Intelligent plate assembling method considering camber form of wide and thick plate rolled piece

Abstract

The invention provides an intelligent plate assembling method giving consideration to camber forms of wide and thick plate rolled pieces, and relates to the technical field of steel production. The method comprises the following steps: firstly, setting parameters and decision variables, and establishing an integer programming model by taking maximization of steel plate utilization quantity, maximization of length and width of a single rolled piece and minimization of edge cutting amount as optimization targets; establishing a hierarchical structure of a hybrid hierarchical network MHN, wherein the hierarchical structure comprises a feature layer, a decision layer and an evaluation layer; and finally, establishing a firework algorithm model coding scheme, performing population initialization and fitness evaluation, designing a firework algorithm explosion mode in combination with an MHN fitness evaluation result, continuously performing feature importance and fitness evaluation and adjustment by using the MHN, and outputting an optimal plate assembly scheme when an algorithm termination condition is reached. According to the method, the traditional mode that large fixed edge cutting amount is reserved for plate assembly design due to the camber form of a rolled piece is changed, the purposes that the plate assembly utilization rate is maximum, the yield is maximum and the edge cutting amount is optimal can be effectively achieved, and cost reduction and efficiency improvement of medium and thick plate enterprises are promoted.

Description

Technical Field

[0001] The invention relates to the technical field of steel production, and in particular to an intelligent plate assembly method taking into account the sickle bend shape of wide and thick plate rolled pieces. Background Art

[0002] During the steel plate manufacturing process, the uneven distribution of residual stress inside the steel plate leads to uneven extension of each longitudinal fiber of the rolled piece, resulting in bending phenomena such as sickle bend. Sickle bend refers to the maximum distance between the side of the rolled piece and the straight line connecting the two end points of the measuring part. It mainly occurs at the head and tail of the rolled piece, reflecting the degree of deviation between the edge of the longitudinal side of the rolled piece and the straight line. The rolled piece needs to be trimmed and the head and tail trimmed, that is, the sickle bend is eliminated before it can become a usable finished steel plate. The rolled piece needs to be trimmed first, and then the long rolled piece is cut into multiple short rolled pieces as needed, and then each short rolled piece is trimmed to form a steel plate that meets the order requirements. Due to the influence of the sickle bend of the rolled piece, when designing the plate assembly of the ordered steel plate, in order to ensure that the required steel plate size can be obtained after trimming the full length of the rolled piece, the steel mill usually adopts a large trimming amount for the order assembly design, resulting in a large trimming loss and a low yield rate. In order to make full use of the sickle bend characteristics of the rolled piece and reduce the trimming amount, it is necessary to dynamically match the sub-plate width according to the position of the sub-plate on the rolled piece.

[0003] Document 1 (Intelligent Optimization Model and System for Medium and Thick Plate Assembly and Slab Design, Steel, 2020, 55(04)) introduces the use of non-fixed-length production orders and the uncertainty of slab specifications to establish an adaptive combinatorial optimization model to achieve one-dimensional and two-dimensional, fixed-length and non-fixed-length medium and thick plate assembly and slab integrated optimization design. Document 2 (Development and Application of Intelligent Assembly System for Wide and Thick Plates, Shandong Metallurgy, 2018, 40(03)) introduces a method for forming the optimal assembly scheme through three modules: steel products, residual steel plates, and residual slabs. From the currently available information, no assembly method that takes into account the sickle bend morphology of wide and thick plate rolled pieces has been found. To this end, the present invention proposes an intelligent assembly method that takes into account the sickle bend morphology of wide and thick plate rolled pieces, which can reduce the cutting width of wide and thick plates and effectively improve the yield rate. Summary of the invention

[0004] In order to solve the above technical problems existing in the prior art and realize the maximum utilization rate of plate assembly, the maximum yield rate and the optimal trimming amount to meet the order requirements, the present invention proposes an intelligent plate assembly method that takes into account the sickle bend morphology of wide and thick plate rolled pieces based on the analysis of the sickle bend characteristics of rolled pieces, designs a hybrid algorithm combining the MHN hybrid hierarchical network and the fireworks algorithm, uses random forests to perform feature importance and fitness evaluation analysis in MHN, and designs an improved explosion update method in the fireworks algorithm in combination with the fitness evaluation results, which promotes MHN to dynamically adjust and update weight features and continuously optimize plate assembly decisions, thereby improving the search efficiency and quality of the solution. The technical scheme is as follows:

[0005] An intelligent plate grouping method that takes into account the camber shape of heavy and wide plate rolling pieces, the method comprising:

[0006] S1. Set parameters and decision variables, and establish an integer programming model with the optimization objectives of maximizing the number of steel plates utilized, maximizing the length and width of a single rolling piece, and minimizing the trimming amount.

[0007] S2. Establish the hierarchical structure of a hybrid hierarchical network MHN, including a feature layer, a decision layer, and an evaluation layer. Process the input data through the feature layer and output the feature vectors of the steel plates and rolling pieces. Use the greedy algorithm in the decision layer to generate the plate grouping positions of each steel plate on the camber rolling pieces to obtain an initial solution. Use the evaluation layer to evaluate the feature importance and fitness through the random forest method, and feedback the evaluation results to the decision layer. The decision layer dynamically updates the feature weights and optimizes the plate grouping decision.

[0008] S3. Establish a coding scheme for the fireworks algorithm model, perform population initialization and fitness evaluation, design the explosion mode of the fireworks algorithm in combination with the fitness evaluation results of MHN, and through the iterative operations of explosion, mutation, and selection of the fireworks algorithm, continuously perform feature importance and fitness evaluation and adjustment using MHN to promote search improvement. When the algorithm termination condition is reached, output the optimal plate grouping scheme.

[0009] The parameters in step S1 include the length, width, thickness, area of the rolling piece, as well as the maximum and minimum values of the length and width; the number of orders, the number of steel plates in each order, the total number of steel plates, the length, width, thickness, and area of the steel plates.

[0010] The decision variables include:

[0011] Set the decision variable x ij , if the j -th steel plate of order m is placed on rolling piece i and occupies one row of space, and the steel plate is placed in a single row, then is 1; if it occupies two rows of space and the steel plate is placed in a double row, then is 1; if the j -th steel plate of order m is not placed on rolling piece i, then is 0;

[0012] The integer programming model is specifically:

[0013] (1) Maximize the number of steel plates utilized, requiring the proportion of the number of grouped steel plates to the total number of steel plates to be as large as possible:

[0014]

[0015] Among them, n is the total number of rolling pieces, m is the order number, q m is the number of steel plates in the m -th order, i is the rolling piece serial number, j is the steel plate serial number, and N is the total number of steel plates in all orders;

[0016] (2) The length and width of each rolled piece formed are maximized, including:

[0017] Maximize the length of a single rolled piece:

[0018]

[0019] Maximize the width of a single rolled piece:

[0020]

[0021] Among them,

[0022] is the maximum width of a single steel plate when the steel plates are placed in a single row;

[0023]

[0024] is the maximum value of the sum of the widths of two steel plates placed side by side vertically when the steel plates are placed in two rows,

[0025] The widths of the j-th and j'-th steel plates of order m placed side by side vertically on the i-th rolled piece are: w mji and w mj′ i;

[0026]

[0027] (3) Minimize the amount of trimmed edges, and require the proportion of the area of the grouped steel plates to the area of the rolled piece to be as large as possible:

[0028]

[0029] Among them, r is the number of orders; q m is the number of steel plates in the m-th order; l mj is the length of the j-th steel plate in the m-th order; w mj is the width of the j-th steel plate in the m-th order; T i is the area of the i-th rolled piece;

[0030] The constraint conditions of the integer programming model include:

[0031] The length, width, and thickness limitations of each rolled piece, the position of the steel plates does not exceed the rolled piece area limitation, the thickness difference between the steel plate thickness and the rolled piece of the grouped plates has a range limitation, and each steel plate can only be placed on a rolled piece in one of the single-row or double-row ways. The specific constraint conditions are as follows:

[0032] (1) The length, width, and thickness limitations of each rolled piece:

[0033] l min ≤ l i≤l max

[0034] w min ≤w i ≤w max

[0035] h min ≤h i ≤h max

[0036] Among them, l i is the length of the rolled piece i, l min is the minimum length required for the rolled piece, l max is the maximum length required for the rolled piece, w i is the width of the rolled piece i, w min is the minimum width required for the rolled piece, w max is the maximum width required for the rolled piece, h i is the thickness of the rolled piece i, h min is the minimum thickness required for the rolled piece, h max is the maximum thickness required for the rolled piece.

[0037] (2) The position of the steel plate does not exceed the rolled piece area, that is, it cannot exceed the upper, lower, left, and right boundary arcs of the camber.

[0038] Upper and lower boundary constraints:

[0039] y mji +w mj ≤f(x mji +l mj )(Upper boundary constraint)

[0040] y mji ≥g(x mji )(Lower boundary constraint)

[0041] Left and right boundary constraints:

[0042] x mji ≥x iA (Left boundary constraint)

[0043] x mji +l mj ≤x iB (Right boundary constraint)

[0044] y mji represents the ordinate of the lower left corner of the placement position of the jth steel plate in order m on the rolled piece i, x mji represents the abscissa of the lower left corner of the placement position of the jth steel plate in order m on the rolled piece i,

[0045] f(x) represents the function of the upper boundary of the camber, defining the upper boundary curve of this area and describing the value of the upper boundary of the camber on the vertical axis at any horizontal axis position x;

[0046] g(x) represents the function of the lower boundary of the camber, defining the lower boundary curve of this area and describing the value of the lower boundary of the camber on the vertical axis at any horizontal axis x;

[0047] x iA represents the abscissa of the bending point at the lower left corner of the cambered rolled piece;

[0048] x iB represents the abscissa of the bending point at the lower right corner of the cambered rolled piece.

[0049] (3) There are range limitations on the thickness difference between the steel plate and the rolled piece of the assembled plates.

[0050] Δh imj ≤τ (τ is a constant determined according to the requirements of each order)

[0051]

[0052] Δh imj is the thickness difference between the steel plate and the rolled piece, h mj is the thickness of the j-th steel plate of the m-th order, h i represents the thickness of the i-th rolled piece.

[0053] (4) Each steel plate can be placed on a rolled piece in at most one of the single-row or double-row ways.

[0054]

[0055] The input data in step S2 includes: the data characteristics of the steel plate and the rolled piece;

[0056] The feature vectors of the steel plate and the rolled piece are:

[0057] F = [f1, f2,......, f k

[0058] f k includes the data characteristics of the steel plate and the rolled piece, representing the length, width, thickness of the steel plate, the number of steel plates combined on a rolled piece, and the length, width and thickness of the rolled piece.

[0059] The feature layer in step S2 is used for preprocessing and feature extraction of the input data, and the features are standardized using the following formula to ensure the consistency of the dimensions of different features:

[0060]

[0061] Among them, ​

[0062] F′ ij : The eigenvalue after standardization;

[0063] F ij : The original eigenvalue;

[0064] μ j : The mean value of feature j;

[0065] σ j： The standard deviation of feature j.

[0066] Extract and output the feature vector to provide support for the evaluation of the adaptability and matching degree of feature matching.

[0067] In step S2, the decision-making layer is responsible for generating the initial scheme and optimizing the feature weights according to the adaptability evaluation to promote the improvement of the search efficiency of the subsequent fireworks algorithm. The decision-making layer designs two main sub-modules:

[0068] The initial scheme generation module and the feature learning and optimization module,

[0069] The initial scheme generation module uses the greedy algorithm to generate the initial splicing scheme;

[0070] The feature learning and optimization module dynamically adjusts the feature weights according to the adaptability evaluation. The adjustment method is as follows:

[0071] Based on the feature importance and adaptability feedback from the evaluation layer, dynamically update the feature weights. The update formula is as follows:

[0072]

[0073] is the updated adaptability;

[0074] is the current adaptability;

[0075] α is the weight factor that controls the balance between the old and new adaptabilities, and α ∈ (0, 1);

[0076] Score(x ij ) is the result of re-evaluating the adaptability based on the performance of the current solution, calculated according to the adaptability calculation formula.

[0077] In step S2, the evaluation layer uses the random forest method to evaluate the feature importance. Specifically:

[0078] Construct the dataset feature matrix F, where each row represents the relevant features of a rolled piece and the steel plates combined on it, and each column represents a feature of the combination on the rolled piece.

[0079] F ijRepresents the numerical values related to the characteristics of the i-th rolled piece and the steel plates combined to form the rolled piece.

[0080] F ij =[L iP W iP H iP Q iP L iS1 W iS1 H iS1 P iS1

[0081] Wherein,

[0082] L iP W iP H iP Q iP respectively represent the length, width, thickness of the i-th rolled piece and the total number of steel plates combined;

[0083] L iS1 W iS1 H iS1 p iS1 respectively represent the length, width, thickness and position of each steel plate forming the i-th rolled piece;

[0084] The target variable Y is set to the fitness of length and width.

[0085] Construct a decision tree and train the decision tree:

[0086] Construct a decision tree: Divide the historical data into a training set and a test set. Randomly draw N samples with replacement from the training set. Each sample represents a combination plan of a rolled piece and multiple steel plates, and these plans form B different subsets. For each subset, construct a decision tree.

[0087] Train the decision tree: Randomly select m features at each node. Traverse all available features at the current node. Determine the splitting plan of the node by calculating the impurity reduction, and measure the importance of the feature by the sum of the Gini impurity reduction.

[0088] Specifically,

[0089] Randomly select m features at each node to determine the best split; at the current node, traverse all available features, calculate the potential split points of each feature; for each feature and each potential split point, calculate the Gini impurity of the current node, calculate the weighted average Gini impurity of the split dataset, and finally calculate the impurity reduction; the importance I k of feature X k is measured by the sum of the Gini impurity reduction it causes in all trees. The formula is as follows:

[0090] ​

[0091] T b All nodes for node splitting using feature X; b is the b-th tree; B is the number of subsets; t is a specific node; ΔI k is the impurity reduction caused by feature X kt in a specific node t of a certain tree b; k

[0092] ΔI kt = Gini(D) - Ginisplit;

[0093]

[0094] where D is the data set of the current node,

[0095] D1 and D2 are the two split data sets for the values of feature X k respectively,

[0096] Gini(D) is the Gini impurity of the current node,

[0097] Gini(D1) and Gini(D2) are the Gini impurities after splitting,

[0098] T is the number of features,

[0099] p i is the probability of feature i in data set D, that is, the ratio of the number of samples of feature i to the total number of samples;

[0100] Ginisplit is the weighted average Gini impurity after splitting,

[0101] F represents the total number of feature sets used at the current node, and F1 and F2 respectively represent the number of features in the two subsets after splitting of the current feature set.

[0102] Through the above calculation, the split point with the largest impurity reduction is used as the splitting scheme for the current node.

[0103] Each tree reaches the leaf node through node splitting; the feature value determines the path of the sample in the tree, which in turn affects the sample set reaching each leaf node, and ultimately affects the predicted value of the leaf node; calculate the predicted value of the leaf node and the actual target value, that is, the length and width of the rolled piece, recursively construct the decision tree, and for each newly created subset, recursively repeat the above process until the stopping condition is reached, and feedback the calculated feature importance result to the decision-making layer.

[0104] The predicted value of the leaf node is output according to the following formula:

[0105] ​

[0106] p is the number of samples reaching this leaf node, indicating how many samples within this leaf node contribute to the predicted value;

[0107] y i is the actual target value of the i-th sample.

[0108] T b (X k ) is the predicted value of the b-th tree for the input feature X k ;

[0109] In the fitness evaluation process in step S2, it is as follows:

[0110] represents the fitness of the steel plate combination for the i-th rolled piece, and the fitness is represented by the length and width of the rolled piece; according to the objective function, within the constraint range, the larger the length and width of a single rolled piece, the better, that is, the larger the fitness value. The fitness calculation formula is as follows:

[0111]

[0112] Where:

[0113]

[0114] w l and w w are the weight coefficients of the length and width. Set w l to 80%, and w w to 20%;

[0115] l mj is the length of the j-th steel plate in the m-th order;

[0116] w mj is the width of the j-th steel plate in the m-th order;

[0117] is the sum of the widths of the two steel plates placed side by side up and down when the steel plates are placed in a double row.

[0118] In step S3, a multi-dimensional array is used for the encoding of the fireworks algorithm. The solution output by the MHN decision layer is used as the initial population, and fitness evaluation is performed. Then, iterative operations are carried out according to the explosion, mutation, and selection processes of the fireworks algorithm. When the maximum number of iterations is reached, the algorithm stops, and the individual with the highest fitness in the current population is output as the optimal plate combination plan.

[0119] The fitness evaluation includes:

[0120] Encoding is performed using a multi-dimensional array, and each element of the array consists of the following information:

[0121] Solution = {O ij , S r , X, Y}

[0122] O ij : The j-th steel plate of the i-th order;

[0123] S r : The rolled piece number

[0124] X, Y: The coordinate positions of the steel plate on the rolled piece;

[0125] Use the combination scheme generated by the MHN decision layer using the greedy algorithm as the initial population of the fireworks algorithm. Take the maximum utilization quantity of the steel plate and the minimum trimming amount as the fitness value, calculate the fitness values of N individuals, and select the maximum value from the N fitness values, that is, the larger the fitness value, the better.

[0126] The fitness represents the utilization quantity of the steel plate and the trimming amount situation, and the formula is as follows:

[0127]

[0128] Where: W u and W e are the weight coefficients;

[0129] f(x i ) is the fitness;

[0130] n is the total number of rolled pieces, m is the order number, i is the rolled piece serial number, j is the steel plate serial number, r is the order quantity, q m is the number of steel plates in the m-th order, is the decision variable for single-row placement of the steel plate, is the decision variable for double-row placement of the steel plate, l mj is the length of the j-th steel plate in the m-th order, w mj is the width of the j-th steel plate in the m-th order, T i is the area of the i-th rolled piece, and N is the total number of steel plates of all orders.

[0131] In the explosion operation of the fireworks algorithm in step S3, combined with the evaluation of the fitness of the matching between the steel plate and the rolled piece by MHN, design the fitness coefficient, so that fireworks with high fitness are set with a relatively small explosion radius and generate relatively more sparks, while fireworks with poor fitness values are set with a relatively large explosion radius and generate relatively fewer sparks. The explosion intensity formula is as follows:

[0132]

[0133] The explosion amplitude formula is as follows:

[0134]

[0135] Where: S i is the explosion intensity, A i is the explosion amplitude, s' and a' are constants respectively, s' is the factor representing the explosion intensity, and its value range is 10 to 200, a' is the factor representing the explosion amplitude, and its value range is 10% to 50%; f(x i ) is the fitness value of the current firework individual; f max = max(f(x i )) and f min = min(f(x i )) are the maximum and minimum values among the fitness values of K fireworks; ε is a very small constant used to ensure that the denominator is not zero; K is the number of fireworks; β is the weight controlling the influence of fitness on the explosion intensity, and its value range is [0,1]; is the average value of fitness; p i is the current fitness value;

[0136] When the fitness is high, that is, A i decreases, resulting in a smaller explosion radius;

[0137] When the fitness is low, that is, A i increases, resulting in a larger explosion radius.

[0138] In order to avoid the problem that excellent fireworks have too strong explosion intensity, too many sparks are generated, wasting computational resources, while poor fireworks have too weak explosion and too few generated sparks, thus reducing the population diversity, the following formula is used to control it:

[0139]

[0140] Where, round(s i ) is the rounding function, which rounds by the method of rounding up or down, and a and b are given constants.

[0141] s i is the explosion intensity, that is, the number of sparks generated by the explosion of the i-th firework, and s' represents the reference value for constraining and adjusting the number of sparks s i .

[0142] For each dimension of the firework individual, a new spark is randomly selected within the range of the explosion amplitude and a displacement operation is performed according to the following formula, represents the position of the i-th firework in the k-th dimension, rand(-1,1) represents a uniform random number within -1 to 1, and A i is the explosion amplitude of the i-th firework.

[0143] When the k dimension of the i-th firework exceeds the boundary, the following calculation formula is used to perform a mapping operation to pull the firework particles that exceed the feasible solution space back into the solution space.

[0144]

[0145] Among them, x k max 、x k min are the maximum and minimum boundaries of the i-th firework in the k-th dimension respectively.

[0146] Then, Gaussian mutation operation is performed to select fireworks individuals, and individuals are selected to be retained for the next generation using the elite strategy + random selection method. MHN is used to evaluate the importance and fitness of features, and feedback is given to the decision layer to promote search improvement. When the maximum number of iterations is reached, the algorithm is stopped and the individual with the highest fitness in the current population is output as the optimal board assembly solution.

[0147] The beneficial effects brought about by the technical solution provided by the embodiment of the present invention include at least:

[0148] The present invention establishes an integer programming model, proposes a hybrid algorithm combining the MHN hybrid hierarchical network and the fireworks algorithm to solve the solution, designs an improved explosion update method, uses MHN to dynamically adjust and update weight characteristics and continuously optimize plate assembly decisions, improves the search efficiency and quality of the solution, and realizes intelligent plate assembly of rolled pieces of different specifications and different bending degrees, effectively promotes the maximum utilization rate of order plate assembly, the maximum yield rate, and the optimal trimming amount, and has a significant promoting effect on production optimization and efficiency improvement of steel enterprises. BRIEF DESCRIPTION OF THE DRAWINGS

[0149] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0150] Figure 1 This is a process flow chart of an intelligent plate assembly method taking into account the sickle shape of wide and thick plate rolled pieces provided by an embodiment of the present invention;

[0151] Figure 2 Schematic diagram of the length and width of a sickle bending workpiece in an embodiment of the present invention;

[0152] Figure 3 It is a schematic diagram for comparing trimming with a fixed trimming amount and trimming with a sickle bending workpiece provided by an embodiment of the present invention;

[0153] Figure 4It is a functional diagram of the hybrid hierarchical network MHN structure provided by an embodiment of the present invention. Detailed implementation manners

[0154] The technical solutions in the present invention will be described below with reference to the accompanying drawings.

[0155] In the embodiments of the present invention, words such as "exemplarily" and "for example" are used to represent examples, illustrations or explanations. Any embodiment or design solution described as an "example" in the present invention should not be construed as being more preferred or more advantageous than other embodiments or design solutions. Exactly speaking, the use of the word "example" is intended to present concepts in a specific manner. In addition, in the embodiments of the present invention, the meaning expressed by "and / or" can be both, or either one of the two can be selected.

[0156] In the embodiments of the present invention, sometimes a subscript such as W1 may be written in a non-subscript form such as W1. When the difference is not emphasized, the meanings to be expressed are the same.

[0157] To make the technical problems, technical solutions and advantages to be solved by the present invention clearer, the following will be described in detail with reference to the accompanying drawings and specific embodiments.

[0158] The embodiments of the present invention provide an intelligent plate grouping method that takes into account the camber shape of heavy plate rolling pieces. As Figure 1 shown in the flow of the intelligent plate grouping method that takes into account the camber shape of heavy plate rolling pieces, the method may include the following steps:

[0159] S1. Set parameters and decision variables, establish an integer programming model with the optimization objectives of maximizing the number of steel plates used, maximizing the length and width of a single rolling piece, and minimizing the trimming amount;

[0160] S2. Establish the hierarchical structure of the hybrid hierarchical network MHN, including a feature layer, a decision layer and an evaluation layer. Process the input data through the feature layer and output the feature vectors of the steel plates and rolling pieces; use the greedy algorithm in the decision layer to generate the plate grouping positions of each steel plate in the camber rolling pieces to obtain an initial solution; use the evaluation layer to evaluate the feature importance and fitness through the random forest method, and feedback the evaluation results to the decision layer. The decision layer dynamically updates the feature weights and optimizes the plate grouping decision;

[0161] S3. Establish a coding scheme for the fireworks algorithm model, perform population initialization and fitness evaluation, design the explosion mode of the fireworks algorithm in combination with the MHN fitness evaluation results, and through the explosion, mutation and selection iteration operations of the fireworks algorithm, continuously perform feature importance and fitness evaluation and adjustment using the MHN to promote search improvement, and output the optimal plate grouping scheme when the algorithm termination condition is reached.

[0162] The following will be described with reference to specific embodiments.

[0163] S1: Set parameters and decision variables, and establish an integer programming model, including:

[0164] Parameters of the rolled piece

[0165] n: Quantity of the rolled pieces

[0166] l i : Length of the i-th rolled piece

[0167] w i : Width of the i-th rolled piece

[0168] h i : Thickness of the i-th rolled piece

[0169] S i : Area of the i-th rolled piece

[0170] l max : Maximum length of the rolled piece

[0171] l min : Minimum length of the rolled piece

[0172] w max : Maximum width of the rolled piece

[0173] w min : Minimum width of the rolled piece

[0174] h max : Maximum thickness of the rolled piece

[0175] h min : Minimum thickness of the rolled piece

[0176] The length of the rolled piece is defined as the distance between the maximum bending points of the two side arcs, that is, the length of A1A2 as shown in Figure 2 shown, and the width of the rolled piece is defined as the distance between the maximum bending points of the camber, that is, the length of C1C2 as shown in Figure 2 shown.

[0177] Parameters of the ordered steel plates:

[0178] r: Quantity of orders

[0179] q m : Quantity of steel plates in the m-th order

[0180] N: Total quantity of steel plates in all orders

[0181] l mj : Length of the j-th steel plate in the m-th order

[0182] w mj : Width of the j-th steel plate in the m-th order

[0183] hmj : Thickness of the j-th steel plate in the m-th order

[0184] A mj : Area of the j-th steel plate in the m-th order

[0185] Decision variable:

[0186] As an implementation manner of the embodiment of the present invention, different from the traditional method of grouping steel plates for rectangular rolled pieces after a fixed trimming amount is completed, instead, the steel plates are directly grouped for the cambered rolled pieces, so that the trimming amount can closely change with the change of the short rolled piece specifications (as Figure 3 shown), the rolled pieces can group wider order steel plates, and the trimming amount can be greatly optimized. The steel plates are arranged on the rolled pieces with the length of the steel plates parallel to the horizontal direction of the rolled pieces. The steel plates can occupy one row of space on the rolled pieces (i.e., single-row arrangement) or two rows of space (i.e., double-row arrangement), as Figure 3 shown, the steel plates P5 and P6 occupy one row of space, and the steel plates P1, P2, P3, and P4 occupy two rows of space.

[0187] Set the decision variable x ij , if the j-th steel plate of order m is placed on the rolled piece i and occupies one row of space, then is 1, if it occupies two rows of space, then is 1, if the steel plate j of order m is not placed on the rolled piece i, then let be 0.

[0188] Objective function:

[0189] As an implementation manner of the embodiment of the present invention, the optimization objective is to maximize the number of steel plates used, that is, to maximize the utilization rate of steel plate grouping, maximize the length and width of each rolled piece formed, and minimize the trimming amount. The optimization objective function is as follows:

[0190] (1) Maximize the number of steel plates used, and require the proportion of the number of grouped steel plates to the total number of steel plates to be as large as possible:

[0191]

[0192] Among them, n is the total number of rolled pieces, m is the order number, q m is the number of steel plates in the m-th order, i is the rolled piece serial number, j is the steel plate serial number, and N is the total number of steel plates in all orders;

[0193] (2) Maximize the length and width of each formed rolled piece,

[0194] The length of the rolled piece composed depends on the sum of the lengths of the steel plates on the rolled piece, which is equal to the sum of the lengths of all the steel plates occupying a single row plus half of the sum of the lengths of the steel plates occupying the double-row space; the width depends on the larger value between the width of the widest steel plate occupying a single-row space and the maximum of the sum of the widths of two steel plates occupying the double-row space.

[0195] Maximize the length of a single rolled piece:

[0196]

[0197] Maximize the width of a single rolled piece:

[0198]

[0199] Among them,

[0200] is the maximum width of a single steel plate when the steel plates are placed in a single row;

[0201]

[0202] is the maximum value of the sum of the widths of two vertically juxtaposed steel plates when the steel plates are placed in a double row,

[0203] The widths of the j-th and j'-th steel plates of order m placed vertically side by side on the i-th rolled piece are: w mji and w mj′ i;

[0204]

[0205] (3) Minimize the trimming amount, and require that the proportion of the area of the steel plate group to the area of the rolled piece is as large as possible:

[0206]

[0207] Among them, r is the number of orders; q m is the number of steel plates in the m-th order; l mj is the length of the j-th steel plate in the m-th order; w mj is the width of the j-th steel plate in the m-th order; T i is the area of the i-th rolled piece;

[0208] The constraint conditions of the integer programming model include:

[0209] The length, width, and thickness limits of each rolled piece, the position of the steel plate does not exceed the rolled piece area limit, the thickness difference between the steel plate thickness and the rolled piece of the grouped plates has a range limit, and each steel plate can only be placed on a rolled piece in one of the single-row or double-row ways, specifically:

[0210] Length, width, and thickness limits of each rolled piece

[0211] l min ≤l i ≤l max

[0212] w min ≤w i ≤w max

[0213] h min ≤h i ≤h max

[0214] The position of the steel plate does not exceed the rolled piece area, that is, it cannot exceed the upper, lower, left, and right boundary arcs of the camber.

[0215] Upper and lower boundary constraints:

[0216] y mji +w mj ≤f(x mji +l mj )(Upper boundary constraint)

[0217] y mji ≥g(x mji )(Lower boundary constraint)

[0218] Left and right boundary constraints:

[0219] x mji ≥x iA (Left boundary constraint)

[0220] x mji +l mj ≤x iB (Right boundary constraint)

[0221] y mji represents the ordinate of the lower left corner of the placement position of the j-th steel plate in order m on the rolled piece i;

[0222] x mji represents the abscissa of the lower left corner of the placement position of the j-th steel plate in order m on the rolled piece i;

[0223] f(x) represents the function of the upper boundary of the camber, defining the upper boundary curve of this area and describing the value of the upper boundary of the camber on the vertical axis at any horizontal axis position x;

[0224] g(x) represents the function of the lower boundary of the camber, defining the lower boundary curve of this area and describing the value of the lower boundary of the camber on the vertical axis at any horizontal axis x;

[0225] x iA represents the abscissa of the lower left corner bending point of the camber rolled piece

[0226] x iB Represents the abscissa of the bending point at the lower right corner of the sickle-shaped rolled piece.

[0227] There are range limitations for the thickness difference between the steel plate and the rolled piece of the assembled plates:

[0228] Δh imj ≤τ (τ is a constant)

[0229]

[0230] Each steel plate can be placed on a rolled piece in at most one of the single-row or double-row ways:

[0231]

[0232] S2: Establish the hierarchical structure of the Mixed Hierarchical Network MHN (see the structure function diagram of the Mixed Hierarchical Network MHN in Figure 4 ), including the feature layer, decision layer, and evaluation layer, specifically as follows:

[0233] Feature layer:

[0234] The feature layer is used for preprocessing of input data and feature extraction. The features are standardized using the following formula to ensure consistent dimensions for different features:

[0235]

[0236] F′ ij : The standardized feature value,

[0237] F ij : The original feature value,

[0238] μ j : The mean of feature j,

[0239] σ j： The standard deviation of feature j.

[0240] Extract and output the feature vector to provide support for the evaluation of the adaptability and matching degree of feature matching.

[0241] The feature vectors of the steel plate and the rolled piece are:

[0242] F = [f1, f2,......, f k

[0243] f k Contains the data features of the steel plate and the rolled piece, representing the length, width, thickness of the steel plate, the number of steel plates combined on a rolled piece, and the length, width, and thickness of the rolled piece. Specifically:

[0244] Steel plate length: l​mj

[0245] Width of steel plate: w mj

[0246] Thickness of steel plate: h mj

[0247] Position of steel plate: p mj

[0248] Number of steel plates combined on one rolled piece: q i

[0249] Length of rolled piece: l i

[0250] Width of rolled piece: w i

[0251] Thickness of rolled piece: h i

[0252] Decision-making layer:

[0253] The decision-making layer is responsible for generating the initial scheme, evaluating and optimizing the feature weights according to the fitness, and promoting the improvement of the search efficiency of the subsequent fireworks algorithm. Two main sub-modules are designed in the decision-making layer:

[0254] Initial scheme generation module: Generate the initial splicing scheme using the greedy algorithm.

[0255] Feature learning and optimization module: Dynamically adjust the feature weights according to the fitness evaluation.

[0256] As an implementation manner of the embodiment of the present invention, the steps of generating the initial splicing scheme using the greedy algorithm are as follows:

[0257] Arrange the steel plates in descending order of area, and select the largest assumed slab size, that is, assume that the size of each rolled piece is the maximum limit value:

[0258] l i = l max , w i = w max , h i = h max

[0259] First, start combining from the largest steel plate on the rolled piece with the largest size, and check whether the steel plates can be combined according to the constraint conditions. If the steel plates are successfully combined, update the size of the assumed rolled piece, reduce the remaining available length and width, and the updated size of the rolled piece:

[0260] l i ‘ = l i - l mj (Update the remaining length)

[0261] w i ‘= w i -w mj (Update the remaining width)

[0262] Select steel plates for combination in descending order of area one by one until there are no more steel plates that can be combined on this rolled piece. Then select a new steel plate and rolled piece for combination until there are no more steel plates that can be combined on any rolled piece, and output a combination plan of steel plates and rolled pieces.

[0263] Arrange the steel plates in descending order of area, and run the greedy algorithm by selecting one steel plate multiple times. Finally, generate N combination plans.

[0264] As an implementation manner of the embodiment of the present invention, the decision-making layer dynamically adjusts the feature weights according to the fitness evaluation as follows:

[0265] Based on the feature importance and fitness situation feedback by the evaluation layer, dynamically update the feature weights. The update formula is as follows:

[0266]

[0267] is the updated fitness;

[0268] is the current fitness;

[0269] α is the weight factor, which controls the balance between the old and new fitnesses, and α ∈ (0, 1);

[0270] Score(x ij ) is the result of re-evaluating the fitness based on the performance of the current solution, and is calculated based on the fitness formula.

[0271] MHN identifies the features that have the greatest impact on the size of the rolled piece composed of steel plates through evaluation, and enhances their weights, while reducing the weights of the features that perform poorly in the current solution, thereby improving the fitness calculation and enhancing the accuracy of matching. Through dynamic adjustment, MHN can encourage the combined use of specific features, dynamically identify the importance of the interaction of multiple features, adjust the weights to reflect their comprehensive impact on the matching result, and thus guide the feature combination. By dynamically adjusting the feature weights, MHN can exclude unimportant feature combinations in the subsequent solution generation, narrow the search space, and thus improve the quality and efficiency of the overall solution.

[0272] Evaluation layer:

[0273] Responsible for evaluating special importance evaluation and fitness evaluation. By evaluating the importance of various features in the current solution, it provides support for the decision-making layer to dynamically adjust feature weights, and promotes the improvement of the search efficiency of the solution space in the subsequent fireworks algorithm. During the steel plate grouping process, check the length and width of the rolled pieces in each group. Through fitness evaluation, it provides support for improving the explosion mode of the fireworks algorithm and promotes the achievement of multiple optimization goals.

[0274] As an implementation manner of the embodiment of the present invention, the random forest method is used to dynamically evaluate the influence degree of each feature on the current solution, and the evaluation result is fed back to the decision-making layer to continuously update the feature weights, and the feature extraction and matching strategies are optimized in real time to improve the flexibility and efficiency of the entire optimization process. The evaluation method of the random forest method is as follows:

[0275] Evaluate feature importance using the random forest method.

[0276] Construct a dataset feature matrix F, where each row represents the relevant features of a rolled piece and the steel plates combined on it, and each column represents a feature of the combination on the rolled piece.

[0277] F ij Represents the numerical value related to the features of the i-th rolled piece and the steel plates combined to form the rolled piece.

[0278] F ij =[L iP W iP H iP Q iP L iS1 W iS1 H iS1 P iS1

[0279] Among them,

[0280] L iP W iP H iP Q iP Respectively represent the length, width, thickness of the i-th rolled piece and the total number of steel plates in the combination;

[0281] L iS1 W iS1 H iS1 p iS1 Respectively represent the length, width, thickness and position of each steel plate forming the i-th rolled piece;

[0282] Set the target variable Y as the fitness of length and width.

[0283] Construct a decision tree:

[0284] ​Divide the historical data into a training set and a test set. Randomly draw N samples with replacement from the training set. Each sample represents a combination plan of a rolled piece and multiple steel plates, and these plans form B different subsets. For each subset, construct a decision tree.

[0285] Training the decision tree:

[0286] Randomly select m features at each node to determine the best split; at the current node, traverse all available features, calculate the potential split points for each feature; for each feature and each potential split point, calculate the Gini impurity of the current node, calculate the weighted average Gini impurity of the split dataset, and finally calculate the impurity reduction.

[0287] Let the dataset at the current node be D, and the feature be X k The two split datasets for the values of X are D1 and D2, corresponding to the feature values less than or greater than a certain split point t. Use the Gini index measurement formula to calculate the Gini impurity G ini (D) of the current node, the Gini impurities G ini (D1) and G ini (D2) after splitting, and the formulas are as follows:

[0288]

[0289] The weighted average Gini impurity after splitting is as follows:

[0290]

[0291] The feature X k The impurity reduction caused by it at a specific node t in a certain tree b is:

[0292] ΔI kt = Gini(D) - Ginisplit;

[0293] Through the above calculations, take the split point with the largest impurity reduction as the split plan for the current node.

[0294] The feature X k The importance I k of the feature X is measured by the sum of the Gini impurity reductions it causes in all trees.

[0295]

[0296] T b is all the nodes that use the feature X k for node splitting, that is, all the nodes in the b-th tree; b is a certain tree; B is the number of subsets; t is a specific node; ΔI kt is the feature X kThe impurity reduction caused in a specific node t of a certain tree b;

[0297] D is the dataset of the current node,

[0298] D1 and D2 are the two split datasets of the feature X k for the two split values,

[0299] Gini(D) is the Gini impurity of the current node,

[0300] Gini(D1) and Gini(D2) are the Gini impurities after splitting,

[0301] T is the number of features,

[0302] p i is the probability of feature i in the dataset D,

[0303] Ginisplit is the weighted average Gini impurity after splitting,

[0304] F represents the total number of feature sets used at the current node, and F1 and F2 represent the number of features in the two subsets after splitting the current feature set respectively.

[0305] Each tree reaches a leaf node through node splitting; the feature values determine the path of the sample in the tree, which in turn affects the sample set reaching each leaf node and ultimately affects the predicted value of the leaf node; the predicted value of the leaf node is output according to the following formula:

[0306]

[0307] P is the number of samples reaching this leaf node, indicating how many samples in this leaf node contribute to the predicted value.

[0308] y i is the actual target value of the i-th sample.

[0309] T b (X k ) is the predicted value of the b-th tree for the input feature X k of.

[0310] Recursively construct a decision tree. For each newly created subset, recursively repeat the above process until the stopping condition is reached, and feedback the calculated feature importance results to the decision layer.

[0311] Fitness evaluation:

[0312] It represents the fitness of the steel plate combination for the i-th rolled piece, and the fitness is represented by the length and width of the rolled piece. According to the objective function, within the constraint range, the larger the length and width of a single rolled piece, the better, that is, the larger the fitness value. The fitness calculation formula is as follows:

[0313]

[0314] Where:

[0315]

[0316] w l and w w are the weight coefficients of the length and width. Set w l to 80%, and w w to 20%;

[0317] l mj is the length of the j-th steel plate in the m-th order;

[0318] w mj is the width of the j-th steel plate in the m-th order;

[0319] is the sum of the widths of the two steel plates placed side by side vertically when the steel plates are placed in a double row.

[0320] S3: Use a multi-dimensional array for the coding of the fireworks algorithm. The solution output by the MHN decision layer is used as the initial population, and fitness evaluation is performed. Then, iterative operations are carried out according to the explosion, mutation, and selection processes of the fireworks algorithm. When the maximum number of iterations is reached, the algorithm stops, and the individual with the highest fitness in the current population is output as the optimal plate combination scheme.

[0321] Coding

[0322] Use a multi-dimensional array for coding. Each element of the array consists of the following information:

[0323] Order number: It represents which order the currently placed steel plate belongs to.

[0324] Steel plate number: It represents the specific steel plate number.

[0325] Rolled piece number: It represents on which rolled piece the steel plate is placed.

[0326] Placement position: It is represented by the coordinates (X, Y) for the placement position of the steel plate on the rolled piece.

[0327] Each steel plate in each solution is represented by a tuple or array in the following format:

[0328] Solution = {O ij , S r , X, Y}

[0329] O ij : The j-th steel plate of the i-th order

[0330] S r : The rolling piece number

[0331] X, Y: The coordinate position of the steel plate on the rolling piece

[0332] Population initialization

[0333] As an implementation manner of the embodiment of the present invention, after encoding the fireworks algorithm, the combination scheme generated by the MHN decision layer using the greedy algorithm is used as the initial population of the fireworks algorithm.

[0334] Fitness evaluation

[0335] Taking the utilization quantity of the steel plate and the trimming amount as the fitness value, calculate the fitness values of N individuals, and among the N fitness values, the larger the fitness value, the better.

[0336] The fitness represents the utilization quantity of the steel plate and the trimming amount situation, and the formula is as follows:

[0337]

[0338] Where: W u And W e Are weight coefficients;

[0339] f(x i ) is the fitness;

[0340] n is the total number of rolling pieces, m is the order number, i is the rolling piece serial number, j is the steel plate serial number, r is the order quantity, q m Is the quantity of steel plates in the m-th order, Is the decision variable for single-row placement of the steel plate, Is the decision variable for double-row placement of the steel plate, l mj Is the length of the j-th steel plate in the m-th order, w mj Is the width of the j-th steel plate in the m-th order, T i Is the area of the i-th rolling piece, and N is the total number of steel plates of all orders.

[0341] Explosion displacement

[0342] Combining the explosion operation of the fireworks algorithm with the evaluation of the matching fitness of the steel plate and the rolling piece by MHN, design the fitness coefficient, so that the fireworks with high fitness are set with a relatively small explosion radius and generate relatively more sparks, while the fireworks with poor fitness values are set with a relatively large explosion radius and generate relatively fewer sparks. The explosion intensity formula is as follows:

[0343]

[0344] The explosion amplitude formula is as follows:

[0345]

[0346] Where: S i is the explosion intensity, A i is the explosion amplitude, s′ and a′ are constants, f(x i ) is the individual fitness value of the current fireworks; f max =max(f(x i )) and f min =min(f(x i ))

[0347] is the maximum and minimum value of the K fireworks fitness values; ε is a very small constant used to ensure that the denominator is not 0; the value of ε is usually 10 -6 to 10 -10 between; K is the number of fireworks; β is the weight of the influence of control fitness on explosion intensity; is the average value of fitness; p i is the current fitness value.

[0348] When the degree of fit is high, When A i Reduced, resulting in a smaller blast radius;

[0349] When the fitness is low, When A i Increases, resulting in a larger blast radius.

[0350] In order to avoid the problem that good fireworks explode too strongly, produce too many fireworks, waste calculations, and bad fireworks explode too weakly, produce too few sparks, and reduce population diversity, the following formula is used to control them:

[0351]

[0352] round(s i ) is a rounding function, which rounds to the nearest integer. a and b are given constants.

[0353] For each dimension of the fireworks individual, within the range of the explosion amplitude, a new spark is generated by randomly selecting and displacing it according to the following formula. It represents the position of the i-th firework in the k-th dimension, and rand(-1,1) represents a uniform random number between -1 and 1.

[0354] When the k dimension of the i-th firework exceeds the boundary, the following calculation formula is used to perform a mapping operation to pull the firework particles that exceed the feasible solution space back into the solution space.

[0355]

[0356] where x k max and x k min are the maximum and minimum boundaries of the i-th firework in the k-th dimension, respectively.

[0357] Mutation operation

[0358] Randomly select a firework individual and perform Gaussian mutation operation on its dimension to generate special sparks as follows:

[0359]

[0360] is the position of the i-th firework in the k-th dimension, and g is a random number subject to Gaussian distribution with both mean and variance equal to 1, i.e., g ~ N(-1, 1)

[0361] Similarly, if the k-th dimension of the i-th firework exceeds the boundary, the above formula is used for mapping operation.

[0362] Selection operation

[0363] Merge the sparks after explosion displacement operation, the sparks generated after mutation operation with the original firework population to form a candidate set V. Use the elitist strategy + random selection method to select N individuals to be retained in the next generation, that is, directly retain the optimal individual in the next generation, and then use the roulette wheel method to select the remaining N - 1 individuals from the candidate set. The probability formula for selecting a certain particle is as follows:

[0364]

[0365] D(x i ) is the sum of distances between the current individual x i and all particles in the candidate set V except x i .

[0366] MHN evaluates the feature importance and fitness, and feeds back to the decision-making layer to promote search improvement.

[0367] Algorithm termination

[0368] When the maximum number of iterations is reached, stop the algorithm and output the individual with the highest fitness in the current population as the optimal template solution; otherwise, return to step S3 for execution.

[0369] As described above, it is only the specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention can easily think of changes or substitutions, which should all be covered within the protection scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claimed rights.

Claims

1. An intelligent plate grouping method that takes into account the camber shape of heavy and wide plate rolling pieces, characterized in that, The method includes: S1. Set parameters and decision variables, and establish an integer programming model with the optimization objectives of maximizing the number of steel plates utilized, maximizing the length and width of a single rolled piece, and minimizing the trimming amount. S2. Establish the hierarchical structure of the hybrid hierarchical network MHN, including the feature layer, decision layer, and evaluation layer. Process the input data through the feature layer and output the feature vectors of the steel plates and rolled pieces. Use the greedy algorithm in the decision layer to generate the grouping positions of each steel plate on the cambered rolled pieces to obtain the initial solution. Use the evaluation layer to evaluate the feature importance and fitness through the random forest method, and feedback the evaluation results to the decision layer. The decision layer dynamically updates the feature weights and optimizes the grouping decision. S3. Establish the coding scheme of the fireworks algorithm model, perform population initialization and fitness evaluation, design the explosion mode of the fireworks algorithm in combination with the fitness evaluation results of MHN, and through the iterative operations of explosion, mutation, and selection of the fireworks algorithm, continuously evaluate and adjust the feature importance and fitness using MHN to promote search improvement. When the algorithm termination condition is reached, output the optimal grouping scheme.

2. The intelligent plate grouping method for considering the camber shape of heavy plate rolling pieces according to claim 1, wherein, The parameters in step S1 include the length, width, thickness, area of the rolled piece, as well as the maximum and minimum values of the length and width. The number of orders, the number of steel plates in each order, the total number of steel plates, the length, width, thickness, and area of the steel plates. The decision variables include: Set the decision variable x ij , if the j-th steel plate of order m is placed on the rolled piece i and occupies one row of space, and the steel plate is placed in a single row, then is 1, if it occupies two rows of space and the steel plate is placed in a double row, then is 1, if the steel plate j of order m is not placed on the rolled piece i, then is 0; The integer programming model is specifically: (1) Maximize the number of steel plates utilized, and require the proportion of the grouped steel plates in the total number of steel plates to be as large as possible: Among them, n is the total number of rolled pieces, m is the order number, q m is the number of steel plates in the m-th order, i is the rolled piece serial number, j is the steel plate serial number, and N is the total number of steel plates in all orders; (2) Maximize the length and width of each formed rolled piece, including: Maximize the length of a single rolled piece: Maximize the width of a single rolled piece: Where, is the maximum width of a single steel plate when the steel plates are placed in a single row; It is the maximum value of the sum of the widths of two steel plates placed side by side, one above the other, when the steel plates are placed in a double row. The widths of the j-th and j'-th steel plates of order m placed one above the other on the i-th rolled piece are: w mji and w mj′ i; (3) Minimize the trimming amount, and require the proportion of the grouped steel plate area in the rolled piece area to be as large as possible: Among them, r is the order quantity; q m is the quantity of steel plates in the m-th order; l mj is the length of the j-th steel plate in the m-th order; w mj is the width of the j-th steel plate in the m-th order; T i is the area of the i-th rolled piece; The constraint conditions of the integer programming model include: The length, width, and thickness limits of each rolled piece, the position of the steel plate does not exceed the rolled piece area limit, the difference in thickness between the steel plate and the grouped rolled piece has a range limit, and each steel plate can only be placed on a rolled piece in one of the single-row or double-row ways.

3. The intelligent plate grouping method that takes into account the camber shape of heavy plate rolling pieces according to claim 1, wherein, The input data in step S2 includes: the data features of the steel plates and rolled pieces. The feature vectors of the steel plates and rolled pieces are: F = [f1, f2,......, f k ​ f k Data characteristics including steel plates and rolled products, representing the length, width, and thickness of the steel plates, the number of steel plates combined on one rolled product, and the length, width, and thickness of the rolled product.

4. The intelligent plate grouping method for taking into account the camber shape of heavy plate rolling pieces according to claim 1, wherein In step S2, the decision layer includes an initial scheme generation module and a feature learning and optimization module. The initial scheme generation module uses the greedy algorithm to generate the initial splicing scheme. The feature learning and optimization module dynamically adjusts the feature weights according to the fitness evaluation, and the adjustment method is as follows: Based on the feature importance and fitness conditions feedback by the evaluation layer, dynamically update the feature weights, and the update formula is as follows: is the updated adaptability; α is the weight factor, which controls the balance between the old and new fitness, and α ∈ (0, 1). is the current adaptation degree; Score(x ij ) is the result of re-evaluating the fitness based on the performance of the current solution, calculated based on the fitness calculation formula.

5. The intelligent plate grouping method considering the camber shape of heavy plate rolling pieces according to claim 1, characterized in that In step S2, the evaluation layer uses the random forest method to evaluate the feature importance, specifically: Construct the dataset feature matrix F. F ij represents the value related to the characteristics of the i-th rolled piece and the steel plates combined to form the rolled piece F ij = [L iP W iP H iP Q iP L iS1 W iS1 H iS1 P iS1 ​ Where, L iP W iP H iP Q iP respectively represent the length, width, thickness of the i-th rolled piece and the total number of combined steel plates; L iS1 W iS1 H iS1 p iS1 respectively represent the length, width, thickness and position of each steel plate that makes up the i-th rolled piece; The target variable Y is set to the fitness of the length and width. Construct the decision tree and train the decision tree: Randomly select m features at each node to determine the best split; at the current node, traverse all available features, calculate the potential split points for each feature; for each feature and each potential split point, calculate the Gini impurity of the current node, calculate the weighted average Gini impurity of the split dataset, and finally calculate the impurity reduction; the importance I k of feature X k is measured by the sum of the Gini impurity reduction it causes in all trees, and the formula is as follows: T b All nodes for performing node splitting using feature X k ; b is a certain tree; B is the number of subsets; t is a specific node; ΔI kt is feature X k causing impurity reduction in the specific node t of a certain tree b; ΔI kt = Gini(D) - Ginisplit; Where, D is the dataset of the current node. D1 and D2 are two split data sets for the values of feature X respectively k and Gini(D) is the Gini impurity of the current node. Gini(D1) and Gini(D2) are the Gini impurities after splitting. T is the number of features. p i is the probability of feature i in dataset D Ginisplit is the weighted average Gini impurity after splitting. F represents the total number of feature sets used at the current node, and F1 and F2 respectively represent the number of features in the two subsets after splitting the current feature set; Each tree reaches the leaf node through node splitting; the feature value determines the path of the sample in the tree, which in turn affects the sample set reaching each leaf node, and ultimately affects the predicted value of the leaf node; calculate the predicted value of the leaf node and the actual target value, i.e., the length and width of the rolled piece, recursively construct the decision tree, and for each newly created subset, recursively repeat the above process until the stopping condition is reached, and feedback the calculated feature importance result to the decision-making layer.

6. The intelligent plate grouping method considering the camber shape of heavy plate rolling pieces according to claim 1, characterized in that In the step S2, the evaluation layer performs the fitness evaluation process as follows: It represents the fitness of the steel plate combination for the i-th rolled piece, and the fitness is represented by the length and width of the rolled piece. According to the objective function, within the constraint range, the larger the length and width of a single rolled piece, the better, that is, the larger the fitness value. The fitness calculation formula is as follows: Where: w l and w w are the weight coefficients of length and width. Set w l to 80% and w w to 20%; l mj is the length of the j-th steel plate in the m-th order; w mj is the width of the j-th steel plate in the m-th order; It is the sum of the widths of the two steel plates placed side by side vertically when the steel plates are placed in a double row.

7. The intelligent plate grouping method that takes into account the camber shape of heavy plate rolling pieces according to claim 1, characterized in that In the step S3, a multi-dimensional array is used for the coding of the fireworks algorithm. The solution output by the MHN decision-making layer is used as the initial population, and the fitness evaluation is carried out. Then, iterative operations are performed according to the explosion, mutation, and selection processes of the fireworks algorithm. When the maximum number of iterations is reached, the algorithm is stopped, and the individual with the highest fitness in the current population is output as the optimal plate combination plan. Among them, the fitness represents the quantity of steel plate utilization and the trimming amount, and the formula is as follows: Where: W u and W e are weight coefficients, W u takes values in the range of [5%, 20%], and W e takes values in the range of [80%, 95%]; f(x i ) is the fitness; n is the total number of rolled pieces, m is the order number, i is the serial number of the rolled piece, j is the serial number of the steel plate, r is the order quantity, q m is the quantity of steel plates for the m-th order, is the decision variable for single-row placement of steel plates, is the decision variable for double-row placement of steel plates, l mj is the length of the j-th steel plate in the m-th order, w mj is the width of the j-th steel plate in the m-th order, T i is the area of the i-th rolled piece, and N is the total number of steel plates for all orders.

8. The intelligent plate grouping method for considering the camber shape of heavy plate rolling pieces according to claim 7, wherein, In the step S3, the explosion operation of the fireworks algorithm combines the evaluation of the fitness of the matching between the steel plate and the rolled piece by the MHN, and designs the fitness coefficient. The explosion intensity formula is as follows: The explosion amplitude formula is as follows: Where: S i is the explosion intensity, A i is the explosion amplitude, s' and a' are constants respectively, s' is the factor representing the explosion intensity, and its value range is 10 - 200, a' is the factor representing the explosion amplitude, and its value range is 10% - 50%; f(x i ) is the fitness value of the current fireworks individual; f max = max(f(x i )) and f min = min(f(x i )) are the maximum and minimum values among the fitness values of K fireworks; ε is a very small constant used to ensure that the denominator is not zero; K is the number of fireworks; β is the weight controlling the influence of fitness on the explosion intensity, and its value range is [0,1]; is the average value of fitness; p i is the current fitness value; When the adaptability is high, that is at this time, A i decreases, resulting in a smaller blast radius; When the adaptability is low, that is , A i increases, resulting in a larger explosion radius.

Images