Converter lime consumption prediction method and system based on convolutional neural network

Abstract

The embodiment of the invention provides a converter lime consumption prediction method and system based on a convolutional neural network, and relates to the technical field of intelligent smelting processes, and the method can obtain converter production parameter information, and calculate a feature distance ranking matrix and a pixel distance ranking matrix according to the converter production parameter information. And performing feature-to-pixel mapping to obtain converter image data, and inputting the converter image data into the prediction model to obtain lime consumption prediction information output by the prediction model. According to the method, each sample in the converter production parameter information can be converted into an image, so that the convolutional neural network can perform multi-task learning by utilizing the space structure of the image, the prediction efficiency and accuracy of the converter lime consumption can be improved, and the prediction information can be used for a steel smelting process to improve the quality of molten steel and reduce the production cost; and thus, higher economic benefits are obtained.

Description

Technical Field

[0001] The present application relates to the field of intelligent smelting process technology, and in particular to a method and system for predicting converter lime consumption based on a convolutional neural network. Background Art

[0002] Slag-forming agents are additives used to form slag during metallurgical processes and are used in metal smelting and refining. Taking steelmaking as an example, the appropriate addition of slag-forming agents to molten steel can effectively improve its quality and reduce inclusion content. Lime is a slag-forming agent in converters, reducing slag viscosity, promoting the conversion of harmful elements, and protecting the converter lining. Precisely controlling lime consumption can improve smelting stability and molten steel quality while reducing production costs.

[0003] To accurately control lime consumption, lime consumption needs to be predicted during the smelting process. For example, the composition and weight of the slag can be calculated in real time by measuring the content of various components in the slag in the molten steel. For example, chemical analysis of slag samples using methods such as titration and inductively coupled plasma atomic emission spectroscopy (ICP-AES) can be performed to analyze components such as lime (CaO) in the slag. Based on the mass balance method, a slag composition calculation model is established, which can then be used to estimate lime consumption.

[0004] However, since the prediction process requires slag sampling and chemical composition analysis, the lime consumption prediction by detecting the slag content in the molten steel cannot accurately reflect the slag composition under complex furnace conditions. In addition, the prediction efficiency is low, making it difficult to achieve real-time consumption prediction, which is not conducive to dynamic adjustment of the converter smelting process. Summary of the Invention

[0005] In view of this, an embodiment of the present application provides a method and system for predicting converter lime consumption based on a convolutional neural network to solve the problem of inaccurate and low efficiency of lime consumption prediction results.

[0006] According to a first aspect of the present application, a method for predicting converter lime consumption based on a convolutional neural network is provided, the method comprising: Obtaining converter production parameter information, wherein the converter production parameter information includes at least molten iron temperature, molten iron volume, scrap steel volume, oxygen consumption, magnesium ball consumption, solid waste ball consumption, return ore consumption, lump ore consumption, and tapping temperature; Calculating a distance ranking matrix based on the converter production parameter information, wherein the distance ranking matrix includes a feature distance ranking matrix and a pixel distance ranking matrix; Performing feature-to-pixel mapping based on the distance ranking matrix to obtain converter image data; the converter image data is a set of pixel points obtained by mapping converter features in the converter production parameter information to pixels according to an optimal mapping relationship; the optimal mapping relationship is a feature-to-pixel mapping relationship obtained by performing feature exchange on the converter features in the distance ranking matrix and continuously iterating based on an error reduction rate after the feature exchange; The converter image data is input into a prediction model to obtain lime consumption prediction information output by the prediction model, where the prediction model is a convolutional neural network model obtained by training multiple optimizers using converter sample data.

[0007] According to a second aspect of the present application, a converter lime consumption prediction system based on a convolutional neural network is provided, the system comprising: An information acquisition module is used to obtain converter production parameter information, wherein the converter production parameter information includes at least molten iron temperature, molten iron volume, scrap steel volume, oxygen consumption, magnesium ball consumption, solid waste ball consumption, return ore consumption, lump ore consumption, and tapping temperature; A matrix calculation module, configured to calculate a distance ranking matrix based on the converter production parameter information, wherein the distance ranking matrix includes a feature distance ranking matrix and a pixel distance ranking matrix; an image mapping module for performing feature-to-pixel mapping based on the distance ranking matrix to obtain converter image data; the converter image data being a set of pixel points obtained by mapping converter features in the converter production parameter information to pixels according to an optimal mapping relationship; the optimal mapping relationship being a feature-to-pixel mapping relationship obtained by performing feature exchange on the converter features in the distance ranking matrix and continuously iterating based on an error reduction rate after the feature exchange; A prediction module is used to input the converter image data into a prediction model to obtain lime consumption prediction information output by the prediction model, wherein the prediction model is a convolutional neural network model obtained by training multiple optimizers using converter sample data.

[0008] According to a third aspect of the present application, a computer device is provided, comprising a storage medium, a processor, and a computer program stored on the storage medium and executable on the processor, wherein when the processor executes the program, the method for predicting converter lime consumption based on a convolutional neural network is implemented.

[0009] According to a fourth aspect of the present application, a storage medium is provided, on which a computer program is stored, and when the program is executed by a processor, the above-mentioned method for predicting converter lime consumption based on convolutional neural network is implemented.

[0010] By means of the above technical solution, the embodiment of the present application provides a method and system for predicting converter lime consumption based on a convolutional neural network. The method can obtain converter production parameter information and calculate a feature distance ranking matrix and a pixel distance ranking matrix based on the converter production parameter information. Feature-to-pixel mapping is then performed to obtain converter image data, and the converter image data is then input into a prediction model to obtain lime consumption prediction information output by the prediction model. The method can convert each sample in the converter production parameter information into an image, so that the convolutional neural network can use the spatial structure of the image for multi-task learning, thereby improving the prediction efficiency and accuracy of converter lime consumption. The prediction information can be used in steel smelting processes to improve the quality of molten steel, reduce production costs, and thereby achieve higher economic benefits.

[0011] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings: Figure 1 A flow chart of a method for predicting converter lime consumption based on a convolutional neural network provided in an embodiment of the present application; Figure 2 A schematic diagram of a converter lime consumption prediction process provided in an embodiment of the present application; Figure 3 A schematic diagram of a process for optimizing feature-to-pixel mapping using the GFIT algorithm provided in an embodiment of the present application; Figure 4 A schematic diagram of a process for exchanging converter features provided in an embodiment of the present application; Figure 5 A schematic diagram of the training prediction model process provided in an embodiment of the present application; Figure 6 A schematic diagram of a training curve for a converter lime consumption prediction model provided in an embodiment of the present application; Figure 7 A schematic diagram of a process for selecting a prediction model from multiple candidate models provided in an embodiment of the present application; Figure 8 A statistical comparison chart of the predicted values ​​of the converter lime consumption prediction model provided in the embodiment of the present application; Figure 9 Schematic diagram of the structure of the converter lime consumption prediction system based on convolutional neural network provided in an embodiment of the present application. DETAILED DESCRIPTION

[0013] The present application will be described in detail below with reference to the accompanying drawings and in combination with embodiments. It should be noted that, unless there is a conflict, the embodiments and features in the embodiments of the present application can be combined with each other.

[0014] In the embodiments of this application, a converter is a type of steelmaking equipment used to decarbonize, desiliconize, dephosphorize, and desulfurize molten iron from a blast furnace through oxidation reactions, adjusting its composition and ultimately producing molten steel. Decarbonization and desiliconization are achieved by blowing oxygen (O2) into the molten iron, oxidizing impurities such as carbon (C) and silicon (Si), which then form gases (CO, CO2) or enter the slag (SiO2). Dephosphorization and desulfurization are achieved by adding slag-forming agents such as calcium oxide (CaO) to convert phosphorus and sulfur into Ca3(PO4)2 and CaS, which then enter the slag.

[0015] Calcium oxide, also known as lime, is an alkaline slag-forming agent used in metallurgical processes to form slag, enabling metal smelting and refining. Taking steelmaking as an example, the appropriate addition of slag-forming agents to molten steel can effectively improve its quality and reduce its inclusion content. Lime is the primary slag-forming agent in converters. Lime helps raise furnace temperatures, reduce slag viscosity, promote the conversion of harmful elements, and protect the converter lining.

[0016] Since lime consumption is a key factor affecting slag properties, dephosphorization and desulfurization efficiency, and molten steel quality, precise control of lime consumption can improve smelting stability and molten steel quality while reducing production costs.

[0017] In some embodiments, to accurately control lime consumption, lime consumption needs to be predicted during the smelting process. For example, the real-time slag composition and weight can be calculated by measuring the content of various components in the slag in the molten steel. For example, slag samples can be taken for chemical analysis using methods such as titration or inductively coupled plasma atomic emission spectroscopy (ICP-AES). The slag is analyzed for components such as lime. Based on the mass balance method, a slag composition calculation model is established, which is then used to estimate lime consumption.

[0018] However, since the prediction process requires slag sampling and chemical composition analysis, the lime consumption prediction by detecting the slag content in the molten steel cannot accurately reflect the slag composition under complex furnace conditions. In addition, the prediction efficiency is low, making it difficult to achieve real-time consumption prediction, which is not conducive to dynamic adjustment of the converter smelting process.

[0019] In some embodiments, the Shared Nearest Neighbor Affinity Propagation (SNNAP) clustering algorithm can also be used to summarize and differentiate smelting data with similar furnace conditions to predict smelting data. For example, historical production data for the converter can be obtained and dimensionally filtered. A clustering model can then be constructed based on the SNN-AP algorithm. The filtered dataset can be input into the clustering model to obtain a set of cluster centers based on the model training dataset and the corresponding cluster datasets. A content prediction model can then be established for each dataset type. Each dataset type is then input into the corresponding prediction model for training, resulting in a mature prediction model. Data items corresponding to the converter and the comparison series are then collected and input into the prediction model for the corresponding category to obtain prediction results. However, when predicting smelting data based on the SNN-AP algorithm, the prediction model used is not further optimized, resulting in insufficient prediction stability, which in turn affects the prediction accuracy of different furnace condition results.

[0020] In order to solve the problem of inaccurate and inefficient lime consumption prediction results, some embodiments of this application provide a method for predicting converter lime consumption based on a convolutional neural network, and the method can be applied to electronic devices with data processing capabilities. The electronic devices include but are not limited to computers, servers, mobile terminals, smart wearable devices, industrial control machines, etc. For ease of description, in the embodiments of this application, electronic devices are used as the execution subjects of the method. It should be understood that the method can also be applied to other types of execution subjects, which are no longer shown one by one in the embodiments of this application. Figure 1 As shown, the method includes: S101. Obtain converter production parameter information.

[0021] To predict lime consumption, a model can be built and trained by collecting relevant parameters from the actual converter production process. This requires first acquiring converter production parameter information. This information refers to a set of parameters that influence lime consumption during the converter production process. This information includes at least molten iron temperature, molten iron volume, scrap steel volume, oxygen consumption, magnesium ball consumption, solid waste ball consumption, return ore consumption, lump ore consumption, and tapping temperature.

[0022] For example, when the converter lime consumption prediction method based on convolutional neural network (CNN) is applied to a 120t converter in a steel plant, parameter information of the actual converter production can be collected, that is, the converter production parameter information includes: molten iron temperature, molten iron volume, scrap steel volume, oxygen consumption, magnesium ball consumption, solid waste ball consumption, return ore consumption, lump ore consumption and tapping temperature as characteristic values, and lime consumption as the target value.

[0023] Converter production parameter information can be collected through sensors used during the production process and recorded through material consumption. For example, temperature information such as molten iron temperature and tapping temperature can be obtained through non-contact temperature sensors such as infrared thermal imagers. Quality information such as molten iron volume, scrap steel volume, oxygen consumption, magnesium ball consumption, solid waste ball consumption, return ore consumption, and lump ore consumption can be obtained through weighing sensors or material addition records.

[0024] To obtain converter production parameter information, the electronic device running the convolutional neural network-based converter lime consumption prediction method can establish a communication connection with a data source device within the converter production environment. The data source device refers to a device capable of providing converter production parameter information, and can include data acquisition devices and data storage devices. For example, the data acquisition device may include a temperature sensor and a weight sensor; the data storage device may include a memory for storing user-entered process parameters.

[0025] When acquiring converter production parameter information, the electronic device may send a data acquisition request to the data source device. After acquiring the data acquisition request, the data source device may respond to the data acquisition request and feed back the converter production parameter information to the electronic device.

[0026] It should be noted that the types of information included in the converter production parameter information are not limited to the aforementioned categories, such as molten iron temperature and molten iron volume. To obtain more accurate prediction results, the types of information included in the converter production parameter information can be appropriately increased or decreased based on the actual production process. Furthermore, the acquired converter production parameter information must meet a certain data volume. In some embodiments, the data volume of the converter production parameter information is greater than or equal to 800 heats.

[0027] For example, for a steel plant's 120t converter, the converter production parameter information includes molten iron temperature, molten iron volume, scrap steel volume, oxygen consumption, magnesium ball consumption, solid waste ball consumption, return ore consumption, lump ore consumption, and tapping temperature. Furthermore, the converter production parameter data volume covers 850 heats.

[0028] like Figure 2As shown, in some embodiments, the data fed back by the data source device to the electronic device may include multiple types of raw production parameters. After obtaining the raw production parameters fed back by the data source device, the electronic device needs to filter the raw production parameters to select appropriate input parameter information related to converter lime consumption and then preprocess the appropriate input parameter information. Preprocessing may include data outlier processing, data standardization, and correlation analysis.

[0029] Therefore, when acquiring converter production parameter information, the electronic device may first acquire the original production parameters and then extract valid parameters from the original production parameters based on a normal value range, wherein the normal value range is a screening range determined by using a boxplot method and an interquartile range with a preset magnification.

[0030] For example, outlier handling utilizes a boxplot approach, using 1.3-1.5 times the interquartile range (IQR) to represent the range of normal values. The interquartile range is a statistic that describes the degree of dispersion in data, representing the distribution range of the middle 50% of the data. The interquartile range is calculated by calculating the difference between the upper quartile (Q3) and the lower quartile (Q1), i.e., IQR = Q3 - Q1. To determine the interquartile range, the converter production parameter information can be sorted first, arranging each category of data in ascending order. Based on this sorting result, the lower quartile (Q1) is then determined as the 25th percentile, representing the cutoff point for the top 25% of the data. The upper quartile (Q3) is then determined as the 75th percentile, representing the cutoff point for the top 75% of the data.

[0031] During outlier detection, the interquartile range (IQR) is used to determine the normal range. The lower limit of the normal range is (Q1 - 1.5 × IQR), and the upper limit is (Q3 + 1.5 × IQR). By comparing each data point with the normal range, values ​​outside the normal range are considered outliers. By eliminating outliers from converter production parameter information, valid parameters can be extracted.

[0032] After obtaining the effective parameters, the electronic device can perform a correlation analysis on the effective parameters, that is, calculate a correlation coefficient based on the effective parameters. The correlation coefficient is the ratio of the covariance of the effective parameters and the lime consumption to the median value; the median value is the product of the standard deviation of the effective parameters and the standard deviation of the lime consumption.

[0033] For example, for parameter information determined to be valid parameters, the electronic device may perform a correlation analysis, wherein the correlation coefficient formula is: ; Where, ρ X,Yis the correlation coefficient of parameter information, cov ( X , Y ) is the input variable X and lime consumption Y covariance of σ X and σ Y Input variables X and lime consumption Y The standard deviation of .

[0034] Correlation analysis can be used to analyze the linear correlation between various input parameters collected during converter production (such as hot metal temperature, hot metal volume, and scrap steel quantity) and lime consumption. The results, calculated using the correlation coefficient formula, quantify the strength and direction of the correlation between different parameters and lime consumption. This helps identify key parameters that significantly influence lime consumption during preprocessing, while eliminating redundant or low-correlation parameters. This provides a more streamlined and efficient input foundation for optimizing feature mapping in subsequent processes, thereby improving the accuracy and efficiency of the overall model or mapping relationship and ensuring that key information of practical significance to the production process is retained when sample features are subsequently converted into images.

[0035] Therefore, after calculating the correlation coefficient, the electronic device can extract the converter production parameters that are correlated with the converter lime consumption from the effective parameters based on the correlation coefficient, and then perform normalization processing on the converter production parameters to obtain converter production parameter information.

[0036] For example, standardization can eliminate differences in data dimensions and distribution, making different features comparable, thereby improving the accuracy and efficiency of data analysis and modeling. Therefore, electronic equipment can perform standardization on converter production parameters based on the following formula to obtain converter production parameter information: ; Where, X i ' To standardize the results; X i The parameter information data to be processed after data screening; X mean It is the average value of parameter information data after data screening; X std is the variance of the parameter information data after data screening.

[0037] S102. Calculate a distance ranking matrix based on converter production parameter information.

[0038] After obtaining the converter production parameter information, a distance ranking matrix can be calculated based on the converter production parameter information. The distance ranking matrix includes a feature distance ranking matrix and a pixel distance ranking matrix. The feature distance ranking matrix is ​​obtained by calculating feature distances based on converter features extracted from the converter production parameter information and sorting based on the feature distances.

[0039] To obtain a feature distance ranking matrix, in some embodiments, when calculating the distance ranking matrix based on converter production parameter information, the electronic device may first extract converter features from the converter production parameter information to obtain a feature set. Then, a raw mapping image of a preset size is selected based on the converter features. The raw mapping image includes multiple pixel locations, and the product of the sizes of the selected raw mapping image can be greater than or equal to the total number of feature values, i.e., the number of pixel locations is greater than or equal to the number of converter features in the feature set.

[0040] For example, when the converter production parameter information includes nine parameter categories, namely, molten iron temperature, molten iron volume, scrap steel volume, oxygen consumption, magnesium ball consumption, solid waste ball consumption, return ore consumption, lump ore consumption, and tapping temperature, at least nine converter features can be extracted from the converter production parameter information. Therefore, an original mapping image with an image size of 3×3 can be selected, and each converter feature in the converter production parameter information is assigned to a pixel position in the image.

[0041] After selecting the original mapping image, the feature distance between any two converter features in the feature set can be calculated, and then the feature distances can be arranged in ascending order to obtain a feature distance ranking matrix.

[0042] For example, for any two converter features in the feature set, that is, features i and features j , the feature distance can be obtained by calculating the Euclidean distance, where the calculation formula of the Euclidean distance is: ; Where, d ij represents feature distance; X ik Representation characteristics i In the sample k (i.e., converter production parameter information or any data category in converter production parameter information); X jk Representation characteristics j In the sample k The value in ; M Represents a sample k The distances between all features are arranged in ascending order and ranked to obtain the feature distance ranking matrix R .

[0043] The pixel distance ranking matrix is ​​a matrix generated based on the distribution of pixel positions in the converter image and the distances between pixel positions in the converter image.

[0044] To obtain a pixel distance ranking matrix, in some embodiments, when calculating the distance ranking matrix based on converter production parameter information, the electronic device may traverse pixel positions in the original mapped image to obtain a set of pixel pairs. The set of pixel pairs includes a plurality of pixel pairs, each of which includes pixel coordinates of any two pixels in the original mapped image.

[0045] For example, for an original mapping image of size 3×3, which contains 9 pixel positions, each pixel position can be used to set a pixel point. Then, for any two different pixel positions among the 9 pixel positions, a pixel pair can be formed ( p i , p j ), the pixel pair includes p i and p j Two pixel points, the coordinates of the two pixel points are p i =( x i , y i )and p =( x j , y j ).

[0046] After obtaining the pixel pair set, the pixel distance of the pixel pair can be calculated based on the pixel coordinates. Then, according to the pixel position in the original mapped image, the pixel distance ranking matrix is ​​generated based on the pixel distance.

[0047] For example, for the pixel coordinates of any two pixels in the original mapped image p i =( x i , y i )and p =( x j , y j ), the spatial distance between pixels can be calculated to obtain the pixel distance. The pixel distance can also be obtained by calculating the Euclidean distance between two pixel positions, that is: ; Where, d (p i , p j ) represents pixel points p i and p j Pixel distance; ( x i , y i ) is a pixel p i Pixel coordinates of ; ( x j , y j ) is a pixel p j The pixel coordinates of . According to the calculated pixel distance values, the pixel distance ranking matrix can be obtained Q . That is, the pixel distance ranking matrix Q It can be expressed as: ; Feature distance ranking matrix calculated based on converter production parameter information R and pixel distance ranking matrix Q It includes data such as feature distance and pixel distance for characterizing distance relationships, which can be used for mapping features to pixels.

[0048] S103 , performing feature-to-pixel mapping based on the distance ranking matrix to obtain converter image data.

[0049] The feature distance ranking matrix calculated based on the converter production parameter information R and pixel distance ranking matrix Q After that, we can use the feature distance ranking matrix R and pixel distance ranking matrix Q Perform feature-to-pixel mapping to obtain converter image data, wherein the converter image data is a set of pixel points obtained by mapping converter features in converter production parameter information to pixels according to an optimal mapping relationship.

[0050] The optimal mapping relationship is obtained by performing feature swapping on the converter features in the distance ranking matrix and iterating based on the error reduction rate after the feature swap. Electronic devices can use the Greedy Feature Iterative Transformation (GFIT) algorithm to optimize the feature-to-pixel mapping.

[0051] Therefore, in some embodiments, when an electronic device performs feature-to-pixel mapping based on a distance ranking matrix to obtain converter image data, algorithm variables may be initialized. The algorithm variables include an iteration index, an initial error, a feature exchange record vector, and a feature ranking vector. The iteration index has an initial value of 0. The initial error is calculated based on the distance ranking matrix. The feature exchange record vector has an initial value of negative infinity. The lengths of the feature exchange record vector and the feature ranking vector are equal to the number of converter features.

[0052] For example, electronic devices can initialize algorithm variables and set iteration indexes ω =0, initial error e 0= err ( R , Q ), where the error represents the feature distance ranking matrix R and pixel distance ranking matrix Q The degree of deviation, that is, the error, can be calculated according to the following formula: ; in, err ( R , Q ) represents the feature distance ranking matrix R and pixel distance ranking matrix Q Error; N Indicates the total number of features; R i,j Represents the feature distance ranking matrix R The i Row, No. j Matrix elements of columns; Q i,j Represents the pixel distance ranking matrix Q The i Row, No. j Matrix elements of columns.

[0053] During the initialization of the algorithm variables, you can also define the feature exchange record vector and the feature sorting vector. ξ Initialize the vector of length 9 to negative infinity, the characteristic sort vector τ is [1, 2, 3, …, 9].

[0054] After initializing the algorithm variables, the electronic device can continuously iterate based on the algorithm variables, that is, continuously update the iteration index as the iteration process progresses, and perform feature exchange on the converter features based on the greedy algorithm and the feature distance ranking matrix, and calculate the error reduction rate during the feature exchange process.

[0055] like Figure 3As shown, in some embodiments, to perform continuous iteration, when performing feature exchange on converter features based on a greedy algorithm and a feature distance ranking matrix, the electronic device may select a first converter feature from the converter features based on a feature exchange record vector, and then determine a second converter feature using the greedy algorithm. The first converter feature is the converter feature that has not been exchanged for the longest time, as recorded in the feature exchange record vector. The second converter feature is the converter feature that, after exchange with the first converter feature, has the maximum iterative error.

[0056] For example, by updating the iteration index ω = ω +1, feature exchange record vector ξ Select the converter feature that has not been exchanged the longest in the converter feature set. f , which is the first converter feature. And the greedy algorithm (Greedy) is used to calculate the first converter feature. f After the exchange is completed, the new error can reach the maximum feature n , that is, the second converter characteristic. The second converter characteristic can be calculated according to the following formula: ; Where, f n Indicates the n The second converter characteristics when exchanging converter characteristics one by one; f Indicates the characteristics of the first converter; argmax () represents the maximum function, which is used to find the maximum value. e n Get the maximum input value; e n Indicates the n The error of exchanging converter characteristics one by one, that is, the iteration error e n = err ( R n , Q ); R n Indicates the n The feature distance ranking matrix is ​​exchanged one converter feature at a time.

[0057] After determining the first converter feature and the second converter feature, feature exchange can be performed on the first converter feature and the second converter feature. Figure 4As shown, in some embodiments, to perform feature exchange, the electronic device may calculate an iterative error based on the pixel distance ranking matrix and the feature distance ranking matrix after feature exchange, and calculate an error reduction rate based on the iterative error and the initial error. The electronic device then determines whether to perform converter feature exchange by comparing the error reduction rate with a preset minimum error reduction rate. If the error reduction rate is less than or equal to the preset minimum error reduction rate, the feature exchange record vector is updated. If the error reduction rate is greater than the preset minimum error reduction rate, the feature exchange record vector, feature ranking vector, and feature distance ranking matrix are updated.

[0058] For example, the minimum error reduction rate can be set according to the preset ε , calculate the error reduction rate after converter feature exchange, the error reduction rate can be calculated according to the following formula: ; in, represents the error reduction rate; err ( R , Q ) represents the initial error e 0, err ( R n , Q ) represents the iteration error e n , that is, n The error of exchanging converter features one by one.

[0059] By comparing the error reduction rate with the preset minimum error reduction rate, if the error reduction rate is greater than the minimum error reduction rate, the converter feature exchange is performed. Update the feature exchange record vector ξ , update the feature ranking vector τ And update the feature distance ranking matrix R Otherwise, only the feature exchange record vector is updated ξ , other parameters return to their original values.

[0060] After calculating the error reduction rate, the electronic device can also continue to iterate based on the error reduction rate until the error reduction rate converges to a preset error reduction rate threshold to obtain the optimal mapping relationship, and then map the converter features to pixels according to the optimal mapping relationship to obtain converter image data.

[0061] For example, when using the GFIT algorithm to optimize the mapping of features to pixels, the iteration index can be continuously updated iteratively, the converter feature that has not been exchanged for the longest time in the vector is selected, and the feature that can reach the maximum value of the new error after the exchange with the feature is completed is calculated feature by feature through the greedy algorithm, and the error reduction rate after the converter feature is exchanged is calculated, and the parameter update step is performed by comparing the error reduction rate with the set minimum error reduction rate until the error reduction rate of the algorithm converges to the preset error reduction rate threshold. t Until the optimal mapping from features to pixels (F2P MAPPING) is obtained, all features of each sample on the converter are then mapped to each pixel in the image through F2P MAPPING, and all samples are converted into images.

[0062] S104: Input the converter image data into the prediction model to obtain lime consumption prediction information output by the prediction model.

[0063] After obtaining converter image data through feature-to-pixel mapping, the electronic device can input the converter image data into a prediction model to obtain lime consumption prediction information output by the prediction model. The prediction model is a convolutional neural network model obtained by training multiple optimizers using converter sample data.

[0064] For example, the neural grid structure in the convolutional neural network of the prediction model is a two-layer convolution with a convolution kernel size of 2 × 2. L2 regularization can be used in the convolutional neural network to prevent overfitting. The L2 regularization coefficient used is 0.01, and the ReLU activation function is used.

[0065] like Figure 5 As shown, in some embodiments, the electronic device can pre-train a prediction model. Specifically, the electronic device can obtain converter sample data. The converter sample image includes a sample production parameter set and a lime consumption label assigned to the sample production parameter set. The converter sample data is then input into the trained model to obtain a predicted lime consumption value output by the trained model. A training loss is then calculated based on the predicted lime consumption value and the lime consumption label.

[0066] By comparing the training loss with the preset loss threshold, it is determined whether the prediction model has converged so as to perform iterative training. If the training loss is greater than the preset loss threshold, it means that the current prediction model has not been trained to convergence. Then, the model parameters of the trained model can be adjusted according to the training loss, and iterative training can be performed based on the trained model after adjusting the model parameters. After multiple iterative trainings, when the training loss corresponding to the trained model output is iterated to be less than or equal to the preset loss threshold, it is determined that the prediction model has converged. Figure 6 As shown, the model parameters of the trained model can be output to obtain a prediction model.

[0067] Based on the pre-trained prediction model, the electronic device can predict lime consumption. To this end, the electronic device can input converter image data into the prediction model. The prediction model can perform feature analysis on the input converter image data to predict lime consumption.

[0068] Therefore, by applying the technical solutions of the above embodiments, the method for predicting converter lime consumption based on a convolutional neural network provided by the above embodiments can realize converter lime consumption prediction based on GFIT and CNN. The method can convert each sample in the converter production parameter information in the form of tabular data into an image, so that the convolutional neural network (CNN) can use the spatial structure of the image for learning, effectively improving the prediction efficiency and accuracy of converter lime consumption, improving the quality of molten steel, reducing production costs, and thereby obtaining higher economic benefits.

[0069] In some embodiments, as a refinement and expansion of the specific implementation of the above embodiment, in order to fully illustrate the specific implementation process of this embodiment, some embodiments of this application also provide a method for predicting converter lime consumption based on a convolutional neural network. The method can use multiple optimizers to optimize the convolutional neural network and train the converted converter image sample data to obtain the final optimization model to predict the converter lime consumption. Figure 7 As shown, the method includes: S201, dividing the converter sample data into a training set and a test set; S202, using the training set to perform model training on multiple optimizers to obtain multiple candidate models based on different optimizers; S203, inputting the test set into the candidate model to obtain test output information; S204, calculating prediction indicators based on test output information; S205: Select a prediction model from multiple candidate models based on the prediction indicators.

[0070] To obtain a suitable prediction model, the electronic device can divide the converter sample data into a training set and a test set. For example, the converter sample data is stratified using shuffle processing, with 80% used as the training set and 20% used as the test set.

[0071] To divide the training and test sets, the training set can be used to train models using multiple optimizers to obtain multiple candidate models based on different optimizers. The test set is then input into the candidate models to obtain test output information. Prediction metrics are then calculated based on the test output information. These metrics include mean squared error and hit rate.

[0072] For example, convolutional neural networks use mean squared error (MSE) and hit rate (HR) as prediction indicators, where the mean squared error is calculated using the following formula: ; Where, MSE is the mean square error; y i is the label vector, is the predicted value; N is the total number of features.

[0073] ; Correspondingly, the hit rate can be calculated as follows: ; Where, error is the error between the actual input value and the predicted value, n y is the number of data whose error meets the threshold, N is the total number of features, that is, the total number of samples.

[0074] After calculating the prediction index, the electronic device selects a prediction model from multiple candidate models based on the prediction index. The prediction model includes an optimal optimizer that ensures that the hit rate of the mean square error of the candidate model's prediction of converter lime consumption within a preset quality range is greater than or equal to a hit rate threshold.

[0075] For example, the multiple optimizers used in the model training process are Adagrad, Nadam, RMsprop, SGD and Adam. The best optimizer makes the hit rate of the convolutional neural network predicting the converter lime consumption with an error of ±20kg greater than 90%.

[0076] By applying the technical solutions of the above embodiments, the converter lime consumption prediction method based on convolutional neural network provided in the above embodiments can select appropriate convolutional neural network and optimizer for different furnace conditions to improve the accuracy of converter lime consumption prediction. The optimizer whose hit rate of error in the preset quality interval is greater than or equal to the hit rate threshold is determined as the best optimizer, thereby improving the stability of smelting production, ensuring the quality of molten steel, and at the same time reducing production costs and improving economic benefits. Figure 8 As shown in the statistical comparison chart of the predicted values ​​of the converter lime consumption prediction model, it can be seen that the prediction performance of the convolutional neural network is better than that of the BP model.

[0077] In some embodiments, as a specific implementation of the converter lime consumption prediction method based on convolutional neural network described in the above embodiments, some embodiments of the present application also provide a converter lime consumption prediction system based on convolutional neural network, such as Figure 9 As shown, the system includes: An information acquisition module is used to obtain converter production parameter information, wherein the converter production parameter information includes at least molten iron temperature, molten iron volume, scrap steel volume, oxygen consumption, magnesium ball consumption, solid waste ball consumption, return ore consumption, lump ore consumption, and tapping temperature; A matrix calculation module is used to calculate a distance ranking matrix based on converter production parameter information, wherein the distance ranking matrix includes a feature distance ranking matrix and a pixel distance ranking matrix; An image mapping module is configured to perform feature-to-pixel mapping based on a distance ranking matrix to obtain converter image data; the converter image data is a set of pixel points obtained by mapping converter features in converter production parameter information to pixels according to an optimal mapping relationship; the optimal mapping relationship is a feature-to-pixel mapping relationship obtained by performing feature exchange on the converter features in the distance ranking matrix and continuously iterating based on an error reduction rate after the feature exchange; The prediction module is used to input the converter image data into the prediction model to obtain the lime consumption prediction information output by the prediction model. The prediction model is a convolutional neural network model obtained by training multiple optimizers using converter sample data.

[0078] By applying the technical solutions of the above-mentioned embodiments, the convolutional neural network-based converter lime consumption prediction system provided in the above-mentioned embodiments can obtain converter production parameter information through the information acquisition module, and the matrix calculation module calculates the feature distance ranking matrix and the pixel distance ranking matrix based on the converter production parameter information. The image mapping module then performs feature-to-pixel mapping to obtain converter image data, and the prediction module then inputs the converter image data into the prediction model to obtain lime consumption prediction information output by the prediction model. The system can convert each sample in the converter production parameter information into an image, allowing the convolutional neural network to utilize the spatial structure of the image for multi-task learning, thereby improving the efficiency and accuracy of converter lime consumption prediction. The prediction information can be used in the steelmaking process to improve the quality of molten steel, reduce production costs, and thus achieve higher economic benefits.

[0079] It should be noted that for other corresponding descriptions of the functional units involved in the converter lime consumption prediction system based on convolutional neural network provided in the embodiment of the present application, reference can be made to the corresponding descriptions in the converter lime consumption prediction method based on convolutional neural network provided in the above embodiment, and will not be repeated here.

[0080] The embodiment of the present application also provides a computer device, which can be specifically a personal computer, a server, a network device, etc. The computer device includes a bus, a processor, a memory and a communication interface, and may also include an input and output interface and a display device. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and computer program in the non-volatile storage medium. The database of the computer device is used to store location information. The network interface of the computer device is used to communicate with an external terminal via a network connection. When the computer program is executed by the processor, the steps in each method embodiment are implemented.

[0081] Those skilled in the art will understand that the structure of the above-mentioned computer device is only a partial structure related to the solution of the present application and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components, or combine certain components, or have a different component arrangement.

[0082] In one embodiment, a computer-readable storage medium is further provided. The computer-readable storage medium may be non-volatile or volatile, and stores a computer program thereon. When the computer program is executed by a processor, the steps in the above-mentioned method embodiments are implemented.

[0083] In one embodiment, a computer program product is further provided, including a computer program, which implements the steps in the above method embodiments when executed by a processor.

[0084] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties.

[0085] Those skilled in the art will understand that all or part of the processes in the above-mentioned embodiment methods can be implemented by instructing related hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods.

[0086] Any reference to a memory, database, or other medium used in the embodiments provided herein may include at least one of a non-volatile memory and a volatile memory. Non-volatile memory may include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc.

[0087] Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM may be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).

[0088] The database involved in each embodiment provided herein may include at least one of a relational database and a non-relational database. Non-relational databases may include, but are not limited to, distributed databases based on blockchain. The processor involved in each embodiment provided herein may be, but is not limited to, a general-purpose processor, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, and the like.

[0089] The technical features of the above embodiments can be combined arbitrarily. In order to make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0090] The above-described embodiments merely represent several implementation methods of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present application. It should be noted that a person of ordinary skill in the art may make various modifications and improvements without departing from the spirit of the present application, and these modifications and improvements fall within the scope of protection of the present application. Therefore, the scope of protection of the present application shall be determined by the appended claims.

Claims

1. A method for predicting converter lime consumption based on convolutional neural network, characterized in that: The method comprises: Obtaining converter production parameter information, wherein the converter production parameter information includes at least molten iron temperature, molten iron volume, scrap steel volume, oxygen consumption, magnesium ball consumption, solid waste ball consumption, return ore consumption, lump ore consumption, and tapping temperature; Calculating a distance ranking matrix based on the converter production parameter information, wherein the distance ranking matrix includes a feature distance ranking matrix and a pixel distance ranking matrix; Performing feature-to-pixel mapping based on the distance ranking matrix to obtain converter image data; the converter image data is a set of pixel points obtained by mapping converter features in the converter production parameter information to pixels according to an optimal mapping relationship; the optimal mapping relationship is a feature-to-pixel mapping relationship obtained by performing feature exchange on the converter features in the distance ranking matrix and continuously iterating based on an error reduction rate after the feature exchange; The converter image data is input into a prediction model to obtain lime consumption prediction information output by the prediction model, where the prediction model is a convolutional neural network model obtained by training multiple optimizers using converter sample data.

2. The method according to claim 1, characterized in that Obtain converter production parameter information, including: Obtain original production parameters; Extracting effective parameters from the original production parameters based on a normal value range, wherein the normal value range is a screening range determined by using a box plot method and an interquartile range with a preset magnification; Calculating a correlation coefficient based on the effective parameter, the correlation coefficient being the ratio of the covariance of the effective parameter and the lime consumption to an intermediate value; the intermediate value being the product of the standard deviation of the effective parameter and the standard deviation of the lime consumption; Extracting the converter production parameter associated with the converter lime consumption from the effective parameters according to the correlation coefficient; A standardization process is performed on the converter production parameters to obtain the converter production parameter information.

3. The method according to claim 1, characterized in that Calculating a distance ranking matrix according to the converter production parameter information includes: extracting converter features from the converter production parameter information to obtain a feature set; selecting an original mapping image of a preset size according to the converter feature, the original mapping image comprising a plurality of pixel positions; the number of the pixel positions being greater than or equal to the number of the converter features in the feature set; Calculating a feature distance between any two converter features in the feature set; Arrange the feature distances in ascending order to obtain the feature distance ranking matrix.

4. The method according to claim 3, characterized in that Calculating a distance ranking matrix according to the converter production parameter information includes: Traversing the pixel positions in the original mapped image to obtain a pixel pair set, wherein the pixel pair set includes a plurality of pixel pairs, and the pixel pairs include pixel coordinates of any two pixel points in the original mapped image; Calculating the pixel distance of the pixel pair according to the pixel coordinates; The pixel distance ranking matrix is ​​generated according to the pixel positions in the original mapping image and the pixel distances.

5. The method according to claim 1, wherein Performing feature-to-pixel mapping based on the distance ranking matrix to obtain converter image data, including: Initialize algorithm variables, including an iteration index, an initial error, a feature exchange record vector, and a feature sorting vector; the initial value of the iteration index is 0; the initial error is calculated based on the distance ranking matrix; the initial value of the feature exchange record vector is negative infinity; the vector lengths of the feature exchange record vector and the feature sorting vector are equal to the number of converter features; updating the iteration index, and performing feature exchange on the converter features based on a greedy algorithm and the feature distance ranking matrix; Calculate the error reduction rate during feature exchange; Continuously iterating based on the error reduction rate until the error reduction rate converges to a preset error reduction rate threshold, so as to obtain the optimal mapping relationship; According to the optimal mapping relationship, the converter features are mapped to pixels to obtain the converter image data.

6. The method according to claim 5, characterized in that Based on the greedy algorithm and the feature distance ranking matrix, feature exchange is performed on the converter features, including: selecting a first converter feature from the converter features according to the feature exchange record vector, wherein the first converter feature is the converter feature that has not been exchanged for the longest time and is recorded in the feature exchange record vector; determining a second converter feature by a greedy algorithm, wherein the second converter feature is a converter feature whose iterative error reaches a maximum value after being exchanged with the first converter feature; A feature swap is performed on the first converter feature and the second converter feature.

7. The method according to claim 6, characterized in that Performing a feature swap on the first converter feature and the second converter feature includes: Calculating an iteration error based on the pixel distance ranking matrix and the feature distance ranking matrix after performing feature exchange; Calculating an error reduction rate based on the iterative error and the initial error; If the error reduction rate is less than or equal to a preset minimum error reduction rate, updating the feature exchange record vector; If the error reduction rate is greater than the preset minimum error reduction rate, the feature exchange record vector, the feature sorting vector and the feature distance ranking matrix are updated.

8. The method according to claim 1, characterized in that The method further comprises: Acquire converter sample data, wherein the converter sample image includes a sample production parameter set and a lime consumption label set for the sample production parameter set; Inputting the converter sample data into a trained model to obtain a lime consumption prediction value output by the trained model; Calculating a training loss based on the lime consumption prediction value and the lime consumption label; If the training loss is greater than a preset loss threshold, adjusting the model parameters of the trained model according to the training loss, and performing iterative training based on the trained model after adjusting the model parameters; If the training loss is less than or equal to a preset loss threshold, the model parameters of the trained model are output to obtain the prediction model.

9. The method according to claim 1, characterized in that The method further comprises: Dividing the converter sample data into a training set and a test set; Using the training set to perform model training on multiple optimizers to obtain multiple candidate models based on different optimizers; Inputting the test set into the candidate model to obtain test output information; Calculating prediction indicators based on the test output information, wherein the prediction indicators include mean square error and hit rate; The prediction model is selected from multiple alternative models based on the prediction index, and the prediction model includes an optimal optimizer, which ensures that the hit rate of the mean square error of the alternative model in predicting the converter lime consumption in a preset quality range is greater than or equal to a hit rate threshold.

10. A converter lime consumption prediction system based on convolutional neural network, characterized in that: The system comprises: An information acquisition module is used to obtain converter production parameter information, wherein the converter production parameter information includes at least molten iron temperature, molten iron volume, scrap steel volume, oxygen consumption, magnesium ball consumption, solid waste ball consumption, return ore consumption, lump ore consumption, and tapping temperature; A matrix calculation module, configured to calculate a distance ranking matrix based on the converter production parameter information, wherein the distance ranking matrix includes a feature distance ranking matrix and a pixel distance ranking matrix; an image mapping module for performing feature-to-pixel mapping based on the distance ranking matrix to obtain converter image data; the converter image data being a set of pixel points obtained by mapping converter features in the converter production parameter information to pixels according to an optimal mapping relationship; the optimal mapping relationship being a feature-to-pixel mapping relationship obtained by performing feature exchange on the converter features in the distance ranking matrix and continuously iterating based on an error reduction rate after the feature exchange; A prediction module is used to input the converter image data into a prediction model to obtain lime consumption prediction information output by the prediction model, wherein the prediction model is a convolutional neural network model obtained by training multiple optimizers using converter sample data.

Images