivery time satisfaction rate is calculated based on the constructed objective function for delivery time satisfaction rate; The switching cost is calculated based on the constructed switching cost objective function; The work-in-process turnover time is calculated based on the constructed objective function for work-in-process turnover time. The rework risk index is calculated based on the constructed objective function of the rework risk index. Multi-objective evolutionary fitness is calculated based on delivery fulfillment rate, changeover cost, work-in-process turnover time, and rework risk index.

5. A production planning and scheduling method for an MES system according to claim 3, characterized in that, Methods for obtaining the Q-layer Pareto sort structure include: S300: Let the initial value of q be 1, and the range of q is from 1 to Q; S301: For all scheduling schemes in the scheduling scheme population, for any pair of scheduling schemes, let them be scheduling scheme one and scheduling scheme two; If scheduling scheme 1 has better objective function values ​​than scheduling scheme 2 in all four dimensions of delivery date fulfillment rate, switching cost, work-in-process turnover time and rework risk index, and the multi-objective evolutionary fitness of scheduling scheme 1 is greater than that of scheduling scheme 2, then scheduling scheme 1 is determined to dominate scheduling scheme 2. The number of times each scheduling scheme is dominated by other scheduling schemes is recorded as the number of dominated schemes. The scheduling scheme with a number of dominated schemes of 0 is recorded as the Pareto optimal frontier of the qth level. The qth Pareto optimal front is taken as the qth layer of the Q-layer Pareto sorting structure; Remove the scheduling scheme corresponding to the Pareto optimal front of the q-th layer from the scheduling scheme population; S302: Let q = q + 1. If q is less than Q, return to S301 to continue execution. If q is greater than Q, end the current process.

6. A production planning and scheduling method for an MES system according to claim 3, characterized in that, The calculation methods for the comprehensive congestion distance index of the scheduling scheme corresponding to each Pareto optimal front include: The Pareto optimal fronts of each layer are obtained sequentially from the Pareto sorting structure. The number of scheduling schemes in each Pareto optimal front is determined, and the scheduling schemes are sorted in ascending order according to their multi-objective evolutionary fitness to form a Pareto optimal front sequence. For each scheduling scheme in the Pareto optimal front sequence, the congestion distance increment is calculated based on the difference in objective function values ​​between adjacent scheduling schemes. The congestion distance increments in each objective function dimension are accumulated to obtain the corresponding comprehensive congestion distance index.

7. A production planning and scheduling method for an MES system according to claim 1, characterized in that, The method for obtaining the comprehensive scheduling value priority sequence includes: Perform data preprocessing on the received N production order data; Extract the process route identifier corresponding to each production order data, and merge production orders with the same process route identifier into the same process route group to form Y process route groups; For each process route group, execute predefined process compatibility judgment rules on the production orders to form M compatibility merging groups; Based on the production order data corresponding to the M compatibility merging groups, a comprehensive scheduling value assessment is performed to obtain the comprehensive scheduling value score corresponding to the M compatibility merging groups; The M compatibility merging groups are sorted in descending order according to their corresponding comprehensive scheduling value scores to form a comprehensive scheduling value priority sequence.

8. A production planning and scheduling method for an MES system according to claim 7, characterized in that, The methods for forming M compatibility merge groups include: Obtain the set of product process parameters for each production order in each process route group, summarize the set of product process parameters for the same group, and construct a grouped process parameter set; For each set of process parameters in a group, the process parameters of any two production orders are compared item by item to calculate the process parameter similarity. If the process parameter similarity is higher than the preset process parameter similarity threshold, they are marked as compatible order pairs and a compatibility tag is generated. It is then determined whether the compatibility tag matches the compatibility tag of an existing compatibility merging group. If they match, the compatible order pair is added to the corresponding compatibility merging group. If they do not match, a new compatibility merging group is created and a compatibility tag is assigned, ultimately forming M compatibility merging groups.

9. A production planning and scheduling method for a MES system according to claim 1, characterized in that, Methods based on verification that generate and execute scheduling instructions from a set of schemes include: For each optimal scheduling scheme in the verified scheme set, the corresponding batch production sequence, process changeover plan, production launch time plan, and resource allocation strategy are extracted, and structured encapsulation is performed based on a predefined production scheduling instruction template to form a scheduling execution instruction file. The scheduling execution instruction file is automatically sent to the manufacturing execution system and execution units for execution.

10. A production planning and scheduling system for a MES system, used to implement the production planning and scheduling method for an MES system according to any one of claims 1-9, characterized in that, include: The order preprocessing module is used to systematically parse, identify compatibility, automatically classify and merge the received N production order data to form a standardized comprehensive scheduling value priority sequence. The scheduling optimization module, based on the comprehensive scheduling value priority sequence, sequentially obtains each compatibility merging group and constructs a population of scheduling schemes that includes batch production sequence, process changeover plan, production time plan and resource allocation strategy. It uses delivery satisfaction rate, changeover cost, work-in-process turnover time and rework risk index to perform multi-objective evolutionary fitness calculation, non-dominated sorting and congestion distance evaluation. It iteratively generates the optimal scheduling scheme through binary tournament selection, crossover and mutation operations. The scheduling scheme verification module is used to verify the optimal scheduling scheme output by the scheduling optimization module in real time, evaluate the feasibility, stability and performance indicators of the optimal scheduling scheme under actual production conditions, and generate a set of schemes that pass verification and a set of schemes that fail verification. If the set of failed verification solutions is not empty, the instruction generation and issuance module will use the set of failed verification solutions as the priority sequence of comprehensive scheduling value and return it to the scheduling optimization module for execution. If the set of solutions that fail verification is empty, then a scheduling instruction is generated and executed based on the set of solutions that pass verification.

Images