An intelligent hanging control method for medium and thick plate blanks combined with machine learning

Abstract

The present invention provides a method for intelligently controlling the hanging of medium and thick plate slabs in combination with machine learning, and relates to the field of automation technology. Based on the hanging relationship between medium and thick plate slabs and orders, the present invention constructs a steelmaking mark substitution relationship list and a set of candidate orders for medium and thick plate slabs, extracts the problem characteristics of hanging medium and thick plate slabs, converts them into mathematical expressions, describes the optimization objectives and process constraints of hanging medium and thick plate slabs, and establishes a system optimization model; designs a heuristic algorithm to generate an initial set of feasible solutions for hanging medium and thick plate slabs; designs a neighborhood tabu search algorithm to optimize the solutions in the initial set of feasible solutions; optimizes the distribution of orders on medium and thick plate slabs according to the large unit weight reduction rule; designs an internal optimization algorithm for a double-layer medium and thick plate slab hanging solution based on machine learning; and re-optimizes the sub-plate sequence in the double-layer medium and thick plate slab hanging solution.

Description

Technical Field

[0001] The present invention relates to the field of automation technology, and in particular to an intelligent hooking control method for medium and thick plate blanks combined with machine learning. Background Art

[0002] Intense global market competition and a relative oversupply of steel production capacity have led to a continuous decline in steel prices, shrinking profit margins. Coupled with increasing market demand for more diverse steel varieties and specifications, this has necessitated flexible production planning and scheduling to achieve agile manufacturing, improve steel production efficiency, and maximize profits with fixed resources, thereby reducing costs. However, due to factors such as market conditions, process technology, and production management, it is difficult for steel companies to achieve the zero inventory goal outlined in discrete manufacturing management. The steel production process inevitably generates a certain amount of unfinished semi-finished product inventory. Rationally and efficiently utilizing this inventory is crucial to addressing the capital and space constraints associated with this excess inventory, ensuring smooth logistics, and improving companies' ability to deliver on time.

[0003] Medium and thick plate surplus refers to slabs that are not matched with orders or slabs that are out of order during the medium and thick plate production process. There are three main reasons for the generation of surplus in medium and thick plate production: (1) Order reasons, which refer to the inconsistency between the order demand for slabs and the quantity of slabs produced by steelmaking furnaces. Both small and large order quantities will produce surplus; (2) Quality reasons, which refer to the production of surplus slabs due to excessive chemical composition, surface or internal quality defects, which change the original steelmaking mark or purpose of the slabs and are inconsistent with the original order. Quality reasons include steel modification for furnace retention, steel modification for slabs of different steel grades, steel modification for slabs at specific locations of special steel grades, and steel modification or degradation of the surface or internal quality of the slabs; (3) Specification reasons, which refer to the fact that the actual length of the slabs is not within the predetermined length range, thus not meeting the order quality requirements and generating surplus slabs. Among them, the slab length discrepancy includes two situations: the slab length discrepancy in online cutting and the slab length discrepancy in offline cutting. The former is because some users have special requirements for the quality of steel coils, and the predetermined length range of slabs is small. When the last batch of molten steel in the tundish is poured, the cutting model cannot completely distribute the pouring length to each slab according to the predetermined length, resulting in one or more slabs being cut to a length that does not meet the predetermined length; the latter is that some slabs in the slab warehouse are cut to a certain length due to defects, which may cause the cut slabs to be converted into leftover slabs because the length does not meet the order requirements.

[0004] The prolonged storage and unclear destination of excess medium and heavy plate stock leads to logistical congestion, resource and capital backlogs, and compromises resource utilization and order fulfillment, complicating production organization and increasing production costs. Therefore, the management of excess medium and heavy plate stock has become a crucial aspect of production management for steel companies. This is particularly true in a dynamic and volatile market environment, where product demand fluctuates unavoidably, while internal production must be balanced. Faced with this imbalance between product demand and production, completely aligning production with orders is unrealistic. When product orders are low and equipment capacity is relatively oversupplied, companies typically produce a certain amount of material based on demand forecasts, in addition to order-based production, to compensate for the shortfall in orders, ensure primary equipment utilization, and meet mass production requirements. Excess medium and heavy plate stock is typically managed by modifying the quality and dimensions of the original slabs, or by downgrading the steel grade, to ensure that the slabs meet the order requirements before rolling.

[0005] The solution to managing excess medium and thick plate stock is to manually match it with orders. In practice, manual matching takes a significant amount of time. Due to attentional limitations, manually selected orders cannot guarantee optimal matching with slabs. Slabs that may not need to be cut are cut, resulting in waste and unavoidable missed and mismatched slabs. Currently, manual slab matching still relies on experience, significantly influenced by subjective factors. This results in poor process stability and low slab utilization. In the patent "A Production Planning Optimization Method for Reducing Steel Plate Sampling Losses in a Manufacturing System," Song Chengzhong et al. studied a production planning optimization method to reduce steel plate sampling losses. In the patent "A Slab Assembly Method and Apparatus," Xing Jianchang et al. designed a heuristic slab assembly method. In the patent "Slab Matching Control Method for Multiple Hot Rolling Lines to Improve Material Utilization in Steel Enterprises," Tang Lixin et al. designed multiple heuristic and intelligent search algorithms to optimize slab matching across multiple hot rolling lines in steel enterprises, improving slab utilization. In their patent application, "Order Configuration and Slab Assembly Virtual Slab Overall Optimization Device and Method," Li Ziqiang et al. proposed a method for rapid order optimization and overall optimization design of slab / slab assembly virtual slabs that meets multiple rule constraints and real-time computing requirements. In their paper "Modeling and scatter search algorithm for dynamic slab allocation problem in iron and steel enterprises," Tang et al. studied the dynamic slab allocation problem and designed a multi-neighborhood search algorithm to approximate the problem, achieving a high-quality near-optimal solution in a short time. Summary of the Invention

[0006] In view of the shortcomings of the existing technology, the present invention provides an intelligent hanging control method for medium and thick plate blanks combined with machine learning.

[0007] A method for intelligently controlling the hooking of medium and thick plate slabs combined with machine learning includes the following steps:

[0008] Step 1: Based on the connection between the medium and thick plate blanks and orders, a list of steel-making mark substitution relationships is constructed;

[0009] The steel mark substitution relationship list describes the correspondence between the steel tapping marks of the medium and thick plate blanks and the steel tapping marks of the production orders.

[0010] Step 1.1: Extract the tapping marks of the medium and thick plate blanks into a set, then sort the tapping marks in descending order, then remove duplicate tapping marks, and finally obtain a unique list of tapping marks of the medium and thick plate blanks;

[0011] Step 1.2: Extract the tapping marks of the production order into a set, then sort the tapping marks in descending order, then remove duplicates from the tapping marks, and finally obtain a unique list of the tapping marks of the production order;

[0012] Step 1.3: Traverse the unique list of tapping marks of the medium and thick plate blanks, and select an unselected tapping mark as the blank tapping mark;

[0013] Step 1.4: traverse the unique list of tapping marks of the production order and select an unselected tapping mark as the production order tapping mark, marking the production order tapping mark as "traversed". This mark is the "production order tapping mark traversal flag";

[0014] Step 1.5: If the production order tapping mark is linked to the remaining billet tapping mark, create a set of available production order tapping marks for the remaining billet tapping mark, and add the production order tapping mark to the set of available production order tapping marks; otherwise, execute step 1.4 until all production order tapping marks are marked as "traversed"; when all production order tapping marks are marked as "traversed", initialize all "production order tapping mark traversal flags" to "not traversed"; then execute step 1.3 until all medium and thick plate tapping marks in the unique list of medium and thick plate remaining billet tapping marks are traversed;

[0015] Step 2: Construct a set of candidate orders for medium and thick plate blanks; the set of candidate orders for medium and thick plate blanks contains multiple sub-sets. Each sub-set corresponds to a candidate order set for medium and thick plate blanks, including all orders linked to the medium and thick plate blanks.

[0016] Step 2.1: Select a tapping mark from the unique list of tapping marks for the medium and heavy plate blank, and extract the corresponding set of available production order tapping marks;

[0017] Step 2.2: Select production orders based on the extracted set of available production order tapping marks, filter the production orders based on the order constraints of the medium and heavy plate slabs, and form the candidate order set for the medium and heavy plate slabs after filtering;

[0018] Step 2.3: If all the tapping marks in the unique list of medium and heavy plate blank tapping marks have been traversed, the set of candidate orders for medium and heavy plate blanks is constructed. Otherwise, execute step 2.1.

[0019] The order constraints for the medium and thick plate blank hooking include:

[0020] (1) The thickness of the production order is within the upper and lower limits of the thickness allowed for the order of medium and thick plate blanks.

[0021] (2) The width of the production order is within the upper and lower limits of the order width allowed for medium and thick plate blanks;

[0022] (3) The length of the generated order is within the upper and lower limits of the order length allowed for medium and thick plate blanks;

[0023] The "upper and lower limits for thickness," "upper and lower limits for width," and "upper and lower limits for length" represent actual production requirements: Due to limitations in production equipment, the thickness, width, and length of a medium-thick plate blank order must be within these limits to meet production quality requirements. Once the thickness, width, length, and tapping mark of the medium-thick plate blank are determined, the corresponding thickness, width, and length will also be determined. The "upper and lower limits for thickness," "upper and lower limits for width," and "upper and lower limits for length" are fixed parameters.

[0024] Step 3: Extract the problem characteristics of medium and thick plate blank hanging, convert them into mathematical expressions, and describe the optimization objectives and process constraints of medium and thick plate blank hanging;

[0025] Step 3.1: Define the decision variables x for the plate blank hanging problem ijl ; Define k1 as the master order for the first layer of the medium and thick plate blank j, k2 as the master order for the second layer of the medium and thick plate blank j, d i is the length of the i-th order, w i is the weight of the i-th order, h i is the width of the ith order, z kkjl is the master order k on the lth layer of the jth medium and thick plate blank. If k exists, then z kkjl is 1, otherwise z kkjl is 0.

[0026] The decision variable x ijl represents the number of sub-slabs of level l attached to uncommissioned slab j by order i;

[0027] Step 3.2: Design the optimization target for the hanging of medium and thick plate blanks;

[0028] The optimization objectives of the medium and thick plate slab hooking include: maximizing the unit weight of the slab, minimizing the cutting loss of the medium and thick plate slab, minimizing the amount of spot goods carried out, and giving priority to large-section slabs, among which:

[0029] (1) Maximizing the unit weight of the blank: The blank weight refers to the weight of the blank attached to the panelization plan; the blank weight includes the weight of the attached sub-plates and the weight of the trimmed edges and the weight of the trimmed ends, that is, the weight of the finished medium and heavy plate; the greater the total weight of the spot and futures finished medium and heavy plate sub-plates, the greater the unit weight of the blank;

[0030] The toleranced hanging thickness of the finished plate is T j , the hook width is the sum of the widths of the two layers of main orders, that is, The hanging length is the maximum of the two hanging lengths. Since the length of the first layer must be greater than the second layer, the hanging length is the sum of the lengths of all sub-boards in the first layer, that is, The cutting amount corresponding to the hanging scheme is S j , the amount of cutting head and tail is A j , so the weight of a single billet is Maximizing the single weight of the blank means maximizing the total design weight of all blanks, that is: ρ is the density of steel.

[0031] (2) Minimize the weight loss of medium and thick plate blanks: The weight loss of medium and thick plate blanks refers to the weight of the slabs that are not used during the hanging process of medium and thick plate blanks. The weight loss of medium and thick plate blanks includes the weight loss of the trimming and cutting heads and tails, as well as the weight loss caused by the different widths of the sub-plates.

[0032] The actual finished product quantity in the hanging plan of medium and thick plate surplus is the total weight of the ordered sub-plate futures and spot goods, that is, Therefore, the total cutting loss of medium and thick plate blanks is the difference between the total weight of the finished medium and thick plate blanks and the total weight of the finished medium and thick plate blanks. Minimizing the cutting loss of medium and thick plate blanks is

[0033]

[0034] (3) Minimize the quantity of spot products brought out: The quantity of spot products brought out refers to the quantity that exceeds the number of pieces ordered in the design of the medium and thick plate blank hanging scheme. If order i brings out spot products, then the number of sub-plates of order i exceeds the number of pieces ordered in order i. If order i does not bring out spot products, the quantity of spot products brought out in order i is considered to be 0; therefore, the quantity of spot products brought out in order i is expressed as where p i is the number of pieces ordered for order i. The optimization objective of minimizing the spot take-out quantity in all medium and thick plate blank hanging schemes is to express it as

[0035] (4) Priority for large-section stock: The stock section is the thickness and width of the medium and thick plate stock. Medium and thick plate stock should be sorted in descending order of thickness, width and cross-section. After sorting, the higher the priority attribute of the medium and thick plate stock, the higher the priority attribute.

[0036] Step 3.3: Analyze the processing technology limitations of the medium and thick plate blank hanging and set the hanging constraints;

[0037] The attachment constraints include attachment limit constraints, production process constraints and main order constraints;

[0038] (1) The linking restriction is: only the medium and thick plate blanks and orders with a steel-tapping mark dominance relationship can be linked together; the lower and upper limits of the thickness range of the medium and thick plate are T min and T max , the lower and upper limits of the width interval are H min and H max Therefore, T min ≤T j ≤T max and H min ≤H j ≤H max , where T j is the thickness of the finished plate j including thickness tolerance, H j is the width of the finished plate j including the trimming amount.

[0039] (2) Production process constraints: (i) Each medium and thick plate blank is allowed to be hung at most once, i.e. (ii) The jump width of each sub-plate on the remaining medium and thick plate cannot exceed the jump width upper limit, that is, |min{1,x ujl}h u -min{1,x vjl}h v |≤H c , where order u and order v are orders with different production order numbers, H c (iii) The length of the plan for hanging the remaining slab of medium and thick plates cannot exceed the limit length, that is, Among them D j (iv) The yield rate of the hanging plan for the medium and thick plate blank shall not be less than the lower limit of the yield rate Y c ;Right now

[0040]

[0041] (3) Master order constraint: Each layer of each medium and thick plate blank can only have one master order, that is, M is a very large number;

[0042] Step 4: Design a heuristic algorithm to generate a set of feasible solutions for initial medium and thick plate blank hanging;

[0043] Step 4.1: Design a width rule priority heuristic algorithm;

[0044] Step 4.1.1: Arrange the orders in the candidate order set for medium and heavy plate blanks in descending order of width;

[0045] Step 4.1.2: Select one of the candidate orders for the medium and thick plate blank as the hook order, calculate the maximum number of hooks for the order under the current remaining length based on the selected hook order, and execute step 4.1.4.

[0046] Step 4.1.3: Reduce the maximum number of attachments by 1.

[0047] Step 4.1.4: Attach orders according to the maximum attachment number and check whether the current solution status meets the requirements for successful attachment of medium and thick plate stock. If so, terminate the attachment of the stock and record the attachment solution. If the requirements for failure to attach medium and thick plate stock are met and the maximum attachment number is greater than 0, release all current attachment orders and execute step 4.1.3. If the requirements for failure to attach medium and thick plate stock are met and the maximum attachment number is less than or equal to 0, execute step 4.1.2. If the requirements for feasible attachment of medium and thick plate stock are met, maintain the current attachment status and execute step 4.1.2.

[0048] The requirement for successful hanging of the medium and thick plate surplus is that the thickness, width and length of the hanging solution meet the thickness, width and length requirements in the order constraints.

[0049] The requirement for failure of hanging the medium and thick plate blank is that at least one of the thickness, width and length of the hanging solution does not meet the requirements of the order constraints.

[0050] The feasible requirements for hanging the medium and thick plate surplus billet are: the thickness and width of the hanging scheme meet the requirements of the order constraints, and the length of the hanging scheme is less than the lower limit of the length required in the order constraints.

[0051] Step 4.2: Design a single-rolled, same-size-first heuristic algorithm

[0052] Step 4.2.1: Arrange the orders in the candidate order set for medium and thick plate surplus slabs in descending order by width and length. Orders with the same width and length are arranged in descending order by the number of unassembled slabs.

[0053] Step 4.2.2: Select one order from the sorted set of candidate orders for medium and heavy plate blanks as the attached order;

[0054] Step 4.2.3: Calculate the maximum number of joins for the selected order given the current remaining length. If this maximum number is less than the number of unjoined panels in the order, update the order attributes, record the panelization plan, and terminate the panelization. If this maximum number is greater than the number of unjoined panels in the order, join the order to the remaining plate stock according to its number of unjoined panels, and then execute Step 4.2.2 until the requirements for successful joining of the remaining plate stock are met. If the requirements for failed joining of the remaining plate stock are met, release all currently joined orders and execute Step 4.2.2. If the requirements for feasible joining of the remaining plate stock are met, execute Step 4.2.2 while retaining the current joining status.

[0055] Step 4.3: Divide the orders in the set of candidate orders for medium and heavy plate blanks into sets and design a heuristic algorithm for set partitioning rules based on machine learning;

[0056] Step 4.3.1: Design a machine learning algorithm to predict the location of the partition point in the order group and form a subset of orders based on the partition point.

[0057] Step 4.3.2: Generate hook plans using the order subsets respectively;

[0058] Step 4.3.3: Evaluate the attachment plans of all order subsets and select the attachment plan of the order subset with the highest value as the actual attachment plan for the medium and thick plate blank;

[0059] The prediction steps of the machine learning algorithm are:

[0060] Step S1: m attribute features are selected according to the attributes of the medium and thick plate slab and its order candidate group, and the attribute data of the order and the medium and thick plate slab are obtained based on the m attribute features. Then, the n1 order data of the same batch and the current medium and thick plate slab data are merged into a (n1+1)×m two-dimensional array. The upper limit of the number of rows of the two-dimensional array is N, and the value 0 is inserted in the position less than N. Finally, the data of each attribute feature is normalized. These data are called the total data set. The attachment scheme corresponding to the medium and thick plate slab and its candidate order set is split to form an order subset included in the attachment scheme and an order subset not included. The dividing point between the two is the order subset division point.

[0061] Step S2: Divide the total data set into a training data set, a test data set, and a validation data set in a ratio of 7:2:1, and then use the support vector machine algorithm, random forest algorithm, and regression algorithm to train their respective mathematical models in the same training data set; using accuracy as an indicator, evaluate the models trained by the three algorithms separately, and select the one with the highest accuracy among the three models as the rule model.

[0062] Step S3: Arrange the features of the medium and thick plate blanks and their corresponding medium and thick plate blank order candidate groups into a two-dimensional array of (n1+1)×m, normalize them, and input them into the rule model. The rule model will give the corresponding order subset partition point results.

[0063] Step 5: According to the judgment conditions of the optimizable solution, optimizable solutions are extracted from the three initial hanging solution sets as the solutions to be improved; an order neighborhood tabu search algorithm is designed to improve the hanging solution of the medium and thick plate blank; if a better solution is found, the original optimizable solution is replaced, otherwise the next optimizable solution is extracted for improvement.

[0064] The optimizable solution judgment condition is that any mounting solution that meets any of the following conditions is an optimizable solution:

[0065] (1) Set the yield rate improvement standard Y lower_bound , yield rate y j The hooking scheme is smaller than the standard, that is, the yield rate meets y j <Y lower_bound The attachment scheme is an optimizable scheme;

[0066] (2) With the same thickness as the sub-plate on the mounting scheme, there are a number of sub-plates that are not assembled, U i Greater than 0, and the order of the corresponding medium and thick plate blank continues to be connected, that is, the order that satisfies U i If the order with value > 0 exists in the candidate order set corresponding to the remaining billet, then the attachment scheme is an optimizable scheme;

[0067] The order neighborhood tabu search algorithm is described as:

[0068] Step D1: Obtain an order neighborhood for each sub-plate according to the sub-plate of the current mounting solution for the medium and thick plate slab; the orders in the order neighborhood consist of orders that can replace the current sub-plate;

[0069] If the current sub-plate is the first sub-plate, then the orders in its order neighborhood have three characteristics: 1) they exist in the same set of candidate orders for medium and thick plate surplus as the first sub-plate; 2) their width is not less than that of the second sub-plate; 3) the number of unassembled sub-plates in the order is greater than 0;

[0070] If the current sub-plate is the last sub-plate, then the orders in its order field have three characteristics: 1) it exists in the same set of candidate orders for medium and thick plate surplus slabs as the first sub-plate; 2) its width is not greater than that of the second-to-last sub-plate; 3) the number of unassembled sub-plates in the order is greater than 0;

[0071] If the current sub-plate is any sub-plate other than the first or last one, then the orders in its order field have three characteristics: 1) it exists in the same set of candidate orders for medium and thick plate surplus as the first sub-plate; 2) its width range is between the widths of the two adjacent sub-plate orders; 3) the number of unassembled sub-plates in the order is greater than 0;

[0072] Step D2: Select any sub-board order from the optimizable attachment schemes, and then replace it according to its order neighborhood. If the objective function satisfies the following formula, the attachment scheme replaces the original scheme; otherwise, continue searching the order neighborhood until the order neighborhood search is completed.

[0073]

[0074] where x' ijl This is the replacement hook scheme for the medium and thick plate blank j. j A' is the trimming amount of the hanging scheme after the replacement of the medium and thick plate blank j. j It is the head and tail cutting amount of the hanging scheme after the replacement of the medium and thick plate blank j.

[0075] Step 6: According to the rule of reducing the weight of large orders, adjust the distribution of orders on the hanging plan of medium and thick plate surplus billets.

[0076] The large unit weight demotion rule is as follows: all the medium and thick plate blanks that have been successfully mounted are sorted in descending order according to unit weight to form a candidate set; all the medium and thick plate blanks that have not been successfully mounted are sorted in descending order according to the unit weight to form a target set; the medium and thick plate blanks in the candidate set and the target set are selected in the sorted order, so that the selected medium and thick plate blanks are in descending order of weight;

[0077] Step 6.1: List the orders in the candidate order set corresponding to the medium and thick plate blanks for which the attachment scheme is generated, and search for medium and thick plate blanks that can be attached to the order based on the steel tapping marks of these orders; if the found medium and thick plate blank is attached to the order, and the order is not the main order, record the blank tuple "(medium and thick plate blank, order)" formed by the medium and thick plate blank and the order; after all orders have been traversed, obtain a list of candidate blanks for the medium and thick plate blanks in the expanded scheme, and sort the blanks in the candidate blank list from largest to smallest according to unit weight;

[0078] Step 6.2: Select the remaining blank tuples from the candidate list of remaining blanks, and reduce the orders contained in the selected tuples by 1 unit, that is, the number of sub-boards of the i-th order is reduced from the original a i Reduced to a i-1, and the reduced part is added to the optimized medium and thick plate blank until the order i completely disappears from the hanging scheme where it is located, or is reduced to the point where it cannot meet the hanging requirements of the medium and thick plate blank where it is located, and then replace it with a sub-plate with a different order number and continue this operation.

[0079] Step 6.3: Check the attachment status of the current attachment scheme of the optimized medium and thick plate blank; if the attachment success condition is met, terminate the process; if the attachment failure condition is met, restore the original attachment scheme of the reduced-dimensional blank, and replace the next blank tuple for an attempt; if the attachment feasibility condition is met, retain the attachment status of the current optimized medium and thick plate blank, and replace the next blank tuple for an attempt.

[0080] Step 7: Design an internal optimization algorithm for the double-layer medium and thick plate blank hanging scheme based on machine learning; re-optimize the sub-plate sequence in the double-layer medium and thick plate blank hanging scheme.

[0081] Step 7.1: Use the internal optimization prediction method to define the set of residual blank design schemes that need to be optimized, and use the sub-plate distribution, spot quantity, yield rate, cutting loss amount and remaining orders that can be attached to the residual blank of the current residual blank design scheme as input features, and input a mark 0 or 1. 0 means no optimization is required, and 1 means optimization is required.

[0082] The internal optimization prediction method comprises the following steps:

[0083] Step B1: The sub-plate distribution vector of the current blank design scheme Spot quantity s, yield rate y, cutting loss u, and remaining orders that can be attached to the remaining billet As a feature vector Where L1 is the number of sub-plates in the design of the remaining plate, L2 is the number of remaining orders, and n3 is the number of eigenvectors that can be obtained. i =(t i ,h i ,d i ), i=1,...,L1, is the (thickness, width, length) triplet of order i in the medium and heavy plate blank design plan. β k =(t k ,h k ,d k ),k=1,...,L2,is the (thickness, width, length) triplet of the remaining order k. So the input matrix The output vector is Is the output feature, the value is 0 or 1; 0 means no optimization is required, 1 means optimization is required. The input matrix and output vector constitute the original dataset.

[0084] Step B2: Divide the original dataset into 10 equal subsets, labeled 0-9. Starting with subset 0, the span 1 is the subset movement step, and 9 is the subset capacity. Each movement step is one step, resulting in a subset containing 9 data copies. After 10 moves, 10 subsets containing 9 data copies are obtained. These 10 subsets are divided into training, testing, and validation data subsets in a 7:2:1 ratio. Θ(χ) is iteratively trained on each of the 10 subsets.

[0085] Step B3: Extract the data features of all current medium and thick plate blank design schemes, construct the input matrix χ, output the predicted label vector O, and then divide the medium and thick plate blank design schemes that need to be optimized according to the predicted labels;

[0086] Step 7.2: Select a blank from the set of blank design solutions to be optimized, optimize the first and second-layer sub-panels in the solution, and sequentially transfer the first-layer sub-panels to the second layer of the blank attachment solution. If the yield rate increases after the transfer, update the original solution; otherwise, try the next-layer sub-panel.

[0087] Step 7.3: Convert the double-layer medium and thick plate blank mounting scheme into a single-layer one; release all the sub-plates in the current mounting scheme, then sort them in descending order according to width, and re-mount them in sequence to form a new mounting scheme;

[0088] Step 7.4: Check whether the new hanging scheme meets the conditions for successful hanging of medium and thick plate blanks. If so, and the yield rate is improved, update the current hanging scheme; otherwise, keep it as it is.

[0089] Step 8: Send the attachment plan to the rolling production, execute the production plan, and complete the attachment control of the medium and thick plate slabs.

[0090] The beneficial effects of adopting the above technical solution are:

[0091] The present invention provides a machine learning-based intelligent hooking control method for medium and thick plate surplus. This method addresses the problem of a large number of uncommissioned slabs and low utilization rates in steel companies. This method improves material utilization during the hooking process, reduces energy costs and raw material consumption caused by repeated steelmaking, and simultaneously reduces excess material and inventory levels. This optimizes the entire hooking process toward the goals of high returns, low losses, and low costs, ensuring timely delivery of urgent orders while improving order integrity and customer satisfaction. The application of machine learning technology reduces the search scope of the optimization algorithm and improves operational efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0092] Figure 1 It is a flow chart of the method for hanging medium and thick plate surplus billets according to the specific implementation of the present invention.

[0093] Figure 2 It is a schematic diagram of hanging the remaining medium and thick plate according to a specific embodiment of the present invention.

[0094] Figure 3 It is a flow chart of the breadth-first heuristic algorithm specifically implemented in the present invention.

[0095] Figure 4 This is a flow chart of the single rolling-same specification priority heuristic algorithm specifically implemented by the present invention.

[0096] Figure 5 It is a flow chart of the set partitioning heuristic algorithm based on machine learning implemented in the present invention. DETAILED DESCRIPTION

[0097] The following embodiments of the present invention are described in further detail with reference to the accompanying drawings and examples. The following examples are used to illustrate the present invention but are not intended to limit the scope of the present invention.

[0098] The specific implementation of the present invention is equipped with corresponding hardware conditions and software conditions. The hardware conditions mainly include: at least one server, at least one IPV4 network interface, at least one network cable, and at least one router. These hardware facilities are connected through a network communication protocol to form a local area network, and the local area network has a direct or indirect information communication channel with the information management platform ERP system. The ERP system refers to a management platform that is based on information technology and uses systematic management ideas to provide decision-making and operation means for enterprise decision-makers and employees. The software conditions mainly include: database software SQL SERVER2008R2 and data remote communication software IBM WebSphere MQ. The database software needs to configure the server address, server port number, database name, user name and password, and the communication software needs to configure the queue manager, queue, channel, port number, and server address.

[0099] Medium and thick plate surplus data and order data are transmitted via communication software. The medium and thick plate surplus data and order data are downloaded to the local server. The downloaded medium and thick plate surplus information fields contain information important for calculations, including: slab number, steel grade code, nominal thickness, nominal width, nominal length, actual weight, theoretical weight, previous process, steel quality code, tapping mark, actual steel quality code, slab mass, actual thickness, actual width, actual length, and actual theoretical weight of the slab. Order data fields include: main production line code, billet design request plan number, production order number, order number, raw material source code, tapping mark, steel quality code, representative steel code, order thickness, designed product thickness, order width, order width 1, order length, order length 1, order weight, number of ordered pieces, order quantity, number of sub-plates missing in current status, and quantity missing in current status.

[0100] The medium and thick plate surplus data and order data are generated by the ERP system and sent to the local server through the communication software IBM WebSphere MQ. When the local server executes the download command, it receives the data from IBM WebSphere MQ, converts the data format, and stores it in the local computer in the form of a string. The string format data is then cut, and the SQL statement is executed to insert the cut string data into the corresponding data table of the SQL SERVER 2008R2 database, and the downloaded medium and thick plate surplus and order data are displayed on the intelligent panelization system interface. After the data is successfully downloaded, the data management function can be used to manually add, delete, and modify the data; the data source (steel grade, representative steel code, order thickness, key orders) can be set independently to filter the data; key orders and key medium and thick plate surplus can be defined independently through the key order setting and key medium and thick plate surplus setting functions; data statistics (total number of data items, total weight of medium and thick plate surplus or orders) can be displayed.

[0101] A method for intelligent hooking control of medium and thick plate blanks combined with machine learning, such as Figure 1 As shown, the following steps are included:

[0102] Step 1: Based on the connection between the remaining plate and the order, Figure 2 As shown, a steel tapping symbol substitution relationship list is constructed;

[0103] The steel mark substitution relationship list describes the correspondence between the steel-out marks of the medium and thick plate blanks and the steel-out marks of the production orders. In the design of the medium and thick plate blank hanging order, the rules for judging whether the hanging is feasible include order constraints and steel-out mark correspondence constraints. Order constraints refer to the fact that in the control of the hanging of medium and thick plate blanks, the medium and thick plate blanks and the orders to which they are hung must meet certain restrictions on thickness, width and length. The steel-out mark correspondence constraint refers to the fact that the steel-out marks of the medium and thick plate blanks and the steel-out marks of the hanging order must comply with the correspondence between the steel-out marks. It is worth noting that the steel-out marks of the medium and thick plate blanks and the steel-out marks of the hanging order are a one-to-many relationship, that is, the steel-out mark of one medium and thick plate blank is allowed to hang orders under multiple steel-out marks.

[0104] Step 1.1: Extract the tapping marks of the medium and thick plate blanks into a set, then sort the tapping marks in descending order, then remove duplicate tapping marks, and finally obtain a unique list of tapping marks of the medium and thick plate blanks;

[0105] Step 1.2: Extract the tapping marks of the production order into a set, then sort the tapping marks in descending order, then remove duplicates from the tapping marks, and finally obtain a unique list of the tapping marks of the production order;

[0106] Step 1.3: Traverse the unique list of tapping marks of the medium and thick plate blanks, and select an unselected tapping mark as the blank tapping mark;

[0107] Step 1.4: traverse the unique list of tapping marks of the production order and select an unselected tapping mark as the production order tapping mark, marking the production order tapping mark as "traversed". This mark is the "production order tapping mark traversal flag";

[0108] Step 1.5: If the production order tapping mark is linked to the remaining billet tapping mark, create a set of available production order tapping marks for the remaining billet tapping mark, and add the production order tapping mark to the set of available production order tapping marks; otherwise, execute step 1.4 until all production order tapping marks are marked as "traversed"; when all production order tapping marks are marked as "traversed", initialize all "production order tapping mark traversal flags" to "not traversed"; then execute step 1.3 until all medium and thick plate tapping marks in the unique list of medium and thick plate remaining billet tapping marks are traversed;

[0109] Step 2: Construct a set of candidate orders for medium and thick plate blanks; the set of candidate orders for medium and thick plate blanks contains multiple sub-sets. Each sub-set corresponds to a candidate order set for medium and thick plate blanks, including all orders linked to the medium and thick plate blanks.

[0110] Step 2.1: Select a tapping mark from the unique list of tapping marks for the medium and heavy plate blank, and extract the corresponding set of available production order tapping marks;

[0111] Step 2.2: Select production orders based on the extracted set of available production order tapping marks, filter the production orders based on the order constraints of the medium and heavy plate slabs, and form the candidate order set for the medium and heavy plate slabs after filtering;

[0112] Step 2.3: If all the tapping marks in the unique list of medium and heavy plate blank tapping marks have been traversed, the set of candidate orders for medium and heavy plate blanks is constructed. Otherwise, execute step 2.1.

[0113] The order constraints for the medium and thick plate blank hooking include:

[0114] (1) The thickness of the production order is within the upper and lower limits of the thickness allowed for the order of medium and thick plate blanks.

[0115] (2) The width of the production order is within the upper and lower limits of the order width allowed for medium and thick plate blanks;

[0116] (3) The length of the generated order is within the upper and lower limits of the order length allowed for medium and thick plate blanks;

[0117] The "upper and lower limits for thickness," "upper and lower limits for width," and "upper and lower limits for length" represent actual production requirements: Due to limitations in production equipment, the thickness, width, and length of a medium-thick plate blank order must be within these limits to meet production quality requirements. Once the thickness, width, length, and tapping mark of the medium-thick plate blank are determined, the corresponding thickness, width, and length will also be determined. The "upper and lower limits for thickness," "upper and lower limits for width," and "upper and lower limits for length" are fixed parameters.

[0118] Step 3: Extract the problem characteristics of medium and thick plate blank hanging, convert them into mathematical expressions, and describe the optimization objectives and process constraints of medium and thick plate blank hanging;

[0119] Step 3.1: Define the decision variables x for the plate blank hanging problem ijl ; Define k1 as the master order for the first layer of the medium and thick plate blank j, k2 as the master order for the second layer of the medium and thick plate blank j, d i is the length of the i-th order, w i is the weight of the i-th order, h i is the width of the ith order, z kkjl is the master order k on the lth layer of the jth medium and thick plate blank. If k exists, then z kkjl is 1, otherwise z kkjl is 0.

[0120] The decision variable x ijl represents the number of sub-slabs of level l attached to uncommissioned slab j by order i;

[0121] Step 3.2: Design the optimization target for the hanging of medium and thick plate blanks;

[0122] The optimization objectives of the medium and thick plate slab hooking include: maximizing the unit weight of the slab, minimizing the cutting loss of the medium and thick plate slab, minimizing the amount of spot goods carried out, and giving priority to large-section slabs, among which:

[0123] (1) Maximizing the unit weight of the blank: The blank weight refers to the weight of the slab attached to the panelization plan; the blank weight includes the weight of the attached sub-plates and the weight of the trimmed edges and the weight of the trimmed ends, that is, the weight of the finished medium and thick plate; the greater the total weight of the spot and futures sub-plates of the finished medium and thick plate, the greater the unit weight of the blank; maximizing the unit weight of the blank optimizes the remaining blank attachment plan towards the direction of maximizing the slab utilization rate.

[0124] The toleranced hanging thickness of the finished plate is T j , the hook width is the sum of the widths of the two layers of main orders, that is, The hanging length is the maximum of the two hanging lengths. Since the length of the first layer must be greater than the second layer, the hanging length is the sum of the lengths of all sub-boards in the first layer, that is, The cutting amount corresponding to the hanging scheme is S j , the amount of cutting head and tail is A j , so the weight of a single billet is Maximizing the single weight of the blank means maximizing the total design weight of all blanks, that is: ρ is the density of steel.

[0125] (2) Minimize the loss of medium and thick plate blanks: The loss of medium and thick plate blanks refers to the weight of the slabs that are not used during the process of hanging medium and thick plate blanks. After the medium and thick plate blanks are hung, they are rolled to become finished medium and thick plates, and then cut into sub-plates corresponding to the production order. Figure 1 As shown in the figure, the weight loss of medium and thick plate slabs includes the weight lost from trimming the edges and ends, as well as the weight loss caused by the different widths of the sub-plates. Although minimizing the amount of cutting loss and maximizing the unit weight of the slab are both aimed at improving the utilization rate of medium and thick plate slabs, there are also differences between the two. Maximizing the unit weight of the slabs tends to combine more sub-plates when the medium and thick plate slabs are mounted, while minimizing the amount of cutting loss optimizes the combination of sub-plates mounted on each medium and thick plate slab to reduce the unused portion of the slab.

[0126] The actual finished product quantity in the hanging plan of medium and thick plate surplus is the total weight of the ordered sub-plate futures and spot goods, that is, Therefore, the total cutting loss of medium and thick plate blanks is the difference between the total weight of the finished medium and thick plate blanks and the total weight of the finished medium and thick plate blanks. Minimizing the cutting loss of medium and thick plate blanks is

[0127]

[0128] (3) Minimize the amount of spot carryout: The amount of spot carryout refers to the amount of excess order quantity in the design of the medium and thick plate blank hanging scheme. This part of the product needs to be sold in the spot market, and its price should be lower than the futures price. Therefore, reducing the amount of spot carryout is conducive to improving production revenue. If order i brings out spot products, then the number of sub-plates of order i exceeds the number of orders for order i. If order i does not bring out spot products, the amount of spot carryout of order i is considered to be 0; therefore, the amount of spot carryout of order i is expressed as where p i is the number of pieces ordered for order i. The optimization objective of minimizing the spot take-out quantity in all medium and thick plate blank hanging schemes is to express it as

[0129] (4) Priority for large-section blanks: The blank section is the thickness and width of the medium and thick plate blank. Large-section blanks are a relative concept. Given any two medium and thick plate blanks, the medium and thick plate blank with the larger thickness and width is the larger blank. In the calculation, the priority of the large-section blank depends on the comparison of the thickness and width of the medium and thick plate blanks. Therefore, the medium and thick plate blanks need to be sorted in descending order of thickness and width. After sorting, the higher the priority of the medium and thick plate blanks, the higher the priority. Large-section blanks have processing advantages, and generally speaking, large-section medium and thick plate blanks are also heavier, so they can be placed on more orders.

[0130] Step 3.3: Analyze the processing technology limitations of the medium and thick plate blank hanging and set the hanging constraints;

[0131] The attachment constraints include attachment limit constraints, production process constraints and main order constraints;

[0132] (1) The linking restriction constraint is: only the medium and thick plate blanks and orders with the steel tapping mark dominance relationship can be linked together; the steel tapping mark dominance relationship structure contains the specification constraints of the sub-plates. After the medium and thick plate blank j is linked to the order, it needs to go through the rolling process to become the finished medium and thick plate j. Due to the limitations of the production process and production equipment, the lower limit and upper limit of the thickness range of the finished medium and thick plate are T min and T max , the lower and upper limits of the width interval are H min and H max Therefore, T min ≤T j ≤T max and H min ≤H j ≤H max , where T jis the thickness of the finished plate j including thickness tolerance, H j is the width of the finished plate j including the trimming amount.

[0133] (2) Production process constraints: (i) Each medium and thick plate blank is allowed to be hung at most once, i.e. In actual production, each medium and thick plate blank has its own blank number, so each medium and thick plate blank has two possibilities: (1) it meets the attachment conditions and is attached successfully; (2) it cannot find an order that meets the attachment conditions, cannot be attached, and continues to remain in the slab inventory. If the medium and thick plate blank j is attached, then its first-level main order must exist, that is, at least one order can be used as the main order for the medium and thick plate blank. The length of the first layer of the medium and thick plate blank must be greater than the length of the second layer, so the second-level main order will only be allowed to exist if the first-level main order exists, that is, (ii) The width jump of each sub-plate on the remaining medium and thick plate cannot exceed the upper limit of the width jump. The width jump refers to the difference in width between the sub-plate with the largest width and the sub-plate with the smallest width in the same paneling solution. ujl}h u -min{1,x vjl}h v |≤H c , where order u and order v are orders with different production order numbers, H c (iii) The length of the plan hanging of the medium and thick plate blank cannot exceed the limit length. The limit length of the hanging refers to the length of the plate plan. This length is related to the rolling length of the medium and thick plate blank. It can also be given manually. Usually the minimum value of the two is taken. In order to facilitate production, it is generally stipulated that in the double-layer hanging plan of the medium and thick plate blank, the length of the first layer hanging plan should not be less than the length of the second layer hanging plan, that is, The process of rolling medium and thick plate slabs into medium and thick plates requires the use of multiple production equipment, such as rolling mills, which have their own processing capacity limitations, such as the maximum rolling length of the rolling mill. Therefore, the minimum value of the processing capacity upper limit of all equipment is used as the maximum length required by production. The total length of all orders for the first layer of medium and thick plate slabs is less than the maximum length required by production. In addition, it can also be given manually, but usually the minimum value of the two is taken, that is, Among them D j (iv) The yield rate of the hanging plan for the medium and thick plate blank shall not be less than the lower limit of the yield rate Y c The yield rate is an important evaluation criterion for the hanging scheme of medium and thick plate surplus billets. The hanging scheme with a yield rate lower than the specified lower limit is considered to have no optimization value. That is, the yield rate of all hanging schemes must not be less than the lower limit of the yield rate.

[0134]

[0135] (3) Master order constraint: Each layer of each medium and thick plate blank can only have one master order, that is, M is a very large number;

[0136] Step 4: Design a heuristic algorithm to generate a set of feasible solutions for initial medium and thick plate blank hanging;

[0137] Step 4.1: Design a width rule priority heuristic algorithm; e.g. Figure 3 shown.

[0138] Step 4.1.1: Arrange the orders in the candidate order set for medium and heavy plate blanks in descending order of width;

[0139] Step 4.1.2: Select one of the candidate orders for the medium and thick plate blank as the hook order, calculate the maximum number of hooks for the order under the current remaining length based on the selected hook order, and execute step 4.1.4.

[0140] Step 4.1.3: Reduce the maximum number of attachments by 1.

[0141] Step 4.1.4: Attach orders according to the maximum attachment number and check whether the current solution status meets the requirements for successful attachment of medium and thick plate stock. If so, terminate the attachment of the stock and record the attachment solution. If the requirements for failure to attach medium and thick plate stock are met and the maximum attachment number is greater than 0, release all current attachment orders and execute step 4.1.3. If the requirements for failure to attach medium and thick plate stock are met and the maximum attachment number is less than or equal to 0, execute step 4.1.2. If the requirements for feasible attachment of medium and thick plate stock are met, maintain the current attachment status and execute step 4.1.2.

[0142] The requirement for successful hanging of the medium and thick plate surplus is that the thickness, width and length of the hanging solution meet the thickness, width and length requirements in the order constraints.

[0143] The requirement for failure of hanging the medium and thick plate blank is that at least one of the thickness, width and length of the hanging solution does not meet the requirements of the order constraints.

[0144] The feasible requirements for hanging the medium and thick plate surplus billet are: the thickness and width of the hanging scheme meet the requirements of the order constraints, and the length of the hanging scheme is less than the lower limit of the length required in the order constraints.

[0145] Step 4.2: Design a single-rolling-same specification priority heuristic algorithm, such as Figure 4 shown.

[0146] Step 4.2.1: Arrange the orders in the candidate order set for medium and thick plate surplus slabs in descending order by width and length. Orders with the same width and length are arranged in descending order by the number of unassembled slabs.

[0147] Step 4.2.2: Select one order from the sorted set of candidate orders for medium and heavy plate blanks as the attached order;

[0148] Step 4.2.3: Calculate the maximum number of joins for the selected order given the current remaining length. If this maximum number is less than the number of unjoined panels in the order, update the order attributes, record the panelization plan, and terminate the panelization. If this maximum number is greater than the number of unjoined panels in the order, join the order to the remaining plate stock according to its number of unjoined panels, and then execute Step 4.2.2 until the requirements for successful joining of the remaining plate stock are met. If the requirements for failed joining of the remaining plate stock are met, release all currently joined orders and execute Step 4.2.2. If the requirements for feasible joining of the remaining plate stock are met, execute Step 4.2.2 while retaining the current joining status.

[0149] 4.3: To increase the diversity of medium and heavy plate slab attachment solutions, the orders in the medium and heavy plate slab candidate order set are partitioned into sets, and machine learning methods are used to assist in the generation of attachment solutions. A heuristic algorithm for set partitioning rules based on machine learning is designed;

[0150] Step 4.3.1: Design a machine learning algorithm to predict the location of the partition point in the order group and form a subset of orders based on the partition point.

[0151] Step 4.3.2: Generate hook plans using the order subsets respectively;

[0152] Step 4.3.3: Evaluate the attachment plans of all order subsets and select the attachment plan of the order subset with the highest value as the actual attachment plan for the medium and thick plate blank;

[0153] The prediction step of the machine learning algorithm is as follows Figure 5 As shown,

[0154] Step S1: m attribute features are selected according to the attributes of the medium and thick plate slabs and their order candidate groups, such as thickness, width, length, delivery time, number of unassembled sub-plates, etc. The attribute data of the order and the medium and thick plate slabs are obtained based on the m attribute features, and then the n1 order data of the same batch and the current medium and thick plate slabs data are merged into a (n1+1)×m two-dimensional array. The upper limit of the number of rows of the two-dimensional array is N, and the value 0 is inserted in the position less than N. Finally, the data of each attribute feature is normalized. These data are called the total data set. The attachment scheme corresponding to the medium and thick plate slabs and their candidate order set is split to form an order subset included in the attachment scheme and an order subset not included. The dividing point between the two is the order subset division point.

[0155] Step S2: Divide the total data set into a training data set, a test data set, and a validation data set in a ratio of 7:2:1, and then use the support vector machine algorithm, random forest algorithm, and regression algorithm to train their respective mathematical models in the same training data set; using accuracy as an indicator, evaluate the models trained by the three algorithms separately, and select the one with the highest accuracy among the three models as the rule model.

[0156] Step S3: Arrange the features of the medium and thick plate blanks and their corresponding medium and thick plate blank order candidate groups into a two-dimensional array of (n1+1)×m, normalize them, and input them into the rule model. The rule model will give the corresponding order subset partition point results.

[0157] Step 5: Select and optimize the initial solution.

[0158] Step 5.1: Based on the optimizable solution judgment conditions, extract optimizable solutions from the three initial attachment solution sets as the solutions to be improved.

[0159] Step 5.2: Execute the order neighborhood tabu search algorithm to improve the hanging scheme of medium and thick plate blanks.

[0160] Step 5.3: If a solution with a better optimization goal is found, replace the original optimizable solution. Otherwise, execute step 5.1 and continue to extract the next optimizable solution for improvement.

[0161] The optimizable solution judgment condition is that any mounting solution that meets any of the following conditions is an optimizable solution:

[0162] (1) Set the yield rate improvement standard Y lower_bound , yield rate y j The hooking scheme is smaller than the standard, that is, the yield rate meets y j <Y lower_bound The attachment scheme is an optimizable scheme;

[0163] (2) With the same thickness as the sub-plate on the mounting scheme, there are a number of sub-plates that are not assembled, U i Greater than 0, and the order of the corresponding medium and thick plate blank continues to be connected, that is, the order that satisfies U i If the order with value > 0 exists in the candidate order set corresponding to the remaining billet, then the attachment scheme is an optimizable scheme;

[0164] The order neighborhood tabu search algorithm is described as:

[0165] Step D1: Obtain an order neighborhood for each sub-plate according to the sub-plate of the current mounting solution for the medium and thick plate slab; the orders in the order neighborhood consist of orders that can replace the current sub-plate;

[0166] If the current sub-plate is the first sub-plate, then the orders in its order neighborhood have three characteristics: 1) they exist in the same set of candidate orders for medium and thick plate surplus as the first sub-plate; 2) their width is not less than that of the second sub-plate; 3) the number of unassembled sub-plates in the order is greater than 0;

[0167] If the current sub-plate is the last sub-plate, then the orders in its order field have three characteristics: 1) it exists in the same set of candidate orders for medium and thick plate surplus slabs as the first sub-plate; 2) its width is not greater than that of the second-to-last sub-plate; 3) the number of unassembled sub-plates in the order is greater than 0;

[0168] If the current sub-plate is any sub-plate other than the first or last one, then the orders in its order field have three characteristics: 1) it exists in the same set of candidate orders for medium and thick plate surplus as the first sub-plate; 2) its width range is between the widths of the two adjacent sub-plate orders; 3) the number of unassembled sub-plates in the order is greater than 0;

[0169] Step D2: Select any sub-board order from the optimizable attachment schemes, and then replace it according to its order neighborhood. If the objective function satisfies the following formula, the attachment scheme replaces the original scheme; otherwise, continue searching the order neighborhood until the order neighborhood search is completed.

[0170]

[0171] where x' ijl This is the replacement hook scheme for the medium and thick plate blank j. j A' is the trimming amount of the hanging scheme after the replacement of the medium and thick plate blank j. j It is the head and tail cutting amount of the hanging scheme after the replacement of the medium and thick plate blank j.

[0172] Step 6: According to the rule of reducing the weight of large orders, adjust the distribution of orders on the hanging plan of medium and thick plate surplus billets.

[0173] The large unit weight demotion rule is as follows: all the medium and thick plate blanks that have been successfully mounted are sorted in descending order according to unit weight to form a candidate set; all the medium and thick plate blanks that have not been successfully mounted are sorted in descending order according to the unit weight to form a target set; the medium and thick plate blanks in the candidate set and the target set are selected in the sorted order, so that the selected medium and thick plate blanks are in descending order of weight;

[0174] Step 6.1: List the orders in the candidate order set corresponding to the medium and thick plate blanks for which the attachment scheme is generated, and search for medium and thick plate blanks that can be attached to the order based on the steel tapping marks of these orders; if the found medium and thick plate blank is attached to the order, and the order is not the main order, record the blank tuple "(medium and thick plate blank, order)" formed by the medium and thick plate blank and the order; after all orders have been traversed, obtain a list of candidate blanks for the medium and thick plate blanks in the expanded scheme, and sort the blanks in the candidate blank list from largest to smallest according to unit weight;

[0175] Step 6.2: Select the remaining blank tuples from the candidate list of remaining blanks, and reduce the orders contained in the selected tuples by 1 unit, that is, the number of sub-boards of the i-th order is reduced from the original a i Reduced to a i -1, and the reduced part is added to the optimized medium and thick plate blank until the order i completely disappears from the hanging scheme where it is located, or is reduced to the point where it cannot meet the hanging requirements of the medium and thick plate blank where it is located, and then replace it with a sub-plate with a different order number and continue this operation.

[0176] Step 6.3: Check the attachment status of the current attachment scheme of the optimized medium and thick plate blank; if the attachment success condition is met, terminate the process; if the attachment failure condition is met, restore the original attachment scheme of the reduced-dimensional blank, and replace the next blank tuple for an attempt; if the attachment feasibility condition is met, retain the attachment status of the current optimized medium and thick plate blank, and replace the next blank tuple for an attempt.

[0177] Step 7: Design an internal optimization algorithm for the double-layer medium and thick plate blank hanging scheme based on machine learning; in order to improve the yield rate, re-optimize the sub-plate sequence in the double-layer medium and thick plate blank hanging scheme to achieve intelligent hanging control of medium and thick plate blanks.

[0178] Step 7.1: Use the internal optimization prediction method to define the set of residual blank design schemes that need to be optimized, and use the sub-plate distribution, spot quantity, yield rate, cutting loss amount and remaining orders that can be attached to the residual blank of the current residual blank design scheme as input features, and input a mark 0 or 1. 0 means no optimization is required, and 1 means optimization is required.

[0179] The internal optimization prediction method is designed based on machine learning methods and can be described as three parts: data set construction, training verification and prediction.

[0180] Step B1: The key to constructing the dataset is the definition of input features and output features. Spot quantity s, yield rate y, cutting loss u, and remaining orders that can be attached to the remaining billet As a feature vector Where L1 is the number of sub-plates in the design of the remaining plate, L2 is the number of remaining orders, and n3 is the number of eigenvectors that can be obtained. i =(t i ,h i ,d i ), i=1,...,L1, is the (thickness, width, length) triplet of order i in the medium and heavy plate blank design plan. β k =(t k ,h k ,d k ),k=1,...,L2,is the (thickness, width, length) triplet of the remaining order k. So the input matrix The output vector is Is the output feature, the value is 0 or 1; 0 means no optimization is required, 1 means optimization is required. The input matrix and output vector constitute the original dataset.

[0181] Step B2: Divide the original dataset into 10 equal subsets, labeled 0-9. Starting with subset 0, the span 1 is the subset movement step, and 9 is the subset capacity. Each movement step is one step, resulting in a subset containing 9 data copies. After 10 moves, 10 subsets containing 9 data copies are obtained. These 10 subsets are divided into training, testing, and validation data subsets in a 7:2:1 ratio. Θ(χ) is iteratively trained on each of the 10 subsets.

[0182] Step B3: Extract the data features of all current medium and thick plate blank design schemes, construct the input matrix χ, output the predicted label vector O, and then divide the medium and thick plate blank design schemes that need to be optimized according to the predicted labels;

[0183] Step 7.2: Select a blank from the set of blank design solutions to be optimized, optimize the first and second-layer sub-panels in the solution, and sequentially transfer the first-layer sub-panels to the second layer of the blank attachment solution. If the yield rate increases after the transfer, update the original solution; otherwise, try the next-layer sub-panel.

[0184] Step 7.3: Convert the double-layer medium and thick plate blank mounting scheme into a single-layer one; release all the sub-plates in the current mounting scheme, then sort them in descending order according to width, and re-mount them in sequence to form a new mounting scheme;

[0185] Step 7.4: Check whether the new hanging scheme meets the conditions for successful hanging of medium and thick plate blanks. If so, and the yield rate is improved, update the current hanging scheme; otherwise, keep it as it is.

[0186] Step 8: Send the attachment plan to the rolling production, execute the production plan, and complete the attachment control of the medium and thick plate slabs.

[0187] This example uses multi-objective optimization to improve various production plans with different characteristics, and sets the default optimal plan. Compared with traditional manual design methods, the number of successful blank attachments increased by 18.4%, the average yield rate of blanks increased by 4.42%, and the spot rate increased by 2.65%.

[0188] The above description is merely a preferred embodiment of the present disclosure and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of the invention involved in the embodiments of the present disclosure is not limited to the technical solutions formed by a specific combination of the above-mentioned technical features, but should also encompass other technical solutions formed by any combination of the above-mentioned technical features or their equivalents without departing from the above-mentioned inventive concept. For example, a technical solution formed by mutually replacing the above-mentioned features with (but not limited to) technical features with similar functions disclosed in the embodiments of the present disclosure.

Claims

1. A method for intelligently controlling the connection of medium and thick plate blanks combined with machine learning, characterized in that: The following steps are involved: Step 1: Based on the connection between the medium and thick plate blanks and orders, a list of steel-making mark substitution relationships is constructed; Step 2: Construct a set of candidate orders for medium and thick plate slabs; the set of candidate orders for medium and thick plate slabs contains multiple sub-sets; each sub-set corresponds to a set of candidate orders for medium and thick plate slabs, including all orders linked to the medium and thick plate slabs; Step 3: Extract the problem characteristics of medium and thick plate blank hanging, convert them into mathematical expressions, and describe the optimization objectives and process constraints of medium and thick plate blank hanging; Step 4: Design a heuristic algorithm to generate a set of feasible solutions for initial medium and thick plate blank hanging; Step 5: Based on the optimizable solution judgment conditions, extract optimizable solutions from the three initial attachment solution sets as the solutions to be improved; An order neighborhood tabu search algorithm is designed to improve the hanging scheme of medium and thick plate blanks. If a better scheme is found, the original optimizable scheme is replaced; otherwise, the next optimizable scheme is extracted for improvement. Step 6: Adjust the order distribution on the hanging plan for medium and thick plate surplus billets according to the rule of reducing the order weight of large orders; The large unit weight demotion rule is as follows: all the medium and thick plate blanks that have been successfully mounted are sorted in descending order according to unit weight to form a candidate set; all the medium and thick plate blanks that have not been successfully mounted are sorted in descending order according to the unit weight to form a target set; the medium and thick plate blanks in the candidate set and the target set are selected in the sorted order, so that the selected medium and thick plate blanks are in descending order of weight; Step 7: Design an internal optimization algorithm for the double-layer medium and thick plate blank hanging scheme based on machine learning; re-optimize the sub-plate sequence in the double-layer medium and thick plate blank hanging scheme; Step 8: Send the attachment plan to the rolling production, execute the production plan, and complete the attachment control of the medium and thick plate slabs.

2. The method for intelligently controlling the connection of medium and thick plate blanks combined with machine learning according to claim 1 is characterized in that: The step 1 specifically includes the following steps: Step 1.1: Extract the tapping marks of the medium and thick plate blanks into a set, then sort the tapping marks in descending order, then remove duplicate tapping marks, and finally obtain a unique list of tapping marks of the medium and thick plate blanks; Step 1.2: Extract the tapping marks of the production order into a set, then sort the tapping marks in descending order, then remove duplicates from the tapping marks, and finally obtain a unique list of the tapping marks of the production order; Step 1.3: Traverse the unique list of tapping marks of the medium and thick plate blanks, and select an unselected tapping mark as the blank tapping mark; Step 1.4: Traverse the unique list of tapping marks for production orders and select an unselected tapping mark as the production order tapping mark. Mark the production order tapping mark as "traversed". This mark is the "production order tapping mark traversal flag"; Step 1.5: If the production order tapping mark is linked to the remaining billet tapping mark, create a set of available production order tapping marks for the remaining billet tapping mark and add the production order tapping mark to the set of available production order tapping marks; otherwise, execute step 1.4 until all production order tapping marks are marked as "traversed"; when all production order tapping marks are marked as "traversed", initialize all "production order tapping mark traversal flags" to "not traversed"; then execute step 1.3 until all medium and heavy plate tapping marks in the unique list of medium and heavy plate remaining billets have been traversed.

3. The method for intelligently controlling the connection of medium and thick plate blanks combined with machine learning according to claim 1 is characterized in that: The steel mark substitution relationship list described in step 1 describes the corresponding relationship between the steel tapping marks of the medium and thick plate blanks and the steel tapping marks of the production order.

4. The method for intelligently controlling the connection of medium and thick plate blanks combined with machine learning according to claim 1 is characterized in that: The step 2 specifically includes the following steps: Step 2.1: Select a tapping mark from the unique list of tapping marks for the medium and heavy plate blank, and extract the corresponding set of available production order tapping marks; Step 2.2: Select production orders based on the extracted set of available production order tapping marks, filter the production orders based on the order constraints of the medium and heavy plate slabs, and form the candidate order set for the medium and heavy plate slabs after filtering; Step 2.3: If all the tapping marks in the unique list of medium and heavy plate blank tapping marks have been traversed, the set of candidate orders for medium and heavy plate blanks is constructed. Otherwise, execute step 2.

1. The order constraints for the medium and thick plate blank hooking include: (1) The thickness of the production order is within the upper and lower limits of the thickness allowed for the medium and heavy plate blank; (2) The width of the production order is within the upper and lower limits of the order width allowed for medium and thick plate blanks; (3) The length of the generated order is within the upper and lower limits of the order length allowed for medium and thick plate blanks; The "upper and lower limits for thickness", "upper and lower limits for width" and "upper and lower limits for length" represent the actual production requirements: due to the limitations of production equipment, the thickness, width and length of a medium and thick plate blank order must be within the upper and lower limits to meet production quality requirements; once the thickness, width, length and tapping mark of the medium and thick plate blank are determined, the corresponding thickness, width and length will also be determined; "upper and lower limits for thickness", "upper and lower limits for width" and "upper and lower limits for length" are fixed parameters.

5. The method for intelligently controlling the connection of medium and thick plate blanks combined with machine learning according to claim 1 is characterized in that: The step 3 specifically includes the following steps: Step 3.1: Define the decision variables x for the plate blank hanging problem ijl ; Define k1 as the master order for the first layer of the medium and thick plate blank j, k2 as the master order for the second layer of the medium and thick plate blank j, d i is the length of the i-th order, w i is the weight of the i-th order, h i is the width of the ith order, z kkjl is the master order k on the lth layer of the jth medium and thick plate blank. If k exists, then z kkjl is 1, otherwise z kkjl is 0; The decision variable x ijl represents the number of sub-slabs of level l attached to uncommissioned slab j by order i; Step 3.2: Design the optimization target for the hanging of medium and thick plate blanks; The optimization objectives of the medium and thick plate slab hooking include: maximizing the unit weight of the slab, minimizing the cutting loss of the medium and thick plate slab, minimizing the amount of spot goods carried out, and giving priority to large-section slabs, among which: (1) Maximizing the unit weight of the blank: The blank weight refers to the weight of the blank attached to the panelization plan; the blank weight includes the weight of the attached sub-plates and the weight of the trimmed edges and the weight of the trimmed ends, that is, the weight of the finished medium and heavy plate; the greater the total weight of the spot and futures finished medium and heavy plate sub-plates, the greater the unit weight of the blank; The toleranced hanging thickness of the finished plate is T j , the hook width is the sum of the widths of the two layers of main orders, that is, The hanging length is the maximum of the two hanging lengths. Since the length of the first layer must be greater than the second layer, the hanging length is the sum of the lengths of all sub-boards in the first layer, that is, The cutting amount corresponding to the hanging scheme is S j , the amount of cutting head and tail is A j , so the weight of a single billet is Maximizing the single weight of the blank means maximizing the total design weight of all blanks, that is: ρ is the density of steel; (2) Minimize the weight loss of medium and thick plate blanks: The weight loss of medium and thick plate blanks refers to the weight of the slabs that are not used during the hanging process of medium and thick plate blanks. The weight loss of medium and thick plate blanks includes the weight loss of the trimming and cutting heads and tails, as well as the weight loss caused by the different widths of the sub-plates. The actual finished product quantity in the hanging plan of medium and thick plate surplus is the total weight of the ordered sub-plate futures and spot goods, that is, Therefore, the total cutting loss of medium and thick plate blanks is the difference between the total weight of the finished medium and thick plate blanks and the total weight of the finished medium and thick plate blanks. Minimizing the cutting loss of medium and thick plate blanks is (3) Minimize the quantity of spot carryout: The quantity of spot carryout refers to the quantity that exceeds the number of pieces ordered in the design of the medium and thick plate blank hanging scheme; if order i brings out spot products, then the number of sub-plates of order i exceeds the number of pieces ordered in order i. If order i does not bring out spot products, the quantity of spot carryout of order i is considered to be 0; therefore, the quantity of spot carryout of order i is expressed as where p i is the number of pieces ordered for order i; the optimization objective of minimizing the spot take-out quantity in all medium and thick plate blank hanging schemes is expressed as (4) Priority is given to large-section blanks: the blank section is the thickness and width cross section of the medium and thick plate blanks; the medium and thick plate blanks need to be sorted in descending order of thickness, width and cross section; after sorting, the higher the priority attribute of the medium and thick plate blanks at the front; Step 3.3: Analyze the processing technology limitations of the medium and thick plate blank hanging and set the hanging constraints; The attachment constraints include attachment limit constraints, production process constraints and main order constraints; (1) The linking restriction is: only the medium and thick plate blanks and orders with a steel-tapping mark dominance relationship can be linked together; the lower and upper limits of the thickness range of the medium and thick plate are T min and T max , the lower and upper limits of the width interval are H min and H max ; Therefore, T min ≤T j ≤T max and H min ≤H j ≤H max , where T j is the thickness of the finished plate j including thickness tolerance, H j is the width of the finished plate j including trimming; (2) Production process constraints: (i) Each medium and thick plate blank is allowed to be hung at most once, i.e. (ii) The jump width of each sub-plate on the remaining medium and thick plate cannot exceed the jump width upper limit, that is, |min{1,x ujl }h u -min{1,x vjl }h v |≤H c , where order u and order v are orders with different production order numbers, H c is the maximum jump width; (iii) the length of the plan hanging of the medium and thick plate blank cannot exceed the limit length, that is Among them D j is the limit length value; (iv) the yield rate of the hanging scheme of the medium and thick plate blank shall not be less than the lower limit of the yield rate Y c ;Right now (3) Master order constraint: Each layer of each medium and thick plate blank can only have one master order, that is, M is a very large number.

6. The method for intelligently controlling the connection of medium and thick plate blanks combined with machine learning according to claim 1 is characterized in that: The step 4 specifically includes the following steps: Step 4.1: Design a width rule priority heuristic algorithm; Step 4.1.1: Arrange the orders in the candidate order set for medium and heavy plate blanks in descending order of width; Step 4.1.2: Select one of the candidate orders for medium and heavy plate slabs as the hook order. Calculate the maximum number of hooks for the selected hook order based on the current remaining length, and then proceed to step 4.1.

4. Step 4.1.3: Reduce the maximum number of attachments by 1; Step 4.1.4: Attach orders according to the maximum attachment number and check whether the current solution status meets the requirements for successful attachment of medium and thick plate stock. If so, terminate the attachment of the stock and record the attachment solution. If the requirements for failure to attach medium and thick plate stock are met and the maximum attachment number is greater than 0, release all current attachment orders and execute step 4.1.

3. If the requirements for failure to attach medium and thick plate stock are met and the maximum attachment number is less than or equal to 0, execute step 4.1.

2. If the requirements for feasible attachment of medium and thick plate stock are met, maintain the current attachment status and execute step 4.1.

2. The requirements for successful hanging of medium and thick plate slabs are: the thickness, width and length of the hanging solution meet the requirements of thickness, width and length in the order constraints; The failure requirement for hanging the medium and thick plate blank is: at least one of the thickness, width and length of the hanging solution does not meet the requirements of the order constraint; The feasible requirements for hanging the medium and thick plate blank are: the thickness and width of the hanging solution meet the requirements of the order constraints, and the length of the hanging solution is less than the lower limit of the length required by the order constraints; Step 4.2: Design a single-rolling-same specification priority heuristic algorithm; Step 4.2.1: Arrange the orders in the candidate order set for medium and thick plate surplus slabs in descending order by width and length. Orders with the same width and length are arranged in descending order by the number of unassembled slabs. Step 4.2.2: Select one order from the sorted set of candidate orders for medium and heavy plate blanks as the attached order; Step 4.2.3: Calculate the maximum number of connections that can be made to the selected connection order based on the current remaining length. If the maximum number is less than the number of unconnected panels in the order, update the order attributes, record the panelization plan, and terminate the panelization. If the maximum number is greater than the number of unconnected panels in the order, connect the order to the remaining medium and thick plate stock according to the number of unconnected panels, and then execute Step 4.2.2 until the requirements for successful connection of the remaining medium and thick plate stock are met. If the requirements for failed connection of the remaining medium and thick plate stock are met, release all currently connected orders and then execute Step 4.2.

2. If the requirements for feasible connection of the remaining medium and thick plate stock are met, execute Step 4.2.2 while retaining the current connection status. Step 4.3: To increase the diversity of medium and heavy plate slab attachment solutions, the orders in the medium and heavy plate slab candidate order set are divided into sets, and a machine learning method is used to assist in the generation of attachment solutions. Design a heuristic algorithm for set partitioning rules based on machine learning; Step 4.3.1: Design a machine learning algorithm to predict the location of the split point in the order group and form a subset of orders based on the split point; Step 4.3.2: Generate hook plans using the order subsets respectively; Step 4.3.3: Evaluate the attachment plans of all order subsets and select the attachment plan of the order subset with the highest value as the actual attachment plan for the medium and thick plate blank; The prediction steps of the machine learning algorithm are: Step S1: m attribute features are selected according to the attributes of the medium and thick plate slabs and their order candidate groups, such as thickness, width, length, delivery time, and number of unassembled sub-plates; attribute data of the orders and medium and thick plate slabs are obtained based on the m attribute features, and then n1 order data of the same batch and the current medium and thick plate slab data are merged into a (n1+1)×m two-dimensional array, where the upper limit of the number of rows of the two-dimensional array is N, and the value 0 is inserted in the position less than N, and finally the data of each attribute feature is normalized; these data are called the total data set; the attachment scheme corresponding to the medium and thick plate slabs and their candidate order set is split to form an order subset included in the attachment scheme and an order subset not included, and the dividing point between the two is the order subset division point; Step S2: The total data set is divided into a training data set, a test data set, and a validation data set in a ratio of 7:2:

1. Then, the support vector machine algorithm, the random forest algorithm, and the regression algorithm are used to train their respective mathematical models on the same training data set. The models trained by the three algorithms are evaluated using accuracy as an indicator, and the one with the highest accuracy among the three models is selected as the rule model. Step S3: Arrange the features of the medium and thick plate blanks and their corresponding medium and thick plate blank order candidate groups into a two-dimensional array of (n1+1)×m, normalize them, and input them into the rule model. The rule model will give the corresponding order subset partition point results.

7. The method for intelligently controlling the connection of medium and thick plate blanks combined with machine learning according to claim 1 is characterized in that: The step 5 specifically includes the following steps: Step 5.1: Based on the optimizable solution judgment criteria, extract optimizable solutions from the three initial attachment solution sets as the solutions to be improved; The optimizable solution judgment condition is that any mounting solution that meets any of the following conditions is an optimizable solution: (1) Set the yield rate improvement standard Y lower_bound , yield rate y j The hooking scheme is smaller than the standard, that is, the yield rate meets y j <Y lower_bound The attachment scheme is an optimizable scheme; (2) With the same thickness as the sub-plate on the mounting scheme, there are a number of sub-plates that are not assembled, U i Greater than 0, and the order of the corresponding medium and thick plate blank continues to be connected, that is, the order that satisfies U i If the order with value > 0 exists in the candidate order set corresponding to the remaining billet, then the attachment scheme is an optimizable scheme; Step 5.2: Execute the order neighborhood tabu search algorithm to improve the hanging scheme of the medium and thick plate blank; The order neighborhood tabu search algorithm is described as: Step D1: Obtain an order neighborhood for each sub-plate according to the sub-plate of the current mounting solution for the medium and thick plate slab; the orders in the order neighborhood consist of orders that can replace the current sub-plate; If the current sub-plate is the first sub-plate, then the orders in its order neighborhood have three characteristics: 1) they exist in the same set of candidate orders for medium and thick plate surplus as the first sub-plate; 2) their width is not less than that of the second sub-plate; 3) the number of unassembled sub-plates in the order is greater than 0; If the current sub-plate is the last sub-plate, then the orders in its order field have three characteristics: 1) it exists in the same set of candidate orders for medium and thick plate surplus slabs as the first sub-plate; 2) its width is not greater than that of the second-to-last sub-plate; 3) the number of unassembled sub-plates in the order is greater than 0; If the current sub-plate is any sub-plate other than the first or last one, then the orders in its order field have three characteristics: 1) it exists in the same set of candidate orders for medium and thick plate surplus as the first sub-plate; 2) its width range is between the widths of the two adjacent sub-plate orders; 3) the number of unassembled sub-plates in the order is greater than 0; Step D2: Randomly select a sub-board order from the optimizable attachment solutions and replace it according to its order neighborhood. If the objective function satisfies the following equation, the attachment solution replaces the original solution. Otherwise, continue searching the order neighborhood until the order neighborhood search is completed. where x' ijl is the mounting scheme after the replacement of the medium and thick plate blank j; S' j is the trimming amount of the hanging scheme after the replacement of the medium and thick plate blank j; A' j is the head and tail cut amount of the hanging scheme after the replacement of the medium and thick plate blank j; Step 5.3: If a solution with a better optimization goal is found, replace the original optimizable solution. Otherwise, execute step 5.1 and continue to extract the next optimizable solution for improvement.

8. The method for intelligently controlling the connection of medium and thick plate blanks combined with machine learning according to claim 1 is characterized in that: The step 6 specifically includes the following steps: Step 6.1: List the orders in the candidate order set corresponding to the medium and thick plate blanks for which the attachment scheme is generated, and search for medium and thick plate blanks that can be attached to the order based on the tapping marks of these orders; if the found medium and thick plate blank is attached to the order, and the order is not the main order, record the blank tuple "(medium and thick plate blank, order)" consisting of the medium and thick plate blank and the order; after all orders have been traversed, obtain a list of candidate blanks for the medium and thick plate blanks in the expanded scheme, and sort the blanks in the candidate blank list from largest to smallest according to unit weight; Step 6.2: Select the remaining blank tuples from the candidate list of remaining blanks, and reduce the orders contained in the selected tuples by 1 unit, that is, the number of sub-boards of the i-th order is reduced from the original a i Reduced to a i -1, and the reduced part is added to the optimized medium and thick plate blanks until the order i completely disappears from the hanging plan where it is located, or is reduced to the point where it cannot meet the hanging requirements of the medium and thick plate blanks where it is located, and then the next sub-plate with a different order number is replaced and the operation is continued; Step 6.3: Check the attachment status of the current attachment scheme of the optimized medium and thick plate blank; if the attachment success condition is met, terminate the process; if the attachment failure condition is met, restore the original attachment scheme of the reduced-dimensional blank, and replace the next blank tuple for an attempt; if the attachment feasibility condition is met, retain the attachment status of the current optimized medium and thick plate blank, and replace the next blank tuple for an attempt.

9. The method for intelligently controlling the connection of medium and thick plate blanks combined with machine learning according to claim 1 is characterized in that: The step 7 specifically includes the following steps: Step 7.1: Use the internal optimization prediction method to define the set of blank design solutions that need to be optimized. The sub-plate distribution, spot quantity, yield rate, cutting loss, and remaining orders that can be attached to the blank of the current blank design solution are used as input features. Enter a flag of 0 or 1; 0 indicates no optimization is required, and 1 indicates optimization is required. The internal optimization prediction method comprises the following steps: Step B1: The sub-plate distribution vector of the current blank design scheme Spot quantity s, yield rate y, cutting loss u, and remaining orders that can be attached to the remaining billet As a feature vector Where L1 is the number of sub-plates in the design of the remaining plate, L2 is the number of remaining orders, and n3 is the number of eigenvectors that can be obtained; α i =(t i ,h i ,d i ), i=1,...,L1, is the (thickness, width, length) triplet of order i in the medium and thick plate blank design scheme; β k =(t k ,h k ,d k ),k=1,...,L2,is the (thickness, width, length) triplet of the remaining order k; thus the input matrix The output vector is o i ∈O,i=1,...,n3,is the output feature, with a value of 0 or 1; 0 means no optimization is required, 1 means optimization is required; the input matrix and output vector constitute the original data set; Step B2: Divide the original data set into 10 equal data subsets and label them 0-9; starting with data subset 0, the span 1 is the moving step of the data subset, 9 is the capacity of the data subset, and each time the step is moved, a subset containing 9 data portions can be obtained; after 10 moves, 10 subsets containing 9 data portions can be obtained; these 10 subsets are divided into training data subset, test data subset, and validation data subset according to the ratio of 7:2:1, and Θ(χ) is iteratively trained in 10 subsets; Step B3: Extract the data features of all current medium and thick plate blank design schemes, construct the input matrix χ, output the predicted label vector O, and then divide the medium and thick plate blank design schemes that need to be optimized according to the predicted labels; Step 7.2: Select a blank from the set of blank design solutions to be optimized, optimize the first and second-layer sub-panels in the solution, and sequentially transfer the first-layer sub-panels to the second layer of the blank attachment solution. If the yield rate increases after the transfer, update the original solution; otherwise, try the next-layer sub-panel. Step 7.3: Convert the double-layer medium and thick plate blank mounting scheme into a single-layer one; release all the sub-plates in the current mounting scheme, then sort them in descending order according to width, and re-mount them in sequence to form a new mounting scheme; Step 7.4: Check whether the new hanging scheme meets the conditions for successful hanging of medium and thick plate blanks. If so, and the yield rate is improved, update the current hanging scheme; otherwise, keep it as it is.

Images