A steel order multi-line distribution optimization method and system based on dynamic clustering

Abstract

The present application belongs to the field of steel production planning and scheduling optimization, and relates to a steel order multi-line distribution optimization method and system based on dynamic clustering. The method obtains steel production order data and line capacity data to be distributed within a rolling time window, establishes a line capacity constraint model, and determines the processable domain of the order at each processing link and the overall processable domain in combination with a process rule library; under the constraint of the frozen area, dynamic clustering, incremental updating and re-clustering trigger judgment are performed on the current set of orders that can be optimized, and the clustering threshold, feature weight and clustering distance parameters are adaptively adjusted based on the order cluster structure change and order feature difference; further, an initial distribution scheme is constructed based on the cluster-level processable domain of the order cluster and the candidate line matching rule, and the multi-line distribution result is obtained through infeasible repair and iterative optimization. The present application can improve the feasibility, stability and line load balancing of order distribution in the scenario of multiple processing links, multiple candidate lines and dynamic order changes.

Description

Technical Field

[0001] This invention relates to the field of steel production planning and scheduling optimization technology, and in particular to a method and system for optimizing the allocation of steel orders across multiple production lines based on dynamic clustering. Background Technology

[0002] As the steel industry moves towards high-end and customized products, market orders are characterized by diverse varieties, small batches, tight delivery schedules, and frequent arrivals. Steel production typically involves multiple processing stages, including steelmaking, continuous casting, hot rolling, cold rolling, pickling, annealing, galvanizing, leveling, shearing, and post-processing. Each processing stage has multiple candidate production lines or equipment units. Different orders vary in specifications, steel grades or materials, product categories, process paths, quality levels, surface treatment requirements, and delivery dates. Therefore, it is necessary to rationally allocate orders to candidate production lines within the corresponding processing stages, while meeting production line capacity constraints, process rule constraints, and order delivery requirements, to form an executable multi-production line production allocation plan.

[0003] Existing methods for allocating steel orders across multiple production lines typically employ manual experience rules, fixed-threshold grouping strategies, or centralized solution methods based on mathematical programming. While manual experience rules can enable rapid decision-making based on actual on-site conditions, they are easily influenced by human experience when dealing with large order sizes, frequent changes in order structure, or significant differences in production line capabilities across multiple processing stages, resulting in poor stability and consistency of allocation results. Fixed-threshold grouping or static clustering methods usually rely on pre-set clustering parameters, making it difficult to adapt to changes in the concentration or dispersion characteristics of order distribution over time, easily leading to unstable clustering results, large differences within order clusters, or infeasibility of subsequent production line allocation. Centralized optimization methods tend to suffer from long solution times and difficulty in responding promptly to rolling plan update requirements when there are many constraints, large order sizes, and frequent changes in production line capacity.

[0004] Furthermore, in the actual production organization of multi-variety, small-batch orders, orders arrive continuously and may be canceled, changed, have delivery dates adjusted, or have changes in order attributes. If all orders are re-clustered and re-solved for allocation schemes every time the order set changes, it will not only increase computational overhead but also easily cause frequent fluctuations in the plan, affecting the stability of on-site execution. To maintain the executability of the production plan, a freeze zone strategy is usually introduced in production management. This strategy keeps the allocated results unchanged for orders that have been issued for execution, started, or are close to execution, and only performs rolling optimization on currently optimizable orders outside the freeze zone.

[0005] Therefore, it is necessary to propose an optimization method and system that can adapt to the rolling changes in steel orders and improve the stability and solution efficiency of multi-production line allocation schemes, so as to realize the rapid and feasible allocation of multi-variety, small-batch steel orders among multiple production lines and support the implementation of production plans. Summary of the Invention

[0006] To address the problems of poor allocation stability, large amount of redundant calculation, reliance on manual setting of clustering parameters, and easy occurrence of infeasibility in existing steel order multi-production line allocation methods under scenarios such as rolling order arrival, changes in order attributes, and changes in production line status, this invention provides a steel order multi-production line allocation optimization method and system based on dynamic clustering.

[0007] To achieve the above-mentioned technical objectives, the present invention provides the following technical solution: A method for optimizing the allocation of steel orders across multiple production lines based on dynamic clustering includes: S1: Obtain steel production order data and production line capacity data to be allocated, construct order feature vectors based on the steel production order data, and establish a production line capacity constraint model; the production line capacity constraint model is used to characterize the processability relationship and capacity carrying capacity relationship between orders and corresponding candidate production lines in each processing stage; S2: Determine the rolling time window according to the preset window length and rolling step size, and determine the order set within the current rolling time window based on the order's entry into the system status, delivery date requirements, and production lead time; set a freeze zone within the rolling time window to keep the orders within the freeze zone unchanged by the predetermined production line allocation, thereby determining the current set of orders that can be optimized; S3: Within the rolling time window, based on the production line capacity constraint model and the process rule base, determine the processable domain of each order in the current optimizable order set under each processing stage, and form the overall processable domain of the order; S4: Within the current rolling time window, orders with similar characteristics and the same or compatible processing domains inherit the clustering results from the previous time window, and use the clustering results as the initial state of the order clusters within the current time window. S5: When the current set of optimizable orders changes, for the changed orders, perform incremental attribution, removal or re-attribution processing based on the updated order feature vector and processable domain, update the order cluster and the corresponding cluster center, and calculate the cluster center offset index at the same time. S6: Determine whether to trigger re-clustering based on the cluster center offset index and clustering threshold parameter, and perform adaptive updates on clustering-related parameters; S7: Under the constraints of the frozen area, determine the cluster-level processable domain based on the current order cluster and the order processable domain, and construct an initial allocation scheme according to the matching rules between the order cluster and the candidate production line; When the initial allocation scheme becomes infeasible, under the constraints of the order-level processable domain and the cluster-level processable domain, the initial allocation scheme is adjusted according to the infeasibility repair rules to restore allocation feasibility and obtain a feasible allocation scheme; allocation optimization processing is performed on the feasible allocation scheme, and the multi-production line allocation result is output.

[0008] Further, in step S1, the production line capacity constraint model is used to determine the processing steps involved in the order according to the process path of the order, and to determine the corresponding candidate production line set for each processing step; based on the specification compatibility, material compatibility, process compatibility, equipment availability status and capacity carrying capacity between the order and the candidate production lines, the processable domain of the order under each processing step is determined, and the processable domain of the order is formed by the processable domain under each processing step.

[0009] Furthermore, in step S2, the frozen area is determined based on one or more of the following: order plan execution status, whether the order has been issued for execution, whether the order has started, whether the order has entered the scheduled execution window, the order delivery date distance, or the latest start time distance. Orders within the frozen zone will retain their predetermined production line allocation results in this round of calculations and will not participate in this round of dynamic clustering processing and multi-production line order allocation optimization. Orders that have not entered the freeze zone within the rolling time window are identified as the current set of orders that can be optimized. The current set of orders that can be optimized includes one or more of the following: orders newly entering the rolling time window, orders left over from the previous rolling time window that have not entered the freeze zone, and orders whose order attributes have changed and are allowed to be re-optimized.

[0010] Further, in step S3, the processable domain of the order is determined based on the production line capacity constraint model and process rule base established in step S1. Specifically, this includes: determining the processing steps involved in the order according to the process path of the order; for each processing step, screening candidate production lines in the corresponding candidate production line set based on the production line capacity constraint model and process rule base to obtain the processable domain of the order under each processing step; forming the overall processable domain of the order from the processable domain of the order under each processing step, and using the overall processable domain to limit the dynamic clustering boundary and multi-production line allocation range of the order.

[0011] Further, in step S5, during the rolling time window Within, when the current set of orders can be optimized When changes occur, incremental processing is performed only on the orders that have changed, and within the order's processable domain. Update order clusters under constraints This is to avoid performing clustering calculations repeatedly on all orders; The incremental processing includes incremental assignment, removal, or reassignment processing; wherein, cluster assignment processing is performed on newly added orders, removal processing is performed on canceled orders, and reassignment processing is performed on orders whose feature vectors have changed or whose processable domains have changed. After completing the incremental processing, update the corresponding order cluster. and cluster center And based on the rolling time window Compared to the previous scrolling time window The cluster center offset index is calculated based on the distance between the cluster centers of corresponding orders. .

[0012] Further, in step S6, based on the cluster center offset index and clustering threshold parameter Determining whether to trigger re-clustering includes: when When, it indicates a scrolling time window. When the degree of change in the internal order cluster structure exceeds the current clustering stability threshold, re-clustering is triggered; when At the same time, the existing clustering structure remains unchanged, and incremental processing is performed on the basis of the existing clustering structure.

[0013] Furthermore, in step S6, performing adaptive updates to clustering-related parameters includes: based on the difference index Clustering threshold parameter Dynamic updates are performed based on dimensional difference indicators. Weights of feature dimensions Perform adaptive updates and adjust the relative weight coefficients based on the dispersion of numerical features and the dispersion of categorical features. Adaptive adjustment is performed; wherein, the clustering threshold parameter Feature dimension weights and relative weighting coefficients Used to adjust the scrolling time window Clustering granularity and distance metric in the internal order clustering process.

[0014] Furthermore, step S7 specifically includes: (7.1) In a rolling time window Internally, regarding the current set of orders that can be optimized. Each order cluster in Based on the order-processable domain Production line capacity constraint model and process rule base to determine order clusters Cluster-level machinable domains at each processing stage and form order clusters Overall cluster-level machinable domain ; In obtaining cluster-level processable domains Then, the candidate production lines are sorted according to the matching rules between order clusters and candidate production lines to determine the order clusters. The optimal production line sequence is selected at each processing stage, and a target production line is selected from the optimal production line sequence to form an initial allocation scheme; (7.2) After constructing the initial allocation scheme, in the cluster-level processable domain Under constraints, a feasibility check is performed on the initial allocation scheme to determine whether it satisfies the production line capacity constraint model, process rule constraints, and freeze zone constraints. When an infeasible allocation is detected in the initial allocation scheme, an infeasibility check is performed in the order processable domain. and cluster-level cultivable domain Under constraints, the initial allocation scheme is adjusted according to the infeasibility repair rule to restore allocation feasibility; After completing the infeasibility repair process, output a feasible allocation scheme that satisfies the production line capacity constraint model; (7.3) After obtaining a feasible allocation scheme that satisfies the production line capacity constraint model, perform iterative optimization on the feasible allocation scheme under the frozen zone constraint until the allocation result is within the current rolling time window. If no significant improvement occurs or the iteration termination condition is met, the multi-production line allocation result is output.

[0015] Further, in step S7, the infeasibility repair rule includes: when the initial allocation scheme violates one or more of the following constraints: production line capacity, process compatibility, equipment status, delivery date, or adjacent processing link connection, the order's processable domain is used as the basis for correction. and cluster-level cultivable domain To constrain the boundaries, for order clusters that cause infeasible allocation, migration, exchange, splitting, or rollback operations are performed between candidate production lines at the corresponding processing stage to restore allocation feasibility.

[0016] A steel order multi-production line allocation optimization system based on dynamic clustering, the system includes a data acquisition and preprocessing module, a time window and freeze zone management module, a processable domain determination module, a dynamic clustering processing module, and a multi-production line allocation optimization module; The data acquisition and preprocessing module is used to acquire steel production order data and production line capacity data, perform missing value processing, outlier removal, feature encoding and unit unification on the steel production order data, construct order feature vectors, and establish a production line capacity constraint model. The time window and freeze zone management module is used to determine the rolling time window according to the preset window length and rolling step size, and to determine the order set within the current rolling time window based on the order entry status, delivery date requirements and production lead time; to determine the freeze zone within the rolling time window, so that the orders within the freeze zone maintain the predetermined production line allocation results, and to determine the orders outside the freeze zone as the current optimizable order set, for subsequent dynamic clustering processing and multi-production line allocation optimization; The processable domain determination module is used to determine the processable domain of each order in the current optimizable order set under each processing stage based on the production line capacity constraint model and process rule library, and form the overall processable domain of the order from the processable domains under each processing stage, so as to limit the dynamic clustering boundary and multi-production line allocation range of the order. The dynamic clustering processing module inherits the clustering results of the previous rolling time window within the rolling time window to obtain the initial order clusters and cluster centers corresponding to the current optimizable order set. When the current optimizable order set is added, canceled, or the order feature vector changes or the order's processable domain changes, incremental assignment, removal, or reassignment processing is performed only on the orders that have changed, and the order clusters and cluster centers are updated accordingly. At the same time, the cluster center offset index is calculated. The dynamic clustering processing module is also used to determine whether to trigger re-clustering based on the cluster center offset index and the clustering threshold parameter, and to perform adaptive updates on the clustering-related parameters. The multi-production line allocation optimization module is used to determine the cluster-level processable domain of the order cluster based on the current order cluster and the order processable domain under the constraints of the frozen area, and to determine the target production line under each processing stage under the constraints of the cluster-level processable domain according to the matching rules between the order cluster and the candidate production line, so as to construct an initial allocation scheme; when the initial allocation scheme is infeasible, the initial allocation scheme is adjusted according to the infeasibility repair rules to obtain a feasible allocation scheme; after obtaining the feasible allocation scheme, iterative optimization is performed on the feasible allocation scheme within the rolling time window, and the multi-production line allocation result is output.

[0017] The beneficial technical effects of this invention are: (1) The method provided by the present invention can improve the utilization level of resources of multiple production lines and reduce load differences. By determining the processable domain of orders in each processing stage based on the production line capacity constraint model and the process rule library, and allocating multiple production lines under the constraint of the processable domain, the overload or idle situation of production lines can be reduced, and the overall capacity utilization level can be improved.

[0018] (2) The method provided by the present invention manages the order set in collaboration with the rolling time window and the frozen zone, so that the orders that have been issued, started or are about to be executed remain unchanged with the predetermined allocation results, and only the current optimizable order set outside the frozen zone is dynamically optimized, which can reduce the impact of frequent adjustments to the rolling plan on the stability of on-site execution.

[0019] (3) The method provided by the present invention inherits the clustering results of the previous rolling time window and performs incremental attribution, removal or re-attribution processing for orders that are added, canceled, have changed features or have changed processable domains, thereby avoiding full clustering every time the order set changes, thereby reducing computational overhead and improving the rolling plan response efficiency.

[0020] (4) The method provided by the present invention determines whether to trigger re-clustering based on the cluster center offset index, and adaptively updates the clustering parameters according to the order feature dispersion, dimensional difference index and the relative contribution of numerical features and categorical features. This can reduce the dependence on manual parameter setting and improve the stability and adaptability of order cluster division.

[0021] (5) The method provided by the present invention constructs an initial allocation scheme based on the overall processable domain of the order cluster, and restores the feasibility by migration, exchange, splitting or rollback when an infeasible allocation occurs. Then, iterative optimization reduces load imbalance, delivery time risk and scheme disturbance, thereby improving the feasibility, stability and load balance of multi-production line order allocation results. Attached Figure Description

[0022] Figure 1 This is a block diagram of the steel order multi-production line allocation optimization system based on dynamic clustering in an embodiment of the present invention.

[0023] Figure 2 This is a flowchart of the dynamic clustering process in an embodiment of the present invention.

[0024] Figure 3 This is a flowchart of multi-production line allocation and infeasibility repair based on cluster-level processable domains in an embodiment of the present invention. Detailed Implementation

[0025] The technical solution of the present invention will be described in detail below with reference to a preferred embodiment. Those skilled in the art should understand that the following embodiments are only for explaining the present invention and are not intended to limit the scope of protection of the present invention; any improvements or equivalent substitutions to the embodiments without departing from the technical concept of the present invention should fall within the scope of protection of the present invention.

[0026] This embodiment provides a method for optimizing the allocation of steel orders across multiple production lines based on dynamic clustering. The method is executed in the following order: data acquisition and constraint modeling, rolling time window and freeze zone management, determination of the overall processable domain of the order, dynamic clustering incremental update, re-clustering triggering and adaptive adjustment of clustering parameters, determination of the overall processable domain of the order cluster, and optimization of the allocation of multiple production lines.

[0027] Within the current rolling time window, this invention determines the order set based on the order's system entry status, delivery date requirements, and production lead time, and sets a freeze zone to ensure that orders within the freeze zone maintain their predetermined production line allocation results. Orders that do not enter the freeze zone within the rolling time window are identified as the current optimizable order set. Subsequently, based on the aforementioned two-level capability constraint model and process rule base, the processable domains of each order in the current optimizable order set under different processing steps are determined, and the processable domains under each processing step form the overall processable domain of the order, thereby limiting the candidate range for orders to participate in dynamic clustering and multi-production line allocation.

[0028] During order clustering, this invention inherits the clustering results from the previous rolling time window as the initial state of order clusters within the current time window. When the current set of optimizable orders is added, canceled, or its characteristics change, or the overall processable domain of the orders changes, incremental attribution, removal, or re-attribution is performed only on the orders that have changed, and the order clusters and cluster centers are updated accordingly. Furthermore, this invention determines whether to trigger re-clustering based on the cluster center offset index and the clustering threshold parameter, and adaptively updates the clustering threshold, feature dimension weights, and distance metric parameters based on the order feature dispersion, dimensionality difference index, and the relative contribution of numerical and categorical features.

[0029] In the multi-production line allocation process, this invention determines the cluster-level processable domain of an order cluster under each processing step based on the current order cluster and the overall processable domain of the order, and forms the overall processable domain of the order cluster. Under the constraints of the overall processable domain of the order cluster, an initial allocation scheme is constructed according to the matching rules between the order cluster and the candidate production line. When the initial allocation scheme violates capacity constraints, process compatibility constraints, equipment status constraints, delivery date constraints, or adjacent processing step connection constraints, migration, exchange, splitting, or rollback processing is performed according to the infeasible allocation repair rules to restore allocation feasibility. After obtaining a feasible allocation scheme, iterative optimization is performed under the frozen area constraints, and the multi-production line order allocation results are output.

[0030] The method specifically includes: S1: Obtain steel production order data and production line capacity data to be allocated, construct order feature vectors based on the steel production order data, and establish a production line capacity constraint model; the production line capacity constraint model is used to characterize the processability relationship and capacity carrying capacity relationship between orders and corresponding candidate production lines in each processing stage; S2: Determine the rolling time window according to the preset window length and rolling step size, and determine the order set within the current rolling time window based on the order's entry into the system status, delivery date requirements, and production lead time; set a freeze zone within the rolling time window to keep the orders within the freeze zone unchanged by the predetermined production line allocation, thereby determining the current set of orders that can be optimized; S3: Within the rolling time window, based on the production line capacity constraint model and the process rule base, determine the processable domain of each order in the current optimizable order set under each processing stage, and form the overall processable domain of the order; S4: Within the current rolling time window, orders with similar characteristics and the same or compatible processing domains inherit the clustering results from the previous time window, and use the clustering results as the initial state of the order clusters within the current time window. S5: When the current set of optimizable orders changes, for the changed orders, perform incremental attribution, removal or re-attribution processing based on the updated order feature vector and processable domain, update the order cluster and the corresponding cluster center, and calculate the cluster center offset index at the same time. S6: Determine whether to trigger re-clustering based on the cluster center offset index and clustering threshold parameter, and perform adaptive updates on clustering-related parameters; S7: Under the constraints of the frozen area, determine the cluster-level processable domain based on the current order cluster and the order processable domain, and construct an initial allocation scheme according to the matching rules between the order cluster and the candidate production line; When the initial allocation scheme becomes infeasible, under the constraints of the order-level processable domain and the cluster-level processable domain, the initial allocation scheme is adjusted according to the infeasibility repair rules to restore allocation feasibility and obtain a feasible allocation scheme; allocation optimization processing is performed on the feasible allocation scheme, and the multi-production line allocation result is output.

[0031] The following combines a scrolling time window The current optimizations include the order set, the frozen order set, the order cluster and its cluster center. Further explanations will be provided on the specific implementation methods of each step.

[0032] Specifically, in step S1, during the rolling time window... The system acquires steel production order data and production line capacity data to be allocated, constructs order feature vectors based on the steel production order data, and establishes a production line capacity constraint model. The production line capacity constraint model is used to determine the processing stages involved in the order according to the order's process path, and to determine a corresponding set of candidate production lines for each processing stage. Based on the specification compatibility, material compatibility, process compatibility, equipment availability, and capacity carrying capacity between the order and the candidate production lines, the system determines the processable domain of the order under each processing stage, and the processable domains under each processing stage form the overall processable domain of the order.

[0033] Let the set of orders within the current scrolling time window be... The frozen order collection is The current set of orders that can be optimized outside the frozen area is: ,in: The steel production order data includes one or more of the following: order number, product category, product class, process specification, steel grade or material, steel grade class, surface treatment requirements, order specifications, order thickness, order width, quality grade, application information, process path, special process requirements, demand quantity, delivery date, and order priority. The steel production order data undergoes preprocessing, which includes handling missing values, removing outliers, feature encoding, and unifying dimensions. The preprocessed data forms an order feature vector for subsequent clustering processing.

[0034] Preferably, orders The order feature vector is represented in a general form as follows: ; in, This indicates the total number of order feature dimensions. Indicates order In the The feature values ​​are taken on each feature dimension. The order feature dimensions include numerical feature dimensions and categorical feature dimensions; the numerical feature dimensions include one or more of the following: order thickness, order width, order demand quantity, delivery time distance, and order priority quantification value; the categorical feature dimensions include one or more of the following: product category, product class, process specification, steel type or material, steel class, surface treatment requirements, quality grade, application information, process path, special process requirements, and process rule grouping identifier.

[0035] In a preferred embodiment, the order feature vector is used to characterize the similarity of orders at the production organization level. This similarity is determined based on one or more of the following: product category, process specification, steel grade, surface treatment requirements, order thickness, order width, and process rule grouping identifier. The order feature vector does not require the construction of independent clustering target specifications for each processing stage; instead, it is used at the order level for subsequent dynamic clustering processing. Whether an order can be processed by the corresponding production line at each processing stage is constrained by the production line capacity constraint model.

[0036] To establish the production line capacity constraint model, let the set of processing stages involved in the steel order production process be: The processing steps include one or more of the following: steelmaking, continuous casting, hot rolling, cold rolling, pickling, annealing, galvanizing, leveling, shearing, and post-treatment. For any order... The set of processing steps corresponding to the required process path is represented as follows: For any processing stage The set of candidate production lines for this processing stage is represented as follows: in, For processing The number of candidate production lines.

[0037] For orders Processing and candidate production lines Define specification compatibility determination variables respectively. Material compatibility determination variables Process adaptability determination variables and device state variables .in: Based on the above decision variables, define the order. In the processing stage Can the production line be used for this purpose? Processability determination variables : Among them, when When, it indicates an order. In the processing stage Next-line production line Specifications, materials, processes, and equipment status requirements; when When, it indicates an order. In the processing stage The following is unsuitable for production line Processing.

[0038] Furthermore, define For processing Off the production line In the scrolling time window Available capacity within, defined Define for frozen orders or existing plans that already occupy production line load. For orders In the processing stage Distributed to production line The required production capacity at that time. The amount can be determined based on order quantity, processing cycle time, yield rate, equipment efficiency, or unit processing time, for example: in, Indicates order The required quantity or required weight Indicates order In the processing stage The production line The unit capacity consumption coefficient during processing.

[0039] Based on the aforementioned processability determination variables and capacity utilization, orders... In the processing stage The manufacturable domain is defined as follows: Order The overall manufacturable domain is represented as: in, Used to characterize orders The set of processable domains under each processing stage involved in its technological path. If the order... If no processable production line exists at any of the mandatory processing stages, the order will not enter the normal allocation process, but will instead enter the infeasible order marking, manual confirmation, or subsequent rolling time window reprocessing process. The conditions for entering the subsequent dynamic clustering, incremental assignment, re-clustering, and multi-production line allocation processes are as follows: Define order In the processing stage Should it be allocated to the production line? Decision variables : The production line capacity constraint model then includes at least the following constraints: Orders can only be assigned to production lines that meet the manufacturability requirements: The order selects one production line for each of its essential processing stages: The load on each production line within the same processing stage must not exceed its available capacity: When there are process connection requirements between adjacent processing steps, the start time, completion time, minimum interval time or maximum waiting time between adjacent processing steps are further constrained according to the process rule library.

[0040] In step S2 of this embodiment, specifically, the scrolling time window is determined according to the preset window length and scrolling step size. The rolling time window is advanced with the plan update cycle. The preset window length is used to limit the future time range covered by each round of plan calculation, and the rolling step size is used to limit the advancement interval between two adjacent rounds of plan calculation. Within the rolling time window... The system combines the order entry status, delivery date requirements, and production lead time to determine the set of orders within the current rolling time window. The order entry status includes one or more of the following: order created, released, surplus material matching and deduction completed, or in a state where planning is allowed; the delivery date requirement characterizes the delivery time constraint of the order; the production lead time characterizes the time reserved for the order from planning to completion of the corresponding production process. Orders that meet the inclusion criteria of the current rolling time window are included in the order set. .

[0041] In the scrolling time window A freeze zone is set up within the order. The freeze zone is determined based on at least one of the following: order execution status, whether the order has been issued for execution, whether the order has started production, whether the order has entered the scheduled execution window, the order delivery date, or the latest start time. Orders within the freeze zone form a freeze zone order set. Orders within the frozen zone include those already issued for execution, those already started, and those entering the scheduled execution window, as well as orders whose delivery date or latest start time is less than a preset threshold. Orders within the frozen zone will retain their predetermined production line allocation results in this round of calculations and will not participate in this round of dynamic clustering processing and multi-production line order allocation optimization.

[0042] Scrolling time window Orders that have not yet entered the freeze zone are identified as the current set of orders that can be optimized. The current set of optimizable orders. This includes newly entered orders within the rolling time window, orders from the previous rolling time window that have not yet entered the freeze zone, and orders whose attributes have changed and allow for re-optimization. The current set of optimizable orders is then... The input is then fed into the subsequent dynamic clustering and multi-production line allocation optimization process.

[0043] In step S3 of this embodiment, specifically, during the rolling time window... Get the current set of orders that can be optimized. Subsequently, based on the production line capacity constraint model and process rule base established in step S1, the current set of orders that can be optimized is determined. Each order Determine its machinable domain Subsequent dynamic clustering, incremental attribution, and multi-production line allocation all occur within the processable domain. It is carried out under the constraints of [specific rules / regulations]. Specifically, for any order... The set of processing steps required is determined based on its technological path. and in each processing stage Corresponding candidate production line set In the process, based on the processability determination results and capacity carrying capacity conditions defined in step S1, orders are selected. In the processing stage The following manufacturable domain This leads to orders. Overall machinable domain The process rule base includes at least one of the following: process compatibility constraints, specification range constraints, quality grade constraints, equipment status constraints, process path constraints, and special process requirement constraints. These constraints, together with the production line capacity constraint model, are used to define the processable domain of an order, the dynamic clustering boundary, and the multi-production line allocation range. It should be noted that the processable domain is used to limit the range of candidate production lines for order clustering and allocation; it does not imply that independent clustering needs to be performed for each processing stage. Dynamic clustering is still performed at the order level based on order feature vectors to form order clusters with similar production organization.

[0044] In step S4 of this embodiment, within the current scrolling time window Within this timeframe, orders with similar characteristics and the same or compatible processable domains inherit from the previous scrolling time window. The clustering results are used as the initial state of the order clusters within the current rolling time window; specifically: In the machinable domain Under constraints, inherit the previous scrolling time window. The clustering results yield the order clusters corresponding to the previous scrolling time window. and its cluster center and the cluster center corresponding to the previous rolling time window As a rolling time window The initial cluster center set. Preferably, based on the process rule library and the order's processable domain. Orders are categorized and limited to those meeting processing requirements and sharing similar production organization attributes, ensuring they are only included in subsequent similarity calculations and cluster assignments. When performing clustering based on the order feature vectors, a rolling time window is used. The current set of optimizable orders within The orders in the set are compared with the initial cluster center set for similarity calculation, and the initial assignment is completed according to the principle of optimal similarity, thereby obtaining the rolling time window. Initial order cluster within and its cluster center It should be noted that the processable domain constraint is used to limit the candidate range of orders that can participate in the affiliation. Dynamic clustering is still performed at the order level based on the order feature vector, and independent clustering is not performed for each processing stage.

[0045] In step S5 of this embodiment, during the rolling time window... Within, when the current set of orders can be optimized When changes occur, incremental processing is performed only on the orders that have changed, and within the processable domain. Incremental assignment, removal, or reassignment is performed under constraints to avoid repeatedly performing clustering calculations on all orders. The incremental processing refers to partial updates for orders that are added, canceled, or have changed attributes during the rolling time window process. Specifically, it includes incremental assignment, removal, and reassignment. For newly added orders, incremental assignment is performed; for canceled orders, removal is performed; and for orders whose feature vectors or processable domains change, reassignment is performed.

[0046] Specifically, the changes include one or more of the following: adding new orders, canceling existing orders, changes in the order feature vector, and changes in the order's processable domain. For new orders, the similarity between their order feature vector and the centers of each order cluster, combined with their processable domain, is considered. The newly added orders are assigned to order clusters that satisfy the processable domain constraint and have the best similarity. For cancelled orders, they are removed from the original order clusters, and the corresponding order clusters and their cluster centers are updated. For orders whose order feature vectors or processable domains have changed, they are first removed from the original order clusters, and then the clusters are updated based on the updated order feature vectors and processable domains. Perform reassignment processing.

[0047] After completing the incremental processing, update the corresponding order cluster. and cluster center And calculate the cluster center offset index. Used to characterize the rolling time window The degree of change in the internal order clustering structure relative to the previous rolling time window. The cluster center offset index. The distance between the cluster centers of the corresponding order clusters in the current scrolling time window and the previous scrolling time window can be determined by comparing them. The preferred calculation method is explained in step S6 in conjunction with the re-clustering trigger determination. Preferably, within the scrolling time window... During the effective period, when an order is added, canceled, or its feature vector changes, or its processable domain changes, the above processing is triggered and executed based on the updated cluster center offset index. It supports subsequent re-clustering trigger determination and adaptive updating of clustering parameters.

[0048] In step S6 of this embodiment, based on the cluster center offset index and clustering threshold parameter Determine whether to trigger re-clustering and perform adaptive updates to clustering-related parameters. The cluster center offset index... Used to characterize the rolling time window The clustering structure of internal orders relative to the previous rolling time window The degree of change of the clustering threshold parameter This is used to characterize the threshold that allows the existing clustering structure to remain stable within the current rolling time window.

[0049] Specifically, when When the change in the order cluster structure exceeds the current clustering stability threshold, it indicates that the cluster structure in the processable domain has changed. Under constraints on the current rolling time window The current set of optimizable orders within Perform re-clustering to update the order clusters. and cluster center ;when At that time, the existing clustering structure remains unchanged, and the incremental processing in step S5 is continued based on the existing clustering structure.

[0050] To enable the clustering granularity to adaptively adjust as order distribution changes, a rolling time window is used. The degree of dispersion of internal order characteristics is used to construct a difference index. And based on this, the clustering threshold parameter Dynamic updates are performed; furthermore, based on each feature dimension. In the scrolling time window Internal differences construct dimensional difference indicators And based on this, weight the feature dimensions. Perform normalization and smoothing updates; simultaneously, adjust the relative weight coefficients between numerical and categorical features. Adaptive updates are performed to reconcile the relative contributions of numerical and categorical features in the clustering distance metric.

[0051] Preferably, the cluster center offset index It can be determined by the following formula: ; in, The number of order clusters included in the comparison. For order clusters In the scrolling time window The cluster center within, For the corresponding order cluster in the previous scrolling time window The cluster center within, The distance function between cluster centers can be Euclidean distance, weighted Euclidean distance, or a distance function consistent with the order clustering distance metric.

[0052] Preferably, let It is a set of numerical feature dimensions. For a set of categorical feature dimensions, Let be the total number of order feature dimensions, and satisfy: For any numerical feature dimension Its rolling time window The degree of dispersion of numerical features within the current optimizable order set can be determined based on this dimension. The standard deviation or normalized standard deviation is determined. Preferably, the overall dispersion of the numerical feature is considered. This is expressed as the average of the dispersion of each numerical feature dimension. For any categorical feature dimension... Its rolling time window The information entropy within is: in, Representing the dimensionality of categorical features The set of possible values, This represents any value of the categorical feature dimension. Indicates the value In the scrolling time window The probability of occurrence within.

[0053] Preferably, the overall discreteness of categorical features It can be determined by the following formula: in, Represents the set of categorical feature dimensions The number of feature dimensions contained therein.

[0054] Preferably, the relative weighting coefficient Used to coordinate the relative contributions of numerical and categorical features in distance metrics, and can be determined by the following formula: in, The overall dispersion of numerical features This represents the overall dispersion of categorical features. When At that time, it can be Set to a preset value or use a method of adding extremely small positive numbers for smoothing.

[0055] Preferably, the difference index Used to characterize the rolling time window The overall dispersion of internal order features, and the overall dispersion of numerical features. Overall dispersion of categorical features Combining the results, we get: in, This represents the relative weighting coefficient between numerical features and categorical features.

[0056] Preferably, the clustering threshold parameter Based on the difference index Perform dynamic updates: in, Indicates the previous scrolling time window The corresponding clustering threshold parameter, Indicates the previous scrolling time window The corresponding difference index, This represents the threshold update coefficient. To suppress threshold fluctuations, the following can be applied: Set a threshold range constraint to satisfy: in, and These represent the lower and upper limits of the clustering threshold parameter, respectively.

[0057] Preferably, dimensional difference index Used to characterize feature dimension In the scrolling time window The degree of difference within. For numerical feature dimensions, the dimension difference index... It can be determined based on the standard deviation, normalized standard deviation, or coefficient of variation of this dimension; for categorical feature dimensions, the dimension difference index... This can be determined based on the information entropy of this dimension. Based on the aforementioned dimensional difference index... The feature dimension weights are then normalized and updated to obtain the normalized weights. : in, For summing indexes of feature dimensions, This represents the total number of order feature dimensions. When... When this is the case, the weights of each feature dimension can be set to uniform weights.

[0058] Preferably, to enhance the stability of weight updates, the feature dimension weights are adjusted. Use smooth updates: in, is the weighted smoothing coefficient, and ; Representing feature dimension In the previous scrolling time window Weight within, Indicates based on the current scrolling time window Normalized weights obtained from the internal dimension difference index.

[0059] Preferably, the clustering distance metric can incorporate feature dimension weights. and relative weight coefficients To coordinate the relative contributions of different feature dimensions and numerical and categorical features, the distance can be determined by the following formula: in, For orders The order feature vector, For scrolling time windows Internal order cluster The cluster center, Indicates cluster center In feature dimension The value on, This is an inconsistency measure function for categorical features; when the order feature value is the same as the cluster center value, The value is 0, when the two are different. The value is 1.

[0060] Preferably, the cluster center offset index can also be determined according to the maximum offset method: in, Indicates a scrolling time window The maximum offset of each order cluster center relative to the previous scrolling time window. When using the maximum offset method for re-clustering triggering judgment, it can be... With clustering threshold parameter Compare them.

[0061] After clustering and parameter updates are completed, multi-production line order allocation is performed under the constraints of the frozen region. (Frozen region order set) Orders within the specified time window will maintain their predetermined production line allocation; The current set of optimizable orders within Based on order clusters Clustering results and order processable domains The process involves determining the cluster-level processable domain corresponding to each order cluster and constructing an initial allocation scheme accordingly. This initial allocation scheme is then validated against production line capacity constraints, process rule constraints, and frozen zone constraints. If an infeasible allocation exists, adjustments are made based on infeasibility repair rules to obtain a feasible allocation scheme. After obtaining the feasible allocation scheme, further allocation optimization processing is performed, and the multi-production line allocation results are output. Figure 3 As shown, under the constraint of the frozen region, the multi-production line allocation process includes steps such as determining the cluster-level processable domain, screening and sorting candidate production lines, constructing the initial allocation scheme, cluster-level repair, and infeasibility repair.

[0062] Step S7 in this embodiment specifically includes: (7.1) In a rolling time window Internally, regarding the current set of orders that can be optimized. Each order cluster in In the order processing domain Under constraints, the order cluster is determined based on the production line capacity constraint model and the process rule base. Cluster-level processable domains.

[0063] Specifically, for order clusters and any of the processing steps involved. According to the orders within the cluster, in the processing stage The following manufacturable domain Order clusters are derived from this. In the processing stage Cluster-level processable domain Preferably, when order clusters Domestic orders require processing. When clusters are jointly assigned to the same candidate production line, the cluster-level processable domains can be determined as follows: in, Indicates order cluster In the processing stage The set of candidate production lines that can be jointly selected. Order cluster. The overall cluster-level processable domain is represented as: in, Indicates order cluster The set of processing steps involved.

[0064] In obtaining cluster-level processable domains Then, in each processing stage The candidate production lines are then sorted according to the matching rules between order clusters and candidate production lines to determine the order clusters. In the processing stage The preferred production line sequence is selected, and the target production line is selected from the preferred production line sequence to form an initial allocation scheme for the order cluster under each processing stage.

[0065] Preferably, for order clusters Processing and candidate production lines Construct a comprehensive matching score for candidate production lines. : in, Indicates order cluster In the processing stage Next, select candidate production lines The overall matching score; Indicates order cluster Assigned to production line The resulting delivery time risk; Indicates production line Order cluster The resulting switching costs; Indicates processing stage Off the production line The load level; Indicates order cluster With production line The degree of process matching; , , , This represents the corresponding scoring weight coefficient.

[0066] According to the comprehensive matching score Candidate production lines are sorted in ascending order of score, with lower-scoring candidate lines prioritized as target lines to balance delivery time risk, switchover costs, production line load balancing, and process compatibility. The matching rules between order clusters and candidate production lines comprehensively consider at least one or more of the following factors: delivery time risk, switchover costs, remaining available production line capacity, production line load balancing contribution, and process path connectivity requirements, to improve the feasibility and stability of the initial allocation scheme.

[0067] Preferably, when order clusters At any processing stage Cluster-level processable domain When empty, the order cluster will be... The order clusters are marked as infeasible and cluster-level repair is performed to restore their matchability under the processable domain constraints. The cluster-level repair preferably employs a split repair method: when order clusters... When the demand exceeds the available capacity of any candidate production line, or when the cluster-level processable domain is empty due to incompatibility of the order processable domains within the cluster, the order cluster will be... Split into at least two sub-clusters; redetermine the sub-cluster-level processable domains for each sub-cluster, and perform local re-clustering or further splitting as necessary based on the degree of difference within the sub-cluster; when the sub-cluster-level processable domains are not empty, perform candidate production line screening, preferred sorting, and cluster-production line matching under the constraints of the sub-cluster-level processable domains; if the sub-cluster-level processable domains are still empty after reaching the preset minimum splitting granularity, mark the corresponding sub-cluster as an object to be further processed, and re-execute the allocation in the subsequent rolling time window in combination with the updated order set and production line status.

[0068] (7.2) After constructing the initial allocation scheme, in the cluster-level processable domain Under constraints, a feasibility check is performed on the initial allocation scheme to determine whether it meets the production line capacity constraint model, process rule constraints, freeze zone constraints, and connection requirements between adjacent processing stages. When an infeasible allocation scheme is detected in the order's processable domain, and cluster-level cultivable domain Under constraints, the initial allocation scheme is adjusted according to the infeasibility repair rule to restore allocation feasibility.

[0069] The infeasible allocation includes at least the following: insufficient remaining available capacity of the target production line leading to unmet capacity constraints; the current load of the target production line exceeding the preset load limit leading to unmet load constraints; and the target production line not belonging to an order cluster. Cluster-level machinable domain under the corresponding processing stage This results in the failure to meet manufacturability constraints and process connection constraints between adjacent processing stages.

[0070] In handling cases where repair is not feasible, the "minimum disturbance" principle is preferred. While maintaining the order allocation results in the frozen area, priority is given to addressing the order clusters that caused the infeasibility. and corresponding processing steps Perform migration repair. The migration repair includes: in the corresponding processing stage... Cluster-level processable domain Select the next preferred production line as the alternative target production line and allocate the order cluster. In the processing stage The process will migrate from the current target production line to the alternative target production line; when the cluster-level processable domain... When multiple feasible alternative production lines exist, the alternative production line with the least impact on delivery time risk, lower switching cost, and greater benefit to load balancing should be selected first.

[0071] If migration repair still fails to restore feasibility, further swap repair, split repair, or rollback repair will be performed. Swap repair includes selecting at least two order clusters between production lines in the same processing stage or with substitutable processing relationships for swap allocation, ensuring that each swapped order cluster meets the processing feasibility and capacity constraints of the corresponding production line, and simultaneously updating the load status and delivery risk assessment of the relevant production lines. Split repair includes removing order clusters that would otherwise become infeasible. The process is split into at least two sub-clusters, and the cluster-level processable domain and candidate production lines are redefined for each sub-cluster. The rollback repair includes fixing order clusters that would otherwise become infeasible. Revert to a higher-compatibility production line in its preferred production line sequence to reconstruct a feasible allocation scheme; if necessary, reassign the already allocated order clusters. Mark it as an object to be processed further, and reassign it in the subsequent rolling time window by combining the updated order set and production line status.

[0072] After completing the infeasibility repair process, output a feasible allocation scheme that satisfies the production line capacity constraint model.

[0073] (7.3) After obtaining a feasible allocation scheme that satisfies the production line capacity constraint model, perform iterative optimization on the feasible allocation scheme under the frozen zone constraint until the allocation result is within the current rolling time window. If no significant improvement occurs or the iteration termination condition is met, the multi-production line allocation result is output.

[0074] Specifically, after obtaining a feasible allocation scheme that satisfies the production line capacity constraint model, iterative optimization is performed on the feasible allocation scheme under the frozen zone constraint to reduce the load imbalance between production lines under each processing stage, and within the cluster-level processable domain. The feasibility and stability of the allocation results are improved under constraints. Preferably, the iterative optimization aims to minimize the comprehensive objective function, which can be expressed as: in, Indicates a scrolling time window The overall optimization target value within; This indicates the degree of load imbalance on the production line at each processing stage; This indicates a risk or penalty for order delivery delays; This represents the switching cost incurred when order clusters are adjusted between production lines; This indicates the cost of disruption caused by infeasible repairs or adjustments to the plan; , , , This corresponds to the target weighting coefficient. Preferably, it represents the production line load imbalance. It can be determined based on the deviation between the production line load rate and the average load rate at each processing stage, for example: in, Indicates processing stage Off the production line In the scrolling time window Internal load rate, Indicates processing stage Each production line in the rolling time window The average load rate within.

[0075] During the iteration process, based on the current load status and remaining available capacity of each production line under each processing stage, order clusters with high load, low allocation feasibility, or high process connection risks in the current allocation plan are identified. Identify and use the order cluster As the target of adjustment, in the corresponding processing stage Cluster-level processable domain Within, for the order cluster The production line allocation is performed by relocation, reallocation, or necessary cluster-level splitting and reassignment to generate adjustment plans.

[0076] Perform constraint verification on the adjustment plan; if it meets the production line capacity constraint model, process rule constraints, and freeze zone constraints, update the current allocation plan based on the load improvement effect, changes in delivery time risk, and changes in switchover cost; otherwise, keep the original allocation plan unchanged. Repeat the above process until the allocation result falls within the current rolling time window. If no significant improvement occurs or the iteration termination condition is met, the multi-production line allocation result is output.

[0077] This invention also provides an embodiment of a steel order multi-production line allocation optimization system based on dynamic clustering. The system includes a data acquisition and preprocessing module, a time window and freeze zone management module, a workable domain determination module, a dynamic clustering processing module, and a multi-production line allocation optimization module. Each module is used to implement order data preprocessing, two-level capability constraint modeling, rolling time window and freeze zone management, overall workable domain determination of the order, dynamic clustering incremental update, adaptive adjustment of clustering parameters, overall workable domain determination of the order cluster, infeasible allocation repair, and multi-production line order allocation result output.

[0078] The data acquisition and preprocessing module is used to acquire net demand production order data and production line capacity data, perform missing value processing, outlier removal, feature encoding and unit unification on the order data, and construct order feature vectors; at the same time, it establishes a production line capacity constraint model to describe the processability relationship, capacity carrying relationship and production line capacity constraints between the order and the candidate production line at each processing stage involved in its process path. The time window and freeze zone management module is used to determine the scrolling time window according to the preset window length and scrolling step. The system combines the order entry status, delivery date requirements, and production lead time to determine the set of orders within the current rolling time window. In the scrolling time window The frozen area is defined internally, and the frozen area order set is created. The orders in the frozen area will maintain the established production line allocation results, and the orders outside the frozen area will be identified as the current set of orders that can be optimized. This is for subsequent dynamic clustering processing and multi-production line allocation optimization; The processable domain determination module is used to determine the current set of optimizable orders based on the production line capacity constraint model and the process rule library. Each order Determine its machinable domain Among them, orders Machinable domain This includes the machinable domains of an order at each processing stage involved in its process path. The processable domain is used to limit the range of candidate production lines for orders to participate in dynamic clustering, incremental updates, re-clustering triggering, and multi-production line allocation optimization. The dynamic clustering processing module is used to perform clustering within a rolling time window. Inherited from the previous scrolling time window The clustering results yield the current set of orders that can be optimized. Corresponding initial order cluster and cluster center When the current set of orders can be optimized When new orders are added, canceled, or when the order feature vector changes or the order's processable domain changes, incremental attribution, removal, or re-attribution processing is performed only on the orders that have changed, and the order cluster is updated accordingly. and cluster center Simultaneously calculate the cluster center offset index. The dynamic clustering processing module is also used to perform cluster center offset based on the cluster center offset index. With clustering threshold parameter Determine whether to trigger re-clustering, and in the determination process, base the decision on the difference index. Clustering threshold parameter Perform dynamic updates based on dimensional difference metrics Weights of feature dimensions Perform adaptive updates and adjust the relative weight coefficients. Perform adaptive adjustments; The multi-production line allocation optimization module is used to optimize the current order cluster under the constraints of the frozen zone. Construct an initial allocation scheme and perform allocation optimization. For each order cluster... The multi-production line allocation optimization module is located in the order processing domain. Under constraints, the cluster-level machinable domains of the order cluster are derived at each processing stage. And form a cluster-level processable domain for the entire order cluster. Within the cluster-level processable domain, target production lines for each processing stage are determined according to the order cluster and candidate production line matching rules to form an initial allocation scheme. When an infeasible allocation occurs in the initial allocation scheme, it is adjusted according to the infeasibility repair rules to obtain a feasible allocation scheme that meets the requirements of the production line capacity constraint model and the process rule base. After obtaining a feasible allocation scheme, a rolling time window is used... The internal process performs iterative optimization on the feasible allocation scheme to reduce the load imbalance between production lines under each processing stage, reduce the risk of constraint violation, and improve the feasibility and stability of the allocation results, and finally outputs the order allocation results for each production line.

[0079] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, modifications or variations can still be made to the technical solutions described above, and these modifications and variations all fall within the protection scope of the present invention.

Claims

1. A dynamic clustering-based optimization method for multi-line assignment of steel orders, characterized in that, include: S1: Obtain steel production order data and production line capacity data to be allocated, construct order feature vectors based on the steel production order data, and establish a production line capacity constraint model; the production line capacity constraint model is used to characterize the processability relationship and capacity carrying capacity relationship between orders and corresponding candidate production lines in each processing stage; S2: Determine the rolling time window according to the preset window length and rolling step, and determine the set of orders within the current rolling time window based on the order's entry status into the system, delivery date requirements, and production lead time; Set a freeze zone within the rolling time window to keep the orders within the freeze zone unchanged by the predetermined production line allocation, thereby determining the current set of orders that can be optimized; S3: Within the rolling time window, based on the production line capacity constraint model and the process rule base, determine the processable domain of each order in the current optimizable order set under each processing stage, and form the overall processable domain of the order; S4: Within the current rolling time window, orders with similar characteristics and the same or compatible processing domains inherit the clustering results from the previous time window, and use the clustering results as the initial state of the order clusters within the current time window. S5: When the current set of optimizable orders changes, for the changed orders, perform incremental attribution, removal or re-attribution processing based on the updated order feature vector and processable domain, update the order cluster and the corresponding cluster center, and calculate the cluster center offset index at the same time. S6: Determine whether to trigger re-clustering based on the cluster center offset index and clustering threshold parameter, and perform adaptive updates on clustering-related parameters; S7: Under the constraints of the frozen area, determine the cluster-level processable domain based on the current order cluster and the order processable domain, and construct an initial allocation scheme according to the matching rules between the order cluster and the candidate production line; When the initial allocation scheme becomes infeasible, under the constraints of the order-level processable domain and the cluster-level processable domain, the initial allocation scheme is adjusted according to the infeasibility repair rule to restore allocation feasibility and obtain a feasible allocation scheme. Perform allocation optimization processing on the feasible allocation scheme and output the multi-production line allocation results.

2. The method according to claim 1, wherein, In step S1, the production line capacity constraint model is used to determine the processing steps involved in the order according to the process path of the order, and to determine the corresponding candidate production line set for each processing step; based on the specification compatibility, material compatibility, process compatibility, equipment availability status and capacity carrying capacity between the order and the candidate production lines, the processable domain of the order under each processing step is determined, and the processable domain of the order is formed by the processable domain under each processing step.

3. The method of claim 1, wherein, In step S2, the frozen area is determined based on one or more of the following: order plan execution status, whether the order has been issued for execution, whether the order has started, whether the order has entered the scheduled execution window, the order delivery date distance, or the latest start time distance. Orders within the frozen zone will retain their predetermined production line allocation results in this round of calculations and will not participate in this round of dynamic clustering processing and multi-production line order allocation optimization. Orders that have not entered the freeze zone within the rolling time window are identified as the current set of orders that can be optimized. The current set of orders that can be optimized includes one or more of the following: orders newly entering the rolling time window, orders left over from the previous rolling time window that have not entered the freeze zone, and orders whose order attributes have changed and are allowed to be re-optimized.

4. The method of claim 1, wherein, In step S3, the processable domain of the order is determined based on the production line capacity constraint model and process rule base established in step S1. Specifically, this includes: determining the processing steps involved in the order according to the process path of the order; for each processing step, screening candidate production lines in the corresponding candidate production line set based on the production line capacity constraint model and process rule base to obtain the processable domain of the order under each processing step; forming the overall processable domain of the order from the processable domains of the order under each processing step, and using the overall processable domain to limit the dynamic clustering boundary and multi-production line allocation range of the order.

5. The method for optimizing the allocation of steel orders across multiple production lines based on dynamic clustering according to claim 1, characterized in that, In step S5, during the rolling time window Within, when the current set of orders can be optimized When changes occur, incremental processing is performed only on the orders that have changed, and within the order's processable domain. Update order clusters under constraints This is to avoid performing clustering calculations repeatedly on all orders; The incremental processing includes incremental assignment, removal, or reassignment processing; wherein, cluster assignment processing is performed on newly added orders, removal processing is performed on canceled orders, and reassignment processing is performed on orders whose feature vectors have changed or whose processable domains have changed. Upon completion of the delta processing, the corresponding order cluster is updated and cluster center , and based on a rolling time window and the distance between the cluster center in the current rolling time window and the cluster center in the previous rolling time window , a cluster center shift metric is calculated.

6. The method for optimizing the allocation of steel orders across multiple production lines based on dynamic clustering according to claim 5, characterized in that, In step S6, the cluster center offset indicator is used as a criterion to determine whether to trigger re-clustering and the cluster threshold parameter The determination of whether to trigger re-clustering includes: when the order cluster structure change degree in the rolling time window exceeds the current cluster stability threshold, re-clustering is triggered; when the existing cluster structure is maintained unchanged, and incremental processing is continued based on the existing cluster structure.

7. The method for optimizing the allocation of steel orders across multiple production lines based on dynamic clustering as described in claim 1, characterized in that, In step S6, adaptive updates to clustering-related parameters include: based on the difference index Clustering threshold parameter Dynamic updates are performed based on dimensional difference indicators. Weights of feature dimensions Perform adaptive updates and adjust the relative weight coefficients based on the dispersion of numerical features and the dispersion of categorical features. Adaptive adjustment is performed; wherein, the clustering threshold parameter Feature dimension weights and relative weighting coefficients Used to adjust the scrolling time window Clustering granularity and distance metric in the internal order clustering process.

8. The method for optimizing the allocation of steel orders across multiple production lines based on dynamic clustering according to claim 1, characterized in that, Step S7 specifically includes: (7.1) In a rolling time window Internally, regarding the current set of orders that can be optimized. Each order cluster in Based on the order-processable domain Production line capacity constraint model and process rule base to determine order clusters Cluster-level machinable domains at each processing stage and form order clusters Overall cluster-level machinable domain ; In obtaining cluster-level processable domains Then, the candidate production lines are sorted according to the matching rules between order clusters and candidate production lines to determine the order clusters. The optimal production line sequence is selected at each processing stage, and a target production line is selected from the optimal production line sequence to form an initial allocation scheme; (7.2) After constructing the initial allocation scheme, in the cluster-level processable domain Under constraints, a feasibility check is performed on the initial allocation scheme to determine whether it satisfies the production line capacity constraint model, process rule constraints, and freeze zone constraints. When an infeasible allocation is detected in the initial allocation scheme, an infeasibility check is performed in the order processable domain. and cluster-level cultivable domain Under constraints, the initial allocation scheme is adjusted according to the infeasibility repair rule to restore allocation feasibility; After completing the infeasibility repair process, output a feasible allocation scheme that satisfies the production line capacity constraint model; (7.3) After obtaining a feasible allocation scheme that satisfies the production line capacity constraint model, perform iterative optimization on the feasible allocation scheme under the frozen zone constraint until the allocation result is within the current rolling time window. If no further significant improvement occurs or the iteration termination condition is met, the multi-production line allocation result is output.

9. The method for optimizing the allocation of steel orders across multiple production lines based on dynamic clustering according to claim 8, characterized in that, In step S7, the infeasibility repair rule includes: when the initial allocation scheme violates one or more of the following constraints: production line capacity, process compatibility, equipment status, delivery date, or adjacent processing link connection, the order's processable domain is used as the basis for correction. and cluster-level cultivable domain To constrain the boundaries, for order clusters that cause infeasible allocation, migration, exchange, splitting, or rollback operations are performed between candidate production lines at the corresponding processing stage to restore allocation feasibility.

10. A steel order multi-production line allocation optimization system based on dynamic clustering, characterized in that, The system includes a data acquisition and preprocessing module, a time window and freeze zone management module, a processable domain determination module, a dynamic clustering processing module, and a multi-production line allocation optimization module. The data acquisition and preprocessing module is used to acquire steel production order data and production line capacity data, perform missing value processing, outlier removal, feature encoding and unit unification on the steel production order data, construct order feature vectors, and establish a production line capacity constraint model. The time window and freeze zone management module is used to determine the rolling time window according to the preset window length and rolling step size, and to determine the set of orders in the current rolling time window based on the order entry status, delivery date requirements and production lead time. Within the rolling time window, a freeze zone is defined to keep the orders within the freeze zone unchanged from the predetermined production line allocation, and the orders outside the freeze zone are defined as the current set of orders that can be optimized, for use in subsequent dynamic clustering and multi-production line allocation optimization. The processable domain determination module is used to determine the processable domain of each order in the current optimizable order set under each processing stage based on the production line capacity constraint model and process rule library, and form the overall processable domain of the order from the processable domains under each processing stage, so as to limit the dynamic clustering boundary and multi-production line allocation range of the order. The dynamic clustering processing module inherits the clustering results of the previous rolling time window within the rolling time window to obtain the initial order clusters and cluster centers corresponding to the current optimizable order set. When the current optimizable order set is added, canceled, or the order feature vector changes or the order's processable domain changes, incremental assignment, removal, or reassignment processing is performed only on the orders that have changed, and the order clusters and cluster centers are updated accordingly. At the same time, the cluster center offset index is calculated. The dynamic clustering processing module is also used to determine whether to trigger re-clustering based on the cluster center offset index and the clustering threshold parameter, and to perform adaptive updates on the clustering-related parameters. The multi-production line allocation optimization module is used to determine the cluster-level processable domain of the order cluster based on the current order cluster and the order processable domain under the constraints of the frozen area, and to determine the target production line under each processing stage under the constraints of the cluster-level processable domain according to the matching rules between the order cluster and the candidate production line, so as to construct an initial allocation scheme. When the initial allocation scheme becomes infeasible, the initial allocation scheme is adjusted according to the infeasibility repair rules to obtain a feasible allocation scheme. After obtaining the feasible allocation scheme, iterative optimization of the feasible allocation scheme is performed within a rolling time window, and the multi-production line allocation results are output.

Images