Full-process scrap steel grade judgment method and system based on data acquisition

Abstract

The invention relates to the technical field of production management, in particular to a full-process scrap steel grade judgment method and system based on data acquisition, and the method comprises the following steps: respectively calculating local difference distribution of feature points based on a scrap steel density change value, a scrap steel component trend value and a scrap steel mechanical characteristic value acquired in a scrap steel processing process, association weights are extracted through distance matrix normalization processing between the feature points, and a distribution result is mapped to a low-dimensional space. According to the method, multi-dimensional information such as waste steel density, components and mechanical characteristics is extracted through data acquisition, the characteristic point association weight is accurately calculated and mapped to a low-dimensional space, simplified expression of the characteristic information is achieved, adjacent point weight optimization enhances the accuracy of waste steel grouping, fuzzy boundaries of the characteristic points are effectively eliminated, and the accuracy of waste steel grouping is improved. Abnormal points and boundary samples are determined through density and probability analysis, classification interference is reduced, key information is guaranteed to be completely reserved, the comprehensiveness and accuracy of scrap steel characteristic grading are further improved through grouping distribution optimization, and the resource utilization efficiency is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of production management, and in particular to a full-process scrap steel grading method and system based on data collection. Background Art

[0002] The field of production management technology refers to a comprehensive system that uses various technologies and methods to plan, organize, coordinate and control production activities. This field covers the whole process management from raw material procurement, production process design, manufacturing process monitoring to finished product delivery, with the aim of improving production efficiency, reducing production costs, optimizing resource allocation and ensuring product quality. With the rapid development of information technology, computer systems and big data analysis technologies are widely used in the field of production management, realizing the intelligentization, informatization and automation of the production process, providing important support for the efficient operation of enterprises and the improvement of their market competitiveness.

[0003] Among them, the whole process scrap steel grading method refers to the method and system for grading and managing the whole life cycle of scrap steel from its initial state to the completion of processing through data collection and analysis. Its purpose is to improve the efficiency of scrap steel recycling and utilization through accurate grading, reduce resource waste, and provide technical support for the efficient utilization of scrap steel and the development of circular economy. This method helps to optimize the waste management process in production management, ensure the quality of production raw materials and control production costs.

[0004] Existing technologies are not capable of comprehensive analysis of multi-dimensional characteristic data of scrap steel. There are fuzzy areas in the extraction of feature point weights and the definition of groups, resulting in low accuracy of classification results. The identification of outliers relies on simple algorithms, which makes it difficult to accurately process boundary samples, easily leading to errors in scrap steel classification, affecting resource recovery efficiency and the stability of production raw material quality. For example, the inability to effectively distinguish between boundary points and outliers may lead to misjudgment, which in turn affects the utilization rate and cost control of scrap steel processing, and limits the refined management capabilities of scrap steel recycling. Summary of the invention

[0005] The purpose of the present invention is to solve the shortcomings of the prior art and to propose a full-process scrap steel grading method and system based on data collection.

[0006] In order to achieve the above object, the present invention adopts the following technical solution: a full-process scrap steel grading method based on data collection, comprising the following steps: S1: Based on the scrap steel density change value, scrap steel composition trend value and scrap steel mechanical property value collected during the scrap steel processing, the local difference distribution of the feature points is calculated respectively, and the correlation weight is extracted by normalizing the distance matrix between the feature points. The distribution results are mapped to a low-dimensional space to generate a low-dimensional feature embedding matrix; S2: Based on the low-dimensional feature embedding matrix, the squared Euclidean distances between the scrap steel density change and scrap steel composition trend feature points and the surrounding neighboring points are calculated respectively, and the weighted matrix structure is formed by normalizing the weight distribution, and the weight distribution is summarized to establish a weighted adjacency graph structure; S3: based on the weighted adjacency graph structure, extract the weight distribution matrix of scrap steel mechanical properties and scrap steel density changes in turn, normalize the weight matrix by calculating the cutting edge weights between feature points, optimize the grouping boundaries of feature points, divide the scrap steel feature points into multiple groups, and generate scrap steel grouping classification results; S4: Based on the scrap steel grouping and classification results, the probability density values ​​of the scrap steel density change and scrap steel composition trend feature points are calculated point by point, the feature points with lower density are marked as boundary samples, and the abnormal points are marked after the neighboring distribution difference is calculated for the isolated feature points, so as to generate scrap steel boundary samples and abnormal point distribution labels; S5: Based on the scrap steel boundary samples and abnormal point distribution labels, the number of weighted adjacent points and the weight distribution matrix are adjusted, the weights of the scrap steel composition trend and the scrap steel mechanical property grouping boundaries are normalized, and the grouping distribution is recalculated through the feature point density kernel parameters to obtain optimized grading feature results.

[0007] The results of the scrap steel grading method include a low-dimensional feature embedding matrix, a weighted adjacency graph structure, scrap steel grouping classification results, scrap steel boundary samples and abnormal point distribution labels, and optimized grading feature results. The low-dimensional feature embedding matrix is ​​specifically the association weight, low-dimensional space distribution mapping, and feature point normalization processing results. The weighted adjacency graph structure includes a weight distribution matrix, a Euclidean distance square value, and a weighted matrix structure. The scrap steel grouping classification results are specifically feature point grouping boundaries, grouping optimization results, and multi-group division results. The scrap steel boundary samples and abnormal point distribution labels are specifically boundary sample marks, abnormal point marks, and neighboring distribution differences. The optimized grading feature results include density kernel parameters, grouping distribution optimization, and grouping boundary weight normalization processing.

[0008] As a further solution of the present invention, the step of obtaining the low-dimensional feature embedding matrix is ​​specifically as follows: S111: calculating the local difference distribution of the feature points respectively according to the scrap steel density change value, scrap steel composition trend value and scrap steel mechanical property value collected during the scrap steel processing, and generating a local difference distribution matrix of the feature points; S112: calculating a distance matrix between feature points based on the local difference distribution matrix of the feature points, and generating a feature point association weight distribution matrix by normalizing the distance matrix; S113: Taking the feature point associated weight distribution matrix as input, combining the normalized local difference values ​​of the feature points and the weight relationship between the feature points, using the formula ; Calculate the weight distribution relationship of low-dimensional embedded feature points and generate an optimized weight distribution matrix; in, Represents the weight distribution value of the low-dimensional embedded feature points, Represents the normalized distance value between feature points, Represents the initial weight value of the feature point, Represents the local difference value of the feature point, represents the normalized weight correction value, is the total number of feature points; S114: Based on the optimized weight distribution matrix, the result is mapped to a low-dimensional space, and a low-dimensional feature embedding matrix is ​​obtained through feature projection.

[0009] As a further solution of the present invention, the step of obtaining the weighted adjacency graph structure is specifically as follows: S211: Based on the low-dimensional feature embedding matrix, the Euclidean distance square values ​​between the scrap steel density change and the scrap steel composition trend feature points are calculated to obtain a distance square matrix; S212: normalizing the distance square matrix according to the weight distribution between the feature points to obtain a normalized distance weight matrix; S213: Combining the normalized distance weight matrix with the original weight value of the feature point, using the formula: ; Calculate the weighted matrix structure, summarize all the values ​​in the normalized weight graph, construct the weighted adjacency graph structure, and complete the spatial structure modeling of scrap steel characteristics; in, represents the weights of nodes i and j in the weighted adjacency graph, Represents the square of the Euclidean distance between node i and node j, is the base of the natural logarithm, indicating that the effect of distance on weight is handled in an exponential decay form. is the original weight value from node i to node j, is the total number of feature points.

[0010] As a further solution of the present invention, the steps for obtaining the scrap steel grouping and classification results are specifically as follows: S311: based on the weighted adjacency graph structure, extracting the weight distribution matrix of scrap steel mechanical properties and scrap steel density changes, calculating the edge cutting weights between feature points, and generating a feature point edge cutting weight matrix; S312: normalizing the weights according to the feature point edge cutting weight matrix, optimizing the weight distribution of the feature points by comparing the relationship between the edge cutting weights and the normalization threshold, and generating the optimized grouping boundary of the feature points; S313: Based on the optimized grouping boundaries of the feature points, the intra-group connectivity and inter-group cutting edge weight difference between each group of feature points are calculated using the formula: ; Generate grouping and classification results of scrap steel feature points; in, Represents the weight distribution evaluation of scrap steel grouping and classification results, Represents the weight distribution between feature points within the group, Represents the optimized boundary weight within the feature point group, represents the difference in the weight of the cut edge between groups, is the total number of groups, is the number of feature points in each group, is the number of all feature points, is the number of inter-group cutting edge feature points.

[0011] As a further solution of the present invention, the steps of obtaining the scrap steel boundary samples and abnormal point distribution labels are specifically as follows: S411: extracting the scrap density change and scrap composition trend feature points of each group from the scrap steel grouping classification results, calculating the probability density value of each feature point by a local probability density estimation method, and generating a probability density distribution matrix; S412: Analyze the probability density distribution matrix, identify feature points whose probability density is lower than a set threshold as boundary samples, and obtain a boundary sample identification list; S413: Based on the boundary sample identification list, for feature points that are not marked as boundary samples, the probability density difference between the feature points and the nearest neighbor feature points is calculated using the formula: ; like Greater than the adjusted threshold , it is marked as an outlier, and scrap steel boundary samples and outlier distribution labels are generated; in, It represents the normalized difference between the standard deviation of the probability density average value of the i-th feature point and the k nearest neighbor feature points. is the probability density of the i-th feature point, is the probability density of the jth neighboring point of the ith feature point, is the number of neighboring points, is the standard deviation of the probability density of neighboring points, It is an empirically set threshold used to determine abnormal points.

[0012] As a further solution of the present invention, the step of obtaining the optimization classification feature result is specifically: S511: extracting the number of weighted adjacent points from the scrap steel boundary samples and the abnormal point distribution labels, adjusting the weights according to the abnormality degree of the multiple feature points, recalculating the weighted adjacent point number matrix, and generating an adjusted weighted adjacent point number matrix; S512: normalizing the adjusted weighted adjacent point quantity matrix, using linear transformation to evenly distribute the weight of each feature point, to obtain a normalized weighted adjacent point weight matrix; S513: Based on the normalized weighted neighboring point weight matrix, a kernel density estimation method is used to recalculate the scrap steel composition trend and the distribution of scrap steel mechanical property groups, using the formula: ; Generate optimized classification feature results; in, represents the optimized classification feature result of the kth group, is the kernel function, which is used to calculate the probability density contribution of feature points. is the attribute value of the i-th feature point, is the attribute value of the k-th group center, is the bandwidth of the kernel function, which controls the smoothness of the kernel density estimate. is the weight of the i-th feature point to the k-th group center, is the weight difference adjusted based on the distribution of outliers, is the weight adjustment coefficient, which is used to balance the influence of normal and abnormal feature points. Represents the total number of feature points involved in kernel density estimation.

[0013] The full-process scrap steel grading system based on data collection is used to execute the full-process scrap steel grading method based on data collection, and the system includes: The data feature extraction module calculates the local difference distribution value based on the scrap steel density change value, scrap steel composition trend value and scrap steel mechanical property value collected during the scrap steel processing process, extracts the density difference value, composition difference value and mechanical property difference value, normalizes the distribution value between feature points, and generates a low-dimensional feature embedding matrix; The association weight construction module calculates the density change value and the component trend value and the square of the Euclidean distance of the neighboring points based on the low-dimensional feature embedding matrix, normalizes the weight distribution matrix, extracts the total weight and generates the adjacency structure, and establishes a weighted adjacency graph matrix; The grouping and classification optimization module extracts the weight distribution matrix of mechanical characteristic values ​​and density change values ​​based on the weighted adjacency graph matrix, calculates the edge cutting weights between feature points and normalizes the matrix, determines the grouping boundary, divides the feature points into multiple groups, and generates a scrap steel feature grouping matrix; The abnormal distribution detection module calculates the probability density values ​​of multiple feature points in the group based on the scrap steel feature grouping matrix, marks the density extreme value points as boundary samples, calculates the neighboring distribution differences of isolated feature points and marks the abnormal points, and obtains the scrap steel boundary sample distribution and abnormal point labels; The feature grading optimization module adjusts the weight distribution and quantity of adjacent points based on the scrap steel boundary sample distribution and abnormal point labels, optimizes the grouping boundaries of component trend values ​​and mechanical property values, re-divides the feature point distribution, and generates optimized grading feature parameters.

[0014] Compared with the prior art, the advantages and positive effects of the present invention are: In the present invention, multi-dimensional information such as scrap steel density, composition and mechanical properties is extracted through data collection, and the associated weights of feature points are accurately calculated and mapped to low-dimensional space to achieve simplified expression of feature information. Adjacent point weight optimization enhances the accuracy of scrap steel grouping and effectively eliminates the fuzzy boundaries of feature points. Density and probability analysis identify abnormal points and boundary samples, reduce classification interference, and ensure that key information is fully retained. Group distribution optimization further improves the comprehensiveness and accuracy of scrap steel characteristic classification, improves resource utilization efficiency, optimizes scrap steel management processes, and provides reliable technical support for the development of circular economy. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 It is a schematic diagram of the workflow of the present invention; Figure 2 A flowchart of the steps for obtaining a low-dimensional feature embedding matrix of the present invention; Figure 3 A flowchart of the steps for obtaining the weighted adjacency graph structure of the present invention; Figure 4 The flowchart of the steps for obtaining the results of the scrap steel grouping and classification of the present invention; Figure 5 A flowchart of the steps for obtaining scrap steel boundary samples and abnormal point distribution labels of the present invention; Figure 6 Flow chart of the steps for obtaining the optimized classification feature results of the present invention. DETAILED DESCRIPTION

[0016] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0017] In the description of the present invention, it should be understood that the terms "length", "width", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside" and the like indicate positions or positional relationships based on the positions or positional relationships shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, in the description of the present invention, "multiple" means two or more, unless otherwise clearly and specifically defined.

[0018] Embodiment 1 See also Figure 1 The present invention provides a technical solution: a full-process scrap steel grading method based on data collection, comprising the following steps: S1: Based on the scrap steel density change value, scrap steel composition trend value and scrap steel mechanical property value collected during the scrap steel processing, the local difference distribution of the feature points is calculated respectively, and the correlation weight is extracted by normalizing the distance matrix between the feature points. The distribution results are mapped to a low-dimensional space to generate a low-dimensional feature embedding matrix; S2: Based on the low-dimensional feature embedding matrix, the squared Euclidean distances between the scrap density change and scrap composition trend feature points and the surrounding neighboring points are calculated, and the weighted matrix structure is formed by normalizing the weight distribution. The weight distribution is summarized to establish a weighted adjacency graph structure. S3: Based on the weighted adjacency graph structure, the weight distribution matrix of scrap steel mechanical properties and scrap steel density changes is extracted in turn, the weight matrix is ​​normalized by calculating the cutting edge weights between feature points, the grouping boundaries of feature points are optimized, the scrap steel feature points are divided into multiple groups, and the scrap steel grouping classification results are generated; S4: Based on the scrap steel grouping and classification results, the probability density values ​​of the scrap steel density change and scrap steel composition trend feature points are calculated point by point, and the feature points with lower density are marked as boundary samples. The difference between the adjacent distributions of isolated feature points is calculated and the abnormal points are marked to generate scrap steel boundary samples and abnormal point distribution labels; S5: Based on the scrap steel boundary samples and abnormal point distribution labels, the number of weighted adjacent points and the weight distribution matrix are adjusted, the weights of the scrap steel composition trend and the scrap steel mechanical property grouping boundary are normalized, and the grouping distribution is recalculated through the feature point density kernel parameter to obtain the optimized classification feature results.

[0019] The results of the scrap steel grading method include a low-dimensional feature embedding matrix, a weighted adjacency graph structure, scrap steel grouping and classification results, scrap steel boundary samples and abnormal point distribution labels, and optimized grading feature results. The low-dimensional feature embedding matrix specifically includes association weights, low-dimensional space distribution mapping, and feature point normalization processing results. The weighted adjacency graph structure includes a weight distribution matrix, a Euclidean distance square value, and a weighted matrix structure. The scrap steel grouping and classification results specifically include feature point grouping boundaries, grouping optimization results, and multi-group division results. The scrap steel boundary samples and abnormal point distribution labels specifically include boundary sample labels, abnormal point labels, and neighboring distribution differences. The optimized grading feature results include density kernel parameters, grouping distribution optimization, and grouping boundary weight normalization processing.

[0020] See also Figure 2 , the specific steps for obtaining the low-dimensional feature embedding matrix are: S111: calculating the local difference distribution of the feature points respectively according to the scrap steel density change value, scrap steel composition trend value and scrap steel mechanical property value collected during the scrap steel processing, and generating a local difference distribution matrix of the feature points; First, a local difference analysis is performed on the scrap steel density change value of each data point. The difference in the local area is determined by calculating the difference between adjacent data points. This step is to capture the slight changes in the scrap steel processing process. Then these local difference data are aggregated into a matrix, which will be used as the basic data set for further analysis. At the same time, the trend value of the scrap steel composition will be analyzed by trend analysis technology to determine the change trend of the scrap steel composition with processing time. Finally, the mechanical property values ​​of the scrap steel will also be processed similarly to ensure that the most valuable information is extracted from each key characteristic. The processing of these data includes not only the collection and preliminary processing of the data, but also a more advanced statistical analysis of the data to ensure that the most useful information can be extracted from the data and generate a local difference distribution matrix of the feature points.

[0021] S112: Based on the local difference distribution matrix of the feature points, a distance matrix between the feature points is calculated, and a feature point association weight distribution matrix is ​​generated by normalizing the distance matrix; By comparing the difference between each feature point and all other feature points, these distances are calculated using Euclidean distance or other related distance measurement methods. These calculated distances will be combined into a distance matrix. This distance matrix provides a quantitative basis for subsequent steps for more in-depth data analysis. Normalization is the next key step, which involves converting the data in the distance matrix into values ​​between 0 and 1. This processing step is necessary because it can remove the impact of the data dimension, thereby allowing data points of different scales to be compared fairly and generate a feature point association weight distribution matrix.

[0022] S113: Taking the feature point associated weight distribution matrix as input, combined with the normalized local difference value of the feature point and the weight relationship between the feature points, the formula is used ; Calculate the weight distribution relationship of low-dimensional embedded feature points and generate an optimized weight distribution matrix; in, Represents the weight distribution value of the low-dimensional embedded feature points, Represents the normalized distance value between feature points, Represents the initial weight value of the feature point, Represents the local difference value of the feature point, represents the normalized weight correction value, is the total number of feature points; formula: ; The benefit of the formula is that it provides a new method for calculating the distribution of associated weights of feature points by comprehensively considering the deviation of distance and weight, the local difference value of each feature point and the normalized weight correction value. This method not only considers the basic distance between points, but also considers the importance and variability of the points, thereby obtaining a more accurate weight adjustment value, improving the accuracy of feature mapping and the efficiency of low-dimensional space mapping.

[0023] Detailed explanation of the formula and the process of formula calculation and derivation: Suppose there are two feature points, and their distance value is 0.5, the initial weight is 0.3, the local difference value is 0.1, normalized weight is 0.2, and the calculation process is as follows: ; This result shows that by considering the distance difference and weight correction method, the relationship between feature points can be evaluated more accurately, which helps to represent high-dimensional data more effectively in low-dimensional space.

[0024] S114: Based on the optimized weight distribution matrix, the result is mapped to a low-dimensional space, and a low-dimensional feature embedding matrix is ​​obtained through feature projection.

[0025] This process is accomplished by using multidimensional scaling techniques or other dimensionality reduction techniques, which can effectively transform high-dimensional data into low-dimensional representations while retaining important information in the original data set. This mapping process not only focuses on the global structure of the data, but also attempts to retain the local relationships between data points. This is achieved by optimizing the weight distribution. The mapping process needs to ensure that the important features of the data are retained while removing unnecessary noise and redundant information. This can result in a clearer and more useful low-dimensional data representation and generate the final low-dimensional feature embedding matrix.

[0026] See also Figure 3 , the specific steps for obtaining the weighted adjacency graph structure are: S211: based on the low-dimensional feature embedding matrix, the Euclidean distance square values ​​between the scrap steel density change and the scrap steel composition trend feature points are calculated to obtain the distance square matrix; The main parameters involved in this process include the coordinates of each feature point in the low-dimensional space. These coordinate values ​​are obtained by multi-dimensional scaling of the scrap steel data. In the calculation of the distance square matrix, the coordinates of each feature point are first determined, and then the distance between the two points is calculated and squared using the Euclidean distance formula. This calculation step ensures the accuracy of the distance and the scientific nature of the operation. The obtained distance square matrix provides the necessary basic data for subsequent steps.

[0027] S212: normalizing the distance square matrix according to the weight distribution between the feature points to obtain a normalized distance weight matrix; This process involves dividing each distance value by the maximum distance value in the distance matrix to achieve normalization. The purpose of normalization is to balance the influence between feature points and ensure that no feature point will have too much impact on the overall structure due to excessive original distance values. This processing method helps to provide standardized data input for subsequent weight calculations and the establishment of graph structures. The obtained normalized distance weight matrix provides accurate input parameters for constructing a weighted adjacency graph.

[0028] S213: Combine the normalized distance weight matrix with the original weight value of the feature point using the formula: ; Calculate the weighted matrix structure, summarize all the values ​​in the normalized weight graph, construct the weighted adjacency graph structure, and complete the spatial structure modeling of scrap steel characteristics; in, represents the weights of nodes i and j in the weighted adjacency graph, Represents the square of the Euclidean distance between node i and node j, is the base of the natural logarithm, indicating that the effect of distance on weight is handled in an exponential decay form. is the original weight value from node i to node j, is the total number of feature points.

[0029] formula: ; The benefit of the formula is that it adjusts the distance between feature points in an exponential decay manner, emphasizing the connection strength between points that are closer, while weakening the weight of points that are farther away. This processing strengthens the model's ability to capture local structures, allowing the adjacency graph to better reflect the spatial structural relationship of scrap steel characteristics.

[0030] Detailed explanation of the formula and the process of formula calculation and derivation: Assume there are three feature points, and their distance matrix is , the weight is , put it into the formula and calculate : ; The result shows that the connection strength between node 1 and node 2 is relatively high, indicating that they are more similar and have a more significant impact on the model. The acquisition of this result reflects the adaptability and accuracy of the model in dealing with the influence of distance, and further deepens the understanding of the spatial structural relationship of scrap steel characteristics.

[0031] See also Figure 4 The specific steps for obtaining the scrap steel grouping and classification results are as follows: S311: based on the weighted adjacency graph structure, extracting the weight distribution matrix of scrap steel mechanical properties and scrap steel density changes, calculating the edge cutting weights between feature points, and generating a feature point edge cutting weight matrix; First, it is necessary to conduct an in-depth analysis of the weighted adjacency graph structure, extract the relevant feature points of the mechanical properties of scrap steel and the changes in scrap steel density, and calculate the edge weights between these feature points based on this. This step requires a full understanding and precise processing of the structure and weight distribution of the weighted adjacency graph, and optimize the weight distribution of feature points through mathematical models and algorithms to ensure that the weight of each feature point can truly reflect the actual situation of its mechanical properties and density changes, so as to accurately perform subsequent normalization processing. Through normalization processing, the grouping boundaries between feature points can be further optimized and refined, making the grouping of scrap steel more reasonable and in line with the needs of actual processing.

[0032] S312: normalizing the weights according to the feature point cutting edge weight matrix, optimizing the weight distribution of the feature points by comparing the relationship between the cutting edge weights and the normalization threshold, and generating the optimized grouping boundary of the feature points; Select appropriate normalization formulas and parameters, which need to be adjusted according to the specific distribution of the cutting edge weights between feature points. The normalized weights will be used to optimize the grouping boundaries of feature points. This step is a precise calculation through a mathematical model, with the aim of making the boundaries between groups more obvious, thereby facilitating the effective grouping of scrap steel feature points according to their mechanical properties and density changes. This grouping is of great significance for the subsequent processing and utilization of scrap steel, and can greatly improve the classification efficiency and processing quality of scrap steel.

[0033] S313: Based on the optimized grouping boundaries of the feature points, the intra-group connectivity and inter-group cutting edge weight difference between each group of feature points are calculated using the formula: ; Generate grouping and classification results of scrap steel feature points; in, Represents the weight distribution evaluation of scrap steel grouping and classification results, Represents the weight distribution between feature points within the group, Represents the optimized boundary weight within the feature point group, represents the difference in the weight of the cut edge between groups, is the total number of groups, is the number of feature points in each group, is the number of all feature points, is the number of inter-group cutting edge feature points.

[0034] formula: ; The benefit of the formula is that by optimizing the grouping boundaries and grouping classification results of the feature points, the classification and subsequent processing of scrap steel can be handled more accurately. The parameters in the formula Represents the weights between feature points within a group, Represents the boundary weight of the optimized feature point grouping, It represents the difference in cutting edge weights between groups. These parameters are obtained through specific data collection and calculation, ensuring the accuracy and practicality of the scrap steel classification results.

[0035] Detailed explanation of the formula and the process of formula calculation and derivation: Assumptions The value of is 0.35, The value of is 0.1, With a value of 0.05, the calculation gives: ; ; The results show that by optimizing the grouping boundaries and weight distribution, the classification of scrap steel is more accurate, which helps to improve the scrap steel processing efficiency and material utilization.

[0036] See also Figure 5 , the specific steps for obtaining scrap steel boundary samples and abnormal point distribution labels are as follows: S411: extracting the scrap density change and scrap composition trend feature points of each group from the scrap steel grouping classification results, calculating the probability density value of each feature point by a local probability density estimation method, and generating a probability density distribution matrix; First, the data of each feature point is collected, including its grouping data and the corresponding density change records. The kernel density estimation method is used to estimate the density of each point by weighting the contribution of the surrounding points. The contribution of each feature point is weighted by the attenuation of its distance function. The distance function is weighted using a Gaussian kernel to calculate the relative probability density of each feature point in its group. This probability density reflects the rarity or commonness of each feature point relative to other points in its group, and a probability density distribution matrix is ​​obtained.

[0037] S412: Analyze the probability density distribution matrix, identify feature points whose probability density is lower than a set threshold as boundary samples, and obtain a boundary sample identification list; The threshold is determined according to the statistical distribution of the density value. The threshold is intended to separate points with lower density as boundary samples. The specific operation is to calculate the density mean of all feature points and set the threshold to a certain proportion of the mean, such as 25% of the mean. Each point in the probability density matrix is ​​traversed. If the density value of the point is lower than this threshold, it is marked as a boundary sample. During the identification process, the spatial distribution of these points must be checked to ensure that the marked boundary samples are representative in space, thereby obtaining a boundary sample identification list.

[0038] S413: Based on the boundary sample identification list, for feature points that are not marked as boundary samples, the probability density difference between the feature points and the nearest neighbor feature points is calculated using the formula: ; like Greater than the adjusted threshold , it is marked as an outlier, and scrap steel boundary samples and outlier distribution labels are generated; in, It represents the normalized difference between the standard deviation of the probability density average value of the i-th feature point and the k nearest neighbor feature points. is the probability density of the i-th feature point, is the probability density of the jth neighboring point of the ith feature point, is the number of neighboring points, is the standard deviation of the probability density of neighboring points, It is an empirically set threshold used to determine abnormal points.

[0039] formula: ; The benefit of the formula is that it not only takes into account the density differences of the feature points themselves, but also introduces standard deviation for normalization, thereby increasing the sensitivity to outliers while maintaining adaptability to different density distributions.

[0040] Detailed explanation of the formula and the process of formula calculation and derivation: Assume that there is a probability density of a feature point , its most recent The densities of neighbor points are , calculate the average density of neighbor points: ; The difference between the probability density of the feature point and the average density of its neighbors is ; Assume that the standard deviation of these density values ​​is , then the normalized difference is ; If you set the threshold , then this feature point is not marked as an outlier.

[0041] This result shows that the density of this feature point is not significantly lower than that of its neighboring points, so it is not considered an outlier under the current threshold setting.

[0042] See also Figure 6 , the specific steps for obtaining the optimized classification feature results are: S511: extracting the number of weighted adjacent points from the scrap steel boundary samples and the abnormal point distribution labels, adjusting the weights according to the abnormality degree of the multiple feature points, recalculating the weighted adjacent point number matrix, and generating an adjusted weighted adjacent point number matrix; This adjustment is based on the abnormality level of each point, where the boundary samples have a higher weight adjustment factor to reflect their importance in the overall sample. In this way, the number of weighted neighboring points is recalculated to ensure the uniformity of the weight distribution within each group. During this process, the original weight of each neighboring point and its adjusted value are recorded for subsequent normalization. This process ensures that the influence of each feature point is appropriately adjusted according to its position in the distribution, thus laying the foundation for the next step of normalization and generating an adjusted weighted neighbor point quantity matrix.

[0043] S512: normalizing the adjusted weighted adjacent point quantity matrix, using linear transformation to evenly distribute the weight of each feature point, and obtaining a normalized weighted adjacent point weight matrix; The linear transformation method used in this step ensures the uniformity of the weight distribution of each feature point. The normalization process involves comparing each weight with the maximum weight in the group to determine the normalization coefficient. The normalized weights are used to balance the influence of each feature point in the calculation process to ensure that no single feature point has a disproportionate impact on the final result due to excessive weight. In this way, a normalized weighted neighboring point weight matrix is ​​obtained, which will be used for subsequent kernel density estimation to optimize the results of hierarchical features.

[0044] S513: Based on the normalized weighted neighboring point weight matrix, the kernel density estimation method is used to recalculate the scrap steel composition trend and the distribution of scrap steel mechanical property groups using the formula: ; Generate optimized classification feature results; in, represents the optimized classification feature result of the kth group, is the kernel function, which is used to calculate the probability density contribution of feature points. is the attribute value of the i-th feature point, is the attribute value of the k-th group center, is the bandwidth of the kernel function, which controls the smoothness of the kernel density estimate. is the weight of the i-th feature point to the k-th group center, is the weight difference adjusted based on the distribution of outliers, is the weight adjustment coefficient, which is used to balance the influence of normal and abnormal feature points. Represents the total number of feature points involved in kernel density estimation.

[0045] formula: ; The formula is useful in that by adjusting the weights Combined with the Gaussian kernel function , allowing the model to more finely control the contribution of each feature point to the grouping results, especially when considering the adjustment of outliers, which enhances the flexibility and accuracy of the model in dealing with outliers.

[0046] Detailed explanation of the formula and the process of formula calculation and derivation: Set the specific grouping to , and its core bandwidth is , consider three feature points, namely The attribute values ​​of the feature points are , the attribute value of the group center , the normalized weight is , the outlier weights are adjusted to , adjustment coefficient , calculate the contribution of each term: ; ; ; Contribution calculation for each point: ; ; ; Aggregate Grouping The results: ; The results show that the group The optimized classification feature result is 0.410, which reflects the average contribution of the feature points around the grouping center. Taking into account the adjustment of abnormal points, the result more truly reflects the actual situation of scrap steel composition trend and mechanical properties, providing a basis for the next decision.

[0047] The full-process scrap steel grading system based on data collection is used to execute the full-process scrap steel grading method based on data collection. The system includes: The data feature extraction module calculates the local difference distribution value based on the scrap steel density change value, scrap steel composition trend value and scrap steel mechanical property value collected during the scrap steel processing process, extracts the density difference value, composition difference value and mechanical property difference value, normalizes the distribution value between feature points, and generates a low-dimensional feature embedding matrix; The association weight construction module is based on the low-dimensional feature embedding matrix, calculates the density change value and the component trend value and the square of the Euclidean distance between the neighboring points, normalizes the weight distribution matrix, extracts the total weight and generates the adjacency structure, and establishes a weighted adjacency graph matrix; The grouping and classification optimization module extracts the weight distribution matrix of mechanical characteristic values ​​and density change values ​​based on the weighted adjacency graph matrix, calculates the edge cutting weights between feature points and normalizes the matrix, determines the grouping boundaries, divides the feature points into multiple groups, and generates the scrap steel feature grouping matrix; The abnormal distribution detection module calculates the probability density values ​​of multiple feature points in the group based on the scrap steel feature grouping matrix, marks the density extreme value points as boundary samples, calculates the neighboring distribution difference of isolated feature points and marks the abnormal points, and obtains the scrap steel boundary sample distribution and abnormal point labels; The feature grading optimization module adjusts the weight distribution and quantity of adjacent points based on the scrap steel boundary sample distribution and abnormal point labels, optimizes the grouping boundaries of composition trend values ​​and mechanical property values, re-divides the feature point distribution, and generates optimized grading feature parameters.

[0048] The above are only preferred embodiments of the present invention and are not intended to limit the present invention in other forms. Any technician familiar with the profession may use the technical contents disclosed above to change or modify them into equivalent embodiments with equivalent changes and apply them to other fields. However, any simple modification, equivalent change and modification made to the above embodiments based on the technical essence of the present invention without departing from the technical solution of the present invention still falls within the protection scope of the technical solution of the present invention.

Claims

1. A full-process scrap steel grading method based on data collection, characterized in that: The following steps are involved: S1: Based on the scrap steel density change value, scrap steel composition trend value and scrap steel mechanical property value collected during the scrap steel processing, the local difference distribution of the feature points is calculated respectively, and the correlation weight is extracted by normalizing the distance matrix between the feature points. The distribution results are mapped to a low-dimensional space to generate a low-dimensional feature embedding matrix; S2: Based on the low-dimensional feature embedding matrix, the squared Euclidean distances between the scrap steel density change and scrap steel composition trend feature points and the surrounding neighboring points are calculated respectively, and the weighted matrix structure is formed by normalizing the weight distribution, and the weight distribution is summarized to establish a weighted adjacency graph structure; S3: based on the weighted adjacency graph structure, extract the weight distribution matrix of scrap steel mechanical properties and scrap steel density changes in turn, normalize the weight matrix by calculating the cutting edge weights between feature points, optimize the grouping boundaries of feature points, divide the scrap steel feature points into multiple groups, and generate scrap steel grouping classification results; S4: Based on the scrap steel grouping and classification results, the probability density values ​​of the scrap steel density change and scrap steel composition trend feature points are calculated point by point, the feature points with lower density are marked as boundary samples, and the abnormal points are marked after the neighboring distribution difference is calculated for the isolated feature points, so as to generate scrap steel boundary samples and abnormal point distribution labels; S5: Based on the scrap steel boundary samples and abnormal point distribution labels, the number of weighted adjacent points and the weight distribution matrix are adjusted, the weights of the scrap steel composition trend and the scrap steel mechanical property grouping boundaries are normalized, and the grouping distribution is recalculated through the feature point density kernel parameters to obtain optimized grading feature results.

2. The full-process scrap steel grading method based on data collection according to claim 1 is characterized in that: The results of the scrap steel grading method include a low-dimensional feature embedding matrix, a weighted adjacency graph structure, scrap steel grouping classification results, scrap steel boundary samples and abnormal point distribution labels, and optimized grading feature results. The low-dimensional feature embedding matrix is ​​specifically the association weight, low-dimensional space distribution mapping, and feature point normalization processing results. The weighted adjacency graph structure includes a weight distribution matrix, a Euclidean distance square value, and a weighted matrix structure. The scrap steel grouping classification results are specifically feature point grouping boundaries, grouping optimization results, and multi-group division results. The scrap steel boundary samples and abnormal point distribution labels are specifically boundary sample marks, abnormal point marks, and neighboring distribution differences. The optimized grading feature results include density kernel parameters, grouping distribution optimization, and grouping boundary weight normalization processing.

3. The full-process scrap steel grading method based on data collection according to claim 2 is characterized in that: The steps of obtaining the low-dimensional feature embedding matrix are specifically as follows: S111: calculating the local difference distribution of the feature points respectively according to the scrap steel density change value, scrap steel composition trend value and scrap steel mechanical property value collected during the scrap steel processing, and generating a local difference distribution matrix of the feature points; S112: calculating a distance matrix between feature points based on the local difference distribution matrix of the feature points, and generating a feature point association weight distribution matrix by normalizing the distance matrix; S113: Taking the feature point associated weight distribution matrix as input, combining the normalized local difference values ​​of the feature points and the weight relationship between the feature points, using the formula ； Calculate the weight distribution relationship of low-dimensional embedded feature points and generate an optimized weight distribution matrix; in, Represents the weight distribution value of the low-dimensional embedded feature points, Represents the normalized distance value between feature points, Represents the initial weight value of the feature point, Represents the local difference value of the feature point, represents the normalized weight correction value, is the total number of feature points; S114: Based on the optimized weight distribution matrix, the result is mapped to a low-dimensional space, and a low-dimensional feature embedding matrix is ​​obtained through feature projection.

4. The full-process scrap steel grading method based on data collection according to claim 3 is characterized in that: The steps for obtaining the weighted adjacency graph structure are specifically as follows: S211: Based on the low-dimensional feature embedding matrix, the Euclidean distance square values ​​between the scrap steel density change and the scrap steel composition trend feature points are calculated to obtain a distance square matrix; S212: normalizing the distance square matrix according to the weight distribution between the feature points to obtain a normalized distance weight matrix; S213: Combining the normalized distance weight matrix with the original weight value of the feature point, using the formula: ； Calculate the weighted matrix structure, summarize all the values ​​in the normalized weight graph, construct the weighted adjacency graph structure, and complete the spatial structure modeling of scrap steel characteristics; in, represents the weights of nodes i and j in the weighted adjacency graph, Represents the square of the Euclidean distance between node i and node j, is the base of the natural logarithm, indicating that the effect of distance on weight is handled in an exponential decay form. is the original weight value from node i to node j, is the total number of feature points.

5. The full-process scrap steel grading method based on data collection according to claim 4 is characterized in that: The steps for obtaining the scrap steel grouping and classification results are specifically as follows: S311: based on the weighted adjacency graph structure, extracting the weight distribution matrix of scrap steel mechanical properties and scrap steel density changes, calculating the edge cutting weights between feature points, and generating a feature point edge cutting weight matrix; S312: normalizing the weights according to the feature point edge cutting weight matrix, optimizing the weight distribution of the feature points by comparing the relationship between the edge cutting weights and the normalization threshold, and generating the optimized grouping boundary of the feature points; S313: Based on the optimized grouping boundaries of the feature points, the intra-group connectivity and inter-group cutting edge weight difference between each group of feature points are calculated using the formula: ； Generate grouping and classification results of scrap steel feature points; in, Represents the weight distribution evaluation of scrap steel grouping and classification results, Represents the weight distribution between feature points within the group, Represents the optimized boundary weight within the feature point group, represents the difference in the weight of the cut edge between groups, is the total number of groups, is the number of feature points in each group, is the number of all feature points, is the number of inter-group cutting edge feature points.

6. The full-process scrap steel grading method based on data collection according to claim 5 is characterized in that: The steps for obtaining the scrap steel boundary samples and abnormal point distribution labels are specifically as follows: S411: extracting the scrap density change and scrap composition trend feature points of each group from the scrap steel grouping classification results, calculating the probability density value of each feature point by a local probability density estimation method, and generating a probability density distribution matrix; S412: Analyze the probability density distribution matrix, identify feature points whose probability density is lower than a set threshold as boundary samples, and obtain a boundary sample identification list; S413: Based on the boundary sample identification list, for feature points that are not marked as boundary samples, the probability density difference between the feature points and the nearest neighbor feature points is calculated using the formula: ； like Greater than the adjusted threshold , it is marked as an outlier, and scrap steel boundary samples and outlier distribution labels are generated; in, It represents the normalized difference between the standard deviation of the probability density average value of the i-th feature point and the k nearest neighbor feature points. is the probability density of the i-th feature point, is the probability density of the jth neighboring point of the ith feature point, is the number of neighboring points, is the standard deviation of the probability density of neighboring points, It is an empirically set threshold used to determine abnormal points.

7. The full-process scrap steel grading method based on data collection according to claim 6 is characterized in that: The steps for obtaining the optimization classification feature results are specifically as follows: S511: extracting the number of weighted adjacent points from the scrap steel boundary samples and the abnormal point distribution labels, adjusting the weights according to the abnormality degree of the multiple feature points, recalculating the weighted adjacent point number matrix, and generating an adjusted weighted adjacent point number matrix; S512: normalizing the adjusted weighted adjacent point quantity matrix, using linear transformation to evenly distribute the weight of each feature point, to obtain a normalized weighted adjacent point weight matrix; S513: Based on the normalized weighted neighboring point weight matrix, a kernel density estimation method is used to recalculate the scrap steel composition trend and the distribution of scrap steel mechanical property groups, using the formula: ； Generate optimized classification feature results; in, represents the optimized classification feature result of the kth group, is the kernel function, which is used to calculate the probability density contribution of feature points. is the attribute value of the i-th feature point, is the attribute value of the k-th group center, is the bandwidth of the kernel function, which controls the smoothness of the kernel density estimate. is the weight of the i-th feature point to the k-th group center, is the weight difference adjusted based on the distribution of outliers, is the weight adjustment coefficient, which is used to balance the influence of normal and abnormal feature points. Represents the total number of feature points involved in kernel density estimation.

8. The whole process scrap steel grading system based on data collection is characterized by: According to any one of claims 1 to 7, the whole process scrap steel grading method based on data collection comprises: The data feature extraction module calculates the local difference distribution value based on the scrap steel density change value, scrap steel composition trend value and scrap steel mechanical property value collected during the scrap steel processing process, extracts the density difference value, composition difference value and mechanical property difference value, normalizes the distribution value between feature points, and generates a low-dimensional feature embedding matrix; The association weight construction module calculates the density change value and the component trend value and the square of the Euclidean distance of the neighboring points based on the low-dimensional feature embedding matrix, normalizes the weight distribution matrix, extracts the total weight and generates the adjacency structure, and establishes a weighted adjacency graph matrix; The grouping and classification optimization module extracts the weight distribution matrix of mechanical characteristic values ​​and density change values ​​based on the weighted adjacency graph matrix, calculates the edge cutting weights between feature points and normalizes the matrix, determines the grouping boundary, divides the feature points into multiple groups, and generates a scrap steel feature grouping matrix; The abnormal distribution detection module calculates the probability density values ​​of multiple feature points in the group based on the scrap steel feature grouping matrix, marks the density extreme value points as boundary samples, calculates the neighboring distribution differences of isolated feature points and marks the abnormal points, and obtains the scrap steel boundary sample distribution and abnormal point labels; The feature grading optimization module adjusts the weight distribution and quantity of adjacent points based on the scrap steel boundary sample distribution and abnormal point labels, optimizes the grouping boundaries of component trend values ​​and mechanical property values, re-divides the feature point distribution, and generates optimized grading feature parameters.

Images