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

Abstract

The embodiment of the application provides a converter lime consumption prediction method and system based on a convolutional neural network, relates to the technical field of intelligent smelting processes, and can acquire converter production parameter information and calculate a feature distance ranking matrix and a pixel distance ranking matrix according to the converter production parameter information. Mapping from features to pixels is then performed to obtain converter image data, and then the converter image data is 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 to perform multi-task learning, the prediction efficiency and accuracy of the converter lime consumption can be improved, the prediction information can be used for a steel smelting process to improve the quality of molten steel, reduce production costs, and thus obtain higher economic benefits.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent smelting process, in particular to a converter lime consumption prediction method and system based on a convolutional neural network. BACKGROUND

[0002] A slag former is an additive used in the metallurgical process to form molten slag, which can be used in metal smelting and refining processes. For example, in the steel smelting process, by properly adding a slag former to the molten steel, the quality of the molten steel can be effectively improved, and the content of inclusions in the steel can be reduced. Lime is a converter slag former that can reduce the viscosity of the slag, promote the conversion of harmful elements, and protect the converter lining. By accurately controlling the lime consumption, the smelting stability and molten steel quality can be improved, and the production cost can be reduced.

[0003] In order to accurately control the lime consumption, it is necessary to predict the lime consumption during the smelting process. For example, the content of various components contained in the slag in the molten steel can be detected to calculate the real-time slag composition and weight. For example, by titration, inductively coupled plasma atomic emission spectroscopy (ICP-AES), and other methods, the slag sample is sampled for chemical analysis, the components such as lime (CaO) in the slag are analyzed, and based on the mass balance method, a slag composition calculation model is established, and then the slag composition calculation model is used to estimate the lime consumption.

[0004] However, since the prediction process requires sampling of the slag and chemical composition analysis, the lime consumption prediction by detecting the content of the slag components in the molten steel cannot accurately reflect the slag composition under complex furnace conditions, and 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

[0005] Therefore, the embodiments of the present application provide a converter lime consumption prediction method and system based on a convolutional neural network to solve the problem of inaccurate and low-efficiency lime consumption prediction results.

[0006] According to a first aspect of the present application, a converter lime consumption prediction method based on a convolutional neural network is provided, the method comprising:

[0007] obtaining converter production parameter information, the converter production parameter information at least including hot metal temperature, hot metal quantity, scrap steel quantity, oxygen consumption, magnesium ball consumption, solid waste ball consumption, return ore consumption, lump ore consumption, and tapping temperature;

[0008] calculate a distance ranking matrix according to the converter production parameter information, the distance ranking matrix comprising a feature distance ranking matrix and a pixel distance ranking matrix;

[0009] perform feature-to-pixel mapping based on the distance ranking matrix to obtain converter image data; the converter image data is a pixel point set obtained by mapping converter features in the converter production parameter information to pixels in a best mapping relationship; the best 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 according to error reduction rates after feature exchange;

[0010] input the converter image data into a prediction model to obtain lime consumption prediction information output by the prediction model, the prediction model being a convolutional neural network model obtained by training multiple optimizers using converter sample data.

[0011] According to a second aspect of the present application, a converter lime consumption prediction system based on a convolutional neural network is provided, and the system comprises:

[0012] an information acquisition module configured to acquire converter production parameter information, the converter production parameter information comprising at least molten iron temperature, molten iron quantity, scrap steel quantity, oxygen consumption quantity, magnesium ball consumption quantity, solid waste ball consumption quantity, return ore consumption quantity, lump ore consumption quantity, and tapping temperature;

[0013] a matrix calculation module configured to calculate a distance ranking matrix according to the converter production parameter information, the distance ranking matrix comprising a feature distance ranking matrix and a pixel distance ranking matrix;

[0014] an image mapping module configured to perform feature-to-pixel mapping based on the distance ranking matrix to obtain converter image data; the converter image data is a pixel point set obtained by mapping converter features in the converter production parameter information to pixels in a best mapping relationship; the best 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 according to error reduction rates after feature exchange;

[0015] a prediction module configured to input the converter image data into a prediction model to obtain lime consumption prediction information output by the prediction model, the prediction model being a convolutional neural network model obtained by training multiple optimizers using converter sample data.

[0016] 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 in the storage medium and executable on the processor, wherein the processor implements the above-mentioned converter lime consumption prediction method based on a convolutional neural network when executing the program.

[0017] According to a fourth aspect of the present application, a storage medium is provided, which stores a computer program, the program being executed by a processor to implement the above-mentioned converter lime consumption prediction method based on a convolutional neural network.

[0018] According to the above technical solution, the present application provides a converter lime consumption prediction method and system 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 according to the converter production parameter information. Then, a feature-to-pixel mapping is performed to obtain converter image data, and then the converter image data is 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, which can improve the prediction efficiency and accuracy of the converter lime consumption. The prediction information can be used for steel smelting processes to improve the quality of molten steel and reduce production costs, thereby achieving higher economic benefits.

[0019] The above description is only a summary of the technical solutions of the present application. In order to more clearly understand the technical means of the present application, the present application can be implemented according to the content of the specification, and in order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the following specific embodiments of the present application are described. BRIEF DESCRIPTION OF DRAWINGS

[0020] The accompanying drawings described herein are used to provide further understanding of the present application, and form a part of the present application. The schematic 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:

[0021] Figure 1 A converter lime consumption prediction method based on a convolutional neural network is provided for the embodiments of the present application;

[0022] Figure 2 A converter lime consumption prediction method based on a convolutional neural network is provided for the embodiments of the present application;

[0023] Figure 3 A converter lime consumption prediction method based on a convolutional neural network is provided for the embodiments of the present application;

[0024] Figure 4 A converter lime consumption prediction method based on a convolutional neural network is provided for the embodiments of the present application;

[0025] Figure 5 A converter lime consumption prediction method based on a convolutional neural network is provided for the embodiments of the present application;

[0026] Figure 6A converter lime consumption prediction model training curve schematic diagram provided by an embodiment of the present application;

[0027] Figure 7 A process schematic diagram for selecting a prediction model from multiple alternative models provided by an embodiment of the present application;

[0028] Figure 8 A prediction value statistical comparison diagram of a converter lime consumption prediction model provided by an embodiment of the present application;

[0029] Figure 9 A converter lime consumption prediction system structure schematic diagram based on a convolutional neural network provided by an embodiment of the present application. DETAILED DESCRIPTION

[0030] Hereinafter, the present application will be described in detail with reference to the accompanying drawings and in conjunction with embodiments. It should be noted that the embodiments in the present application and the features in the embodiments can be combined with each other without conflict.

[0031] In an embodiment of the present application, the converter is a kind of steel smelting equipment, which is used to decarburize, desiliconize, dephosphorize, desulfurize the blast furnace molten iron through oxidation reaction, and adjust the composition, and finally smelt into molten steel. Among them, decarburization and desiliconization are achieved by blowing oxygen (O2) into the molten iron, so that carbon (C), silicon (Si) and other impurity elements are oxidized to form gas (CO, CO2) or enter the slag (SiO2). Dephosphorization and desulfurization are achieved by adding calcium oxide (CaO) and other slagging agents to convert phosphorus and sulfur into Ca3(PO4)2 and CaS to form into slag.

[0032] Calcium oxide, also known as lime, as a basic slagging agent, can be used to form molten slag in the metallurgical process to realize metal smelting and refining process. For example, in the steel smelting process, by properly adding slagging agent to the molten steel, the quality of the molten steel can be effectively improved, and the content of inclusions in the steel can be reduced. The converter slagging agent is mainly lime, which can improve the furnace temperature, reduce the viscosity of the slag, promote the conversion of harmful elements, and protect the lining of the converter in the converter smelting.

[0033] Since the consumption of lime is a key factor affecting the performance of the slag, the efficiency of dephosphorization and desulfurization, and the quality of the molten steel, accurate control of the lime consumption can improve the smelting stability and the quality of the molten steel, while reducing the production cost.

[0034] In some embodiments, in order to accurately control the lime consumption, it is necessary to predict the lime consumption in the smelting process. For example, the real-time slag composition and weight can be calculated by detecting the content of various components contained in the slag in the molten steel. For example, by titration, inductively coupled plasma atomic emission spectroscopy (ICP-AES) and other methods, the slag sample is sampled for chemical analysis, the lime and other components in the slag are analyzed, and based on the mass balance method, a slag composition calculation model is established, and then the lime consumption is estimated using the slag composition calculation model.

[0035] However, since the prediction process requires sampling of the slag and chemical composition analysis, when the lime consumption is predicted by detecting the content of the slag composition in the molten steel, the slag composition under complex furnace conditions cannot be accurately reflected, the prediction efficiency is low, real-time consumption prediction is difficult to achieve, and dynamic adjustment of the converter smelting process is not conducive.

[0036] In some embodiments, the Shared Nearest Neighbor Affinity Propagation (SNNAP) clustering algorithm can also be used to classify and distinguish smelting data of similar furnace conditions to predict smelting data. For example, historical production data of the converter can be obtained and dimensionally filtered, and then a clustering model can be constructed based on the SNN-AP algorithm, the filtered data set can be input into the clustering model, and the clustering center point set and the corresponding clustering data set based on the model training data set can be obtained. A content prediction model is established for each type of data set. Each category of data set is input into the corresponding prediction model for training to obtain a mature prediction model. Then, the data items corresponding to the comparison series of the converter are collected and input into the corresponding prediction model to obtain the prediction result. However, when the SNN-AP algorithm is used to predict smelting data, the prediction model used is not further optimized, which affects the stability of the prediction and further affects the prediction accuracy of different furnace conditions.

[0037] To solve the problem of inaccurate and low-efficiency lime consumption prediction, a converter lime consumption prediction method based on a convolutional neural network is provided in some embodiments of the present application. 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 the present application, the electronic device is taken as the execution subject of the method. It should be understood that the method can also be applied to other types of execution subjects, which will not be described one by one in the embodiments of the present application. As shown in the figure, the method comprises the following steps: Figure 1 ​

[0038] S101、acquire converter production parameter information.

[0039] In order to realize the prediction of lime consumption, the model can be established and trained by collecting relevant parameters of the actual production process of the converter, that is, the converter production parameter information needs to be acquired first. The converter production parameter information refers to a set of parameter information related to lime consumption in the converter production process. The converter production parameter information at least includes molten iron temperature, molten iron quantity, scrap steel quantity, oxygen consumption, magnesium ball consumption, solid waste ball consumption, return ore consumption, lump ore consumption and tapping temperature.

[0040] For example, when the converter lime consumption prediction method based on the convolutional neural network (CNN) is applied to a 120t converter in a certain steel plant, the parameter information of the actual production of the converter can be collected, that is, the converter production parameter information includes: molten iron temperature, molten iron quantity, scrap steel quantity, 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 target value.

[0041] The converter production parameter information can be obtained by sensor equipment in the production process and by material consumption record. For example, for temperature information such as molten iron temperature and tapping temperature, non-contact temperature sensors such as infrared thermal imagers can be used for detection. For mass information such as molten iron quantity, scrap steel quantity, oxygen consumption, magnesium ball consumption, solid waste ball consumption, return ore consumption and lump ore consumption, weighing sensors or material addition records can be used for detection.

[0042] In order to acquire the converter production parameter information, the electronic device running the converter lime consumption prediction method based on the convolutional neural network can establish a communication connection with the data source device in the converter production environment. The data source device refers to a device capable of providing converter production parameter information, which can include data acquisition equipment and data storage equipment. For example, the data acquisition equipment can include temperature sensors and weight sensors; the data storage equipment can include a memory for storing user input process parameters.

[0043] When acquiring the converter production parameter information, the electronic device can send a data acquisition request to the data source device, and the data source device can feed back the converter production parameter information to the electronic device in response to the data acquisition request after acquiring the data acquisition request.

[0044] It should be noted that the information categories included in the converter production parameter information are not limited to the above-mentioned categories such as hot metal temperature, hot metal quantity, etc. In order to obtain more accurate prediction results, the information categories included in the converter production parameter information can also be appropriately increased or reduced according to the actual production process. Moreover, the obtained converter production parameter information should meet a certain amount of data, and in some embodiments, the data amount of the converter production parameter information is greater than or equal to 800 heats.

[0045] For example, for a 120t converter of a certain steel plant, the converter production parameter information includes hot metal temperature, hot metal quantity, scrap quantity, oxygen consumption, magnesium ball consumption, solid waste ball consumption, return ore consumption, lump ore consumption, and tapping temperature. Moreover, the data amount of the converter production parameter information is 850 heats.

[0046] As shown in FIG. 1, in some embodiments, the data fed back by the data source device to the electronic device can include various types of original production parameters. Then, after obtaining the original production parameters fed back by the data source device, the electronic device needs to screen the original production parameters to select appropriate input parameter information related to the converter lime consumption, and then pre-process the appropriate input parameter information data. The pre-processing can include data outlier processing, data standardization, and correlation analysis, etc. Figure 2

[0047] Therefore, when the electronic device performs the obtaining of the converter production parameter information, the electronic device can first obtain the original production parameters, and then extract effective parameters from the original production parameters based on a normal value range. The normal value range is a screening range determined by the box plot method according to a preset interquartile range.

[0048] For example, the outlier processing uses the box plot method to represent the range of normal values using 1.3-1.5 times the interquartile range (IQR). The interquartile range is a statistical quantity describing the degree of data dispersion, representing the distribution range of the middle 50% of the data. The interquartile range is obtained by calculating the difference between the upper quartile (Q3) and the lower quartile (Q1), i.e. IQR = Q3-Q1. When determining the interquartile range, the converter production parameter information can be sorted first, and each category of data in the converter production parameter information can be arranged in ascending order. Then, according to the arrangement result, the lower quartile Q1 is determined as the 25th percentile, i.e. the boundary point of the first 25% of the data. Then, the upper quartile Q3 is determined as the 75th percentile, i.e. the boundary point of the first 75% of the data.

[0049] ​In the execution of the outlier detection process, the normal value range can be determined according to the interquartile range. Among them, the lower limit of the normal value range is (Q1-1.5xIQR); the upper limit is (Q3+1.5xIQR). Then by comparing each data with the normal value range, the values exceeding the normal value range are regarded as outliers. By eliminating the outliers in the converter production parameter information, the effective parameters can be extracted.

[0050] After obtaining the effective parameters, the electronic device can perform correlation analysis on the effective parameters, that is, calculate the correlation coefficient according to the effective parameters. Among them, the correlation coefficient is the ratio of the covariance of the effective parameters and the lime consumption to the intermediate value; the intermediate value is the product of the standard deviation of the effective parameters and the standard deviation of the lime consumption.

[0051] For example, for the parameter information determined as the effective parameter, the electronic device can perform correlation analysis, and the correlation coefficient formula is:

[0052] ;

[0053] In the formula, p X,Y is the correlation coefficient of the parameter information, cov ( X , Y ) is the covariance of the input variable X and the lime consumption Y ; s X and s Y are the standard deviations of the input variable X and the lime consumption Y , respectively.

[0054] Correlation analysis can be used to analyze the degree of linear correlation between each input parameter (such as hot metal temperature, hot metal quantity, scrap steel quantity, etc.) collected in the converter production and the lime consumption. The result calculated by the correlation coefficient formula can quantify the correlation strength and direction between different parameters and the lime consumption, which helps to identify key parameters that have a significant impact on the lime consumption in the preprocessing stage, while excluding redundant or highly correlated parameters. Provide a more streamlined and effective input basis for subsequent process optimization feature mapping, thereby improving the accuracy and efficiency of the overall model or mapping relationship, and ensuring that the key information with practical significance for the production process is retained when the sample features are converted into images.

[0055] Therefore, after calculating the correlation coefficient, the electronic device can extract the converter production parameters having a correlation relationship with the converter lime consumption from the effective parameters according to the correlation coefficient. Then perform standardization processing on the converter production parameters to obtain the converter production parameter information.

[0056] For example, the standardization processing can eliminate data dimension and distribution differences, make different features comparable, and thus improve the accuracy and efficiency of data analysis and modeling. Therefore, the electronic device can perform standardization processing on the converter production parameters based on the following formula to obtain converter production parameter information:

[0057] ;

[0058] In the formula, X i ' is the standardization result; X i is the parameter information data after data filtering to be processed; X mean is the average value of the parameter information data after data filtering; X std is the variance of the parameter information data after data filtering.

[0059] S102, calculating a distance ranking matrix according to the converter production parameter information.

[0060] After obtaining the converter production parameter information, a distance ranking matrix can be calculated according to the converter production parameter information, wherein the distance ranking matrix includes a feature distance ranking matrix and a pixel distance ranking matrix. The feature distance ranking matrix is calculated according to the feature distance of the converter features extracted from the converter production parameter information, and is obtained by sorting based on the feature distance.

[0061] In order to obtain the feature distance ranking matrix, in some embodiments, when the electronic device performs the step of calculating a distance ranking matrix according to the converter production parameter information, it can first extract the converter features from the converter production parameter information to obtain a feature set. Then, a pre-set size of the original mapping image is selected according to the converter features. The original mapping image includes a plurality of pixel positions, and the product of the size of the selected original mapping image can be greater than or equal to the total number of feature values, i.e., the number of pixel positions is greater than or equal to the number of converter features in the feature set.

[0062] For example, when the converter production parameter information includes 9 parameter categories of hot metal temperature, hot metal quantity, scrap steel quantity, oxygen consumption, magnesium ball consumption, solid waste ball consumption, return ore consumption, lump ore consumption, and tapping temperature, at least 9 converter features can be extracted from the converter production parameter information, so that an original mapping image with an image size of 3x3 can be selected, and each converter feature in the converter production parameter information can be assigned to a pixel position in the image.

[0063] After selecting the original mapping image, the feature distance between any two converter features in the feature set can be calculated, and the feature distance can be arranged in ascending order to obtain the feature distance ranking matrix.

[0064] For example, for any two converter features in the feature set, i.e., features i and characteristics j The feature distance can be obtained by calculating the Euclidean distance, where the formula for calculating the Euclidean distance is:

[0065] ;

[0066] In the formula, d ij Indicates the feature distance; X ik Representation of features i In the sample k (i.e., the value in the converter production parameter information or any data category within the converter production parameter information); X jk Representation of features j In the sample k The value in; M Indicates sample k The capacity is determined by sorting the distances between all features in ascending order and ranking them, resulting in a feature distance ranking matrix. R .

[0067] The pixel distance ranking matrix is ​​a matrix generated based on the pixel position distribution in the converter image and the distance between pixel positions in the converter image.

[0068] To obtain the pixel distance ranking matrix, in some embodiments, when the electronic device performs the calculation of the distance ranking matrix based on converter production parameter information, it can traverse the pixel positions in the original mapped image to obtain a set of pixel pairs. The set of pixel pairs includes multiple pixel pairs, and each pixel pair includes the pixel coordinates of any two pixels in the original mapped image.

[0069] For example, given a 3×3 original mapped image containing 9 pixel locations, each pixel location can be used to set a pixel. Then, any two different pixel locations among these 9 locations can form a pixel pair. p i , p j The pixel pair includes p i and p j Two pixels, the coordinates of the two pixels are p i = ( x i , y i )and p = (x j , y j ).

[0070] After obtaining the pixel pair set, the pixel distance of the pixel pair can be calculated according to the pixel coordinates. Then, according to the pixel positions in the original mapping image, the pixel distance ranking matrix is generated according to the pixel distance.

[0071] For example, for the pixel coordinates of any two pixel points in the original mapping image p i = ( x i , y i ) and p = ( x j , y j ), the spatial distance between the pixels can be calculated to obtain the pixel distance. The pixel distance can also be obtained by calculating the Euclidean distance between the two pixel positions, that is:

[0072] ;

[0073] In the formula, d ( p i , p j ) represents the pixel distance of the pixel points p i and p j ; ( x i , y i ) is the pixel coordinate of the pixel point p i ; ( x j , y j ) is the pixel coordinate of the pixel point p j . Then, according to the pixel distance value obtained by calculation, the pixel distance ranking matrix Q can be obtained. That is, the pixel distance ranking matrix Q can be expressed as:

[0074] ;

[0075] According to the feature distance ranking matrix R obtained by calculating the converter production parameter information and the pixel distance ranking matrix QThe distance ranking matrix includes data for characterizing distance relationships, such as feature distance and pixel distance, which can be used to perform feature-to-pixel mapping.

[0076] In S103, feature-to-pixel mapping is performed based on the distance ranking matrix to obtain converter image data.

[0077] In S103, feature-to-pixel mapping is performed based on the distance ranking matrix to obtain converter image data. R and the pixel distance ranking matrix Q After that, feature-to-pixel mapping can be performed based on the feature distance ranking matrix R and the pixel distance ranking matrix Q to obtain converter image data. The converter image data is a set of pixel points obtained by mapping the converter features in the converter production parameter information to pixels according to the optimal mapping relationship.

[0078] The optimal mapping relationship is the feature-to-pixel mapping relationship obtained by performing feature exchange on the converter features in the distance ranking matrix and continuously iterating according to the error reduction rate after feature exchange. The electronic device can use a Greedy Feature Iterative Transformation (GFIT) algorithm to optimize the feature-to-pixel mapping.

[0079] Therefore, in some embodiments, when the electronic device performs feature-to-pixel mapping based on the distance ranking matrix to obtain converter image data, it can first initialize algorithm variables. The algorithm variables include iteration index, initial error, feature exchange record vector, and feature sorting vector; the initial value of the iteration index is 0; the initial error is calculated according to the distance ranking matrix; the initial value of the feature exchange record vector is negative infinity; and the vector length of the feature exchange record vector and the feature sorting vector is equal to the number of converter features.

[0080] For example, the electronic device can initialize the algorithm variables to set the iteration index w = 0, the initial error e 0= err ( R , Q ), where error represents the degree of deviation of the feature distance ranking matrix R and the pixel distance ranking matrix Q , i.e., the error can be calculated as follows:

[0081] ;

[0082] where err ( R , Q ) represents the feature distance ranking matrixR And pixel distance ranking matrix Q The error; N Indicates the total number of features; R i,j Represents the feature distance ranking matrix R The first in i line, number j Column matrix elements; Q i,j Represents the pixel distance ranking matrix Q The first in i line, number j Columns of matrix elements.

[0083] During the initialization of algorithm variables, feature exchange record vectors and feature sorting vectors can also be defined. Among them, the feature exchange record vector... x A feature sorting vector initialized to negative infinity for a length of 9. t The sequence is [1, 2, 3, ..., 9].

[0084] After initializing the algorithm variables, the electronic device can continuously iterate based on the algorithm variables. That is, it continuously updates the iteration index as the iteration process progresses, and performs feature exchange on the converter features based on the greedy algorithm and the feature distance ranking matrix, and calculates the error reduction rate during the feature exchange process.

[0085] like Figure 3 As shown, in some embodiments, in order to perform continuous iteration, when the electronic device performs feature exchange on converter features based on a greedy algorithm and a feature distance ranking matrix, it can select a first converter feature from the converter features according to the feature exchange record vector, and then determine a second converter feature using a greedy algorithm. The first converter feature is the converter feature that has not been exchanged for the longest time recorded in the feature exchange record vector. The second converter feature is the converter feature whose iteration error reaches its maximum value after being exchanged with the first converter feature.

[0086] For example, by updating the iterative index w = w +1, Feature exchange record vector x Select the converter feature that has not been exchanged the longest from the converter feature set. f This refers to the first converter characteristic. A greedy algorithm is used to calculate the characteristics related to this converter feature feature one by one. f The feature that the new error can reach its maximum value after the exchange is completed. n This refers to the characteristics of the second converter. The characteristics of the second converter can be calculated using the following formula:

[0087] ;

[0088] In the formula,f n denotes the first n converter feature; f denotes the first converter feature; argmax denotes the second converter feature; e n denotes the input value that takes the maximum value; e n denotes the first n converter feature; e n denotes the error after the err R n Q th converter feature exchange; R n denotes the feature distance ranking matrix after the n th converter feature exchange.

[0089] After the first converter feature and the second converter feature are determined, the feature exchange can be performed on the first converter feature and the second converter feature. As shown in FIG. 6, in some embodiments, in order to perform the feature exchange, the electronic device can calculate the iteration error according to the pixel distance ranking matrix and the feature distance ranking matrix after the feature exchange, and calculate the error reduction rate according to the iteration error and the initial error. Then, by comparing the error reduction rate with the preset minimum error reduction rate, it is determined whether to exchange between the converter features. 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, the feature sorting vector and the feature distance ranking matrix are updated. Figure 4

[0090] For example, the error reduction rate after the converter feature exchange can be calculated according to the preset minimum error reduction rate e , and the error reduction rate can be calculated according to the following formula:

[0091] ;

[0092] wherein, denotes the error reduction rate; err R , Q denotes the initial error e 0, err R n , Q denotes the iteration error e n , i.e., the error after the n th converter feature exchange. ​​​​​

[0093] If the error reduction rate is greater than the minimum error reduction rate, the converter features are exchanged between each other. The feature exchange record vector is updated x , the feature ranking vector is updated t , and the feature distance ranking matrix is updated R . Otherwise, only the feature exchange record vector is updated x , and the other parameters are returned to their original values.

[0094] After calculating the error reduction rate, the electronic device can also perform continuous iteration based on the error reduction rate until the error reduction rate converges to a preset error reduction rate threshold to obtain the best mapping relationship, and then map the converter features to the pixels according to the best mapping relationship to obtain the converter image data.

[0095] For example, when using the GFIT algorithm to optimize the mapping of features to pixels, the iteration index can be continuously updated, the converter feature that has not been exchanged for the longest time in the selection vector is selected, and the feature whose new error can reach the maximum value after the exchange of the feature is calculated feature by feature through the greedy algorithm, and the error reduction rate after the exchange of the converter features is calculated, and the parameters are updated by comparing the error reduction rate with the set minimum error reduction rate until the algorithm error reduction rate converges to the preset error reduction rate threshold t , to obtain a best feature to pixel mapping (F2P MAPPING). Then, all features of each sample on the converter are mapped to each pixel point in the image through the F2P MAPPING, and all samples are converted into an image.

[0096] S104, input the converter image data into the prediction model to obtain the lime consumption amount prediction information output by the prediction model.

[0097] After obtaining the converter image data through the mapping of features to pixels, the electronic device can input the converter image data into the prediction model to obtain the lime consumption amount 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.

[0098] For example, the neural grid structure in the convolutional neural network of the prediction model is two layers of convolution, and the convolution kernel size is 2x2. 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.

[0099] For example, the neural grid structure in the convolutional neural network of the prediction model is two layers of convolution, and the convolution kernel size is 2x2. 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. Figure 5As shown, in some embodiments, the electronic device can obtain the prediction model in advance through model training, that is, the electronic device can obtain the 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. Then input the converter sample data into the trained model to obtain the lime consumption prediction value output by the trained model. Then calculate the training loss according to the lime consumption prediction value and the lime consumption label.

[0100] By comparing the training loss with the preset loss threshold, it is determined whether the prediction model converges or not, 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 the trained model based on the adjusted model parameters is executed for iterative training. After multiple iterations of training, when the corresponding training loss output by the trained model is iterated to be less than or equal to the preset loss threshold, it is determined that the prediction model has converged, as shown in Figure 6 Therefore, the model parameters of the trained model can be output to obtain the prediction model.

[0101] Based on the prediction model obtained through pre-training, the electronic device can predict the lime consumption, and for this purpose, the electronic device can input the converter image data into the prediction model. The prediction model can perform feature analysis on the input converter image data, thereby predicting the lime consumption.

[0102] Therefore, by applying the technical solutions of the above embodiments, the converter lime consumption prediction method based on the 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 form of table data of converter production parameter information into an image, so that the convolutional neural network (CNN) can learn using the spatial structure of the image, effectively improve the prediction efficiency and accuracy of the converter lime consumption, improve the quality of the molten steel, reduce the production cost, and thus obtain higher economic benefits.

[0103] In some embodiments, as a refinement and extension of the specific implementation of the above embodiments, in order to completely describe the specific implementation process of the present embodiment, the present application further provides a converter lime consumption prediction method 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 optimized model to predict the converter lime consumption. As shown in Figure 7 The method comprises:

[0104] S201, divide the converter sample data into a training set and a test set;

[0105] S202, performing model training on the plurality of optimizers using the training set to obtain a plurality of candidate models based on different optimizers;

[0106] S203, inputting the test set into the candidate models to obtain test output information;

[0107] S204, calculating a prediction index according to the test output information;

[0108] S205, selecting a prediction model from the plurality of candidate models based on the prediction index.

[0109] In order 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 processed in layers using shuffle, wherein 80% is used as the training set and 20% is used as the test set.

[0110] The training set and the test set can be divided. The training set can be used to perform model training on a plurality of optimizers to obtain a plurality of candidate models based on different optimizers. The test set can be input into the candidate models to obtain test output information. Then, a prediction index can be calculated according to the test output information. The prediction index includes mean squared error and hit rate.

[0111] For example, the convolutional neural network selects mean squared error (MSE) and hit rate (HR) as the prediction index. The mean squared error is calculated using the following formula:

[0112] ;

[0113] In the formula, MSE is the mean squared error; y i is the label vector, is the predicted value; N is the total number of features.

[0114] ;

[0115] Correspondingly, the hit rate can be calculated using the following formula:

[0116] ;

[0117] In the formula, error is the error between the actual input value and the predicted value, n y is the number of data satisfying the threshold value, N is the total number of features, i.e., the total number of samples.

[0118] After the prediction index is calculated, the electronic device selects a prediction model from the plurality of candidate models based on the prediction index. The prediction model includes an optimal optimizer that causes a hit rate of a mean square error of the candidate model predicting the converter lime consumption amount in a preset quality interval to be greater than or equal to a hit rate threshold.

[0119] For example, the plurality of optimizers used in the model training process are Adagrad, Nadam, RMsprop, SGD, and Adam, respectively. The optimal optimizer causes the hit rate of the error of the convolutional neural network predicting the converter lime consumption amount to be greater than 90% in ±20 kg.

[0120] By applying the technical solutions of the above embodiments, the converter lime consumption prediction method based on the convolutional neural network provided in the above embodiments can select suitable convolutional neural networks and optimizers for different furnace conditions to improve the accuracy of converter lime consumption prediction. The optimizer with a hit rate greater than or equal to the hit rate threshold in the preset quality interval is determined as the optimal optimizer, thereby improving the stability of smelting production, ensuring the quality of molten steel, reducing production costs, and improving economic benefits. Figure 8 As shown in the converter lime consumption prediction model prediction value statistical comparison chart, the prediction performance of the convolutional neural network is better than that of the BP model.

[0121] In some embodiments, as a specific implementation of the converter lime consumption prediction method based on the convolutional neural network described in the above embodiments, some embodiments of the present application also provide a converter lime consumption prediction system based on a convolutional neural network, as shown in Figure 9 The system includes:

[0122] An information acquisition module is configured to acquire converter production parameter information, the converter production parameter information including at least hot metal temperature, hot metal quantity, scrap steel quantity, oxygen consumption, magnesium ball consumption, solid waste ball consumption, return ore consumption, lump ore consumption, and tapping temperature;

[0123] A matrix calculation module is configured to calculate a distance ranking matrix according to the converter production parameter information, the distance ranking matrix including a feature distance ranking matrix and a pixel distance ranking matrix;

[0124] An image mapping module is configured to perform feature-to-pixel mapping based on the distance ranking matrix to obtain converter image data; the converter image data is a pixel point set 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 according to an error reduction rate after the feature exchange;

[0125] The prediction module is configured 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 trained by using converter sample data and a plurality of optimizers.

[0126] By applying the technical solutions of the above embodiments, the converter lime consumption prediction system based on the convolutional neural network provided in the above embodiments can obtain converter production parameter information through the information acquisition module, and calculate a feature distance ranking matrix and a pixel distance ranking matrix according to the converter production parameter information through the matrix calculation module. The image mapping module performs feature-to-pixel mapping to obtain converter image data, and then the prediction module inputs the converter image data into a 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, so that the convolutional neural network can use the spatial structure of the image to perform multi-task learning, and the prediction efficiency and accuracy of the converter lime consumption can be improved. The prediction information can be used for a steel smelting process to improve the quality of molten steel and reduce production costs, thereby obtaining higher economic benefits.

[0127] It should be noted that other corresponding descriptions of the various functional units involved in the converter lime consumption prediction system based on the convolutional neural network provided in the embodiments of the present application can be referred to the corresponding descriptions in the converter lime consumption prediction method based on the convolutional neural network provided in the above embodiments, which will not be described here.

[0128] The embodiments of the present application also provide a computer device, which can be a personal computer, a server, a network device, etc. The computer device includes a bus, a processor, a memory, and a communication interface, and can also include an input / output interface and a display device. The processor of the computer device is configured 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 operating system and the computer program in the non-volatile storage medium to run. The database of the computer device is configured to store location information. The network interface of the computer device is configured to communicate with an external terminal through a network connection. The computer program is executed by the processor to implement the steps in the method embodiments.

[0129] Those skilled in the art can understand that the structure of the computer device described above is only part of the structure related to the scheme of the present application, and does not constitute a limitation on the computer device to which the scheme of the present application is applied. A specific computer device can include more or fewer components, or combine certain components, or have a different arrangement of components.

[0130] In an embodiment, a computer readable storage medium is also provided, which can be non-volatile or volatile, and has stored thereon a computer program which, when executed by a processor, implements the steps of any of the above method embodiments.

[0131] In an embodiment, a computer program product is also provided, comprising a computer program which, when executed by a processor, implements the steps of any of the above method embodiments.

[0132] It should be noted that the user information (including but not limited to user equipment information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in the present application are all information and data authorized by the user or authorized by all parties.

[0133] It can be understood by those skilled in the art that all or part of the processes in the above embodiments can be completed by a computer program instructing related hardware, and 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 above embodiments.

[0134] Any reference to memory, database or other medium used in the embodiments provided by the present application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical storage, high-density embedded non-volatile memory, resistive memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric memory (FRAM), phase change memory (PCM), graphene memory, etc.

[0135] Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration but not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc.

[0136] The database involved in each of the embodiments provided in the present application can include at least one of a relational database and a non-relational database. The non-relational database can include a distributed database based on a blockchain, and the like, without being limited thereto. The processor involved in each of the embodiments provided in the present application can be 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, without being limited thereto.

[0137] Each of the technical features of the above embodiments can be combined arbitrarily. In order to make the description simple, each of the technical features in the above embodiments is not described in all possible combinations, however, as long as the combination of the technical features does not exist contradictory, it should be considered as the scope of the present disclosure.

[0138] The above-described embodiments only express several implementation manners of the present application, and the description is relatively specific and detailed, but it should not be understood as a limitation on the scope of the patent of the present application. It should be pointed out that, for ordinary skilled persons in the art, several modifications and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the appended claims.

Claims

1. A method for predicting converter lime consumption based on convolutional neural networks, characterized in that, The method includes: Obtain converter production parameter information, which includes at least the following: molten iron temperature, molten iron quantity, scrap steel quantity, 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, the distance ranking matrix including a feature distance ranking matrix and a pixel distance ranking matrix; calculating the distance ranking matrix based on 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 based on the converter features, the original mapping image including multiple pixel positions; the number of pixel positions being greater than or equal to the number of converter features in the feature set; calculating the feature distance between any two converter features in the feature set; arranging the feature distances in ascending order to obtain the feature distance ranking matrix; traversing the pixel positions in the original mapping image to obtain a set of pixel pairs, the set of pixel pairs including multiple pixel pairs, the pixel pairs including the pixel coordinates of any two pixels in the original mapping image; calculating the pixel distance of the pixel pairs based on the pixel coordinates; and generating the pixel distance ranking matrix based on the pixel positions in the original mapping image and the pixel distances. Based on the distance ranking matrix, feature-to-pixel mapping is performed to obtain converter image data. The converter image data is a set of pixels obtained by mapping converter features in the converter production parameter information to pixels according to the optimal mapping relationship. The optimal mapping relationship is obtained by performing feature exchange on the converter features in the distance ranking matrix and continuously iterating according to the error reduction rate after feature exchange. Performing feature-to-pixel mapping based on the distance ranking matrix to obtain converter image data includes: initializing algorithm variables, including an iteration index, initial error, feature exchange record vector, and 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 length of the feature exchange record vector and the feature sorting vector is 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. Performing feature exchange on the converter features based on the greedy algorithm and the feature distance ranking matrix includes: according to the feature exchange record vector, from the converter features... The process involves selecting a first converter feature, which is the converter feature that has not been exchanged for the longest time recorded in the feature exchange record vector; determining a second converter feature using a greedy algorithm, which is the converter feature whose iteration error reaches its maximum value after being exchanged with the first converter feature; performing feature exchange on the first and second converter features; and performing feature exchange on the first and second converter features, including: calculating the iteration error based on the pixel distance ranking matrix and the feature distance ranking matrix after the feature exchange; calculating the error reduction rate based on the iteration error and the initial error; updating the feature exchange record vector if the error reduction rate is less than or equal to a preset minimum error reduction rate; updating the feature exchange record vector, the feature ranking vector, and the feature distance ranking matrix if the error reduction rate is greater than the preset minimum error reduction rate; calculating the error reduction rate during the feature exchange process; continuously iterating 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 mapping the converter feature to pixels according to the optimal mapping relationship to obtain the converter image data. The converter image data is input 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.

2. The method according to claim 1, characterized in that, Obtain converter production parameter information, including: Obtain raw production parameters; Based on the normal value range, effective parameters are extracted from the original production parameters. The normal value range is a screening range determined by using a box plot method according to the interquartile range of a preset multiple. The correlation coefficient is calculated based on the effective parameters, and 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. Based on the correlation coefficient, extract the converter production parameters that are related to the converter lime consumption from the effective parameters; The converter production parameters are standardized to obtain converter production parameter information.

3. The method according to claim 1, characterized in that, The method further includes: Acquire converter sample data, wherein the converter sample image includes a set of sample production parameters and a lime consumption label set for the set of sample production parameters; The converter sample data is input into the trained model to obtain the lime consumption prediction value output by the trained model. The training loss is calculated based on the predicted lime consumption value and the lime consumption label; If the training loss is greater than a preset loss threshold, the model parameters of the trained model are adjusted according to the training loss, and iterative training is performed 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.

4. The method according to claim 1, characterized in that, The method further includes: The converter sample data is divided into a training set and a test set; The training set is used to train models on multiple optimizers to obtain multiple candidate models based on different optimizers; The test set is input into the candidate model to obtain test output information; The prediction metrics are calculated based on the test output information, and the prediction metrics include mean squared error and hit rate. Based on the prediction index, the prediction model is selected from multiple candidate models. The prediction model includes an optimal optimizer, which makes the mean square error of the candidate model in predicting converter lime consumption within a preset quality range greater than or equal to a hit rate threshold.

5. A converter lime consumption prediction system based on convolutional neural networks, characterized in that, The system includes: The information acquisition module is used to acquire converter production parameter information, which includes at least the following: molten iron temperature, molten iron quantity, scrap steel quantity, 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 the converter production parameter information. The distance ranking matrix includes a feature distance ranking matrix and a pixel distance ranking matrix. Calculating the distance ranking matrix based on 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 based on the converter features, the original mapping image including multiple pixel positions; the number of pixel positions being greater than or equal to the number of converter features in the feature set; calculating the feature distance between any two converter features in the feature set; arranging the feature distances in ascending order to obtain the feature distance ranking matrix; traversing the pixel positions in the original mapping image to obtain a set of pixel pairs, the set of pixel pairs including multiple pixel pairs, each pixel pair including the pixel coordinates of any two pixels in the original mapping image; calculating the pixel distance between the pixel pairs based on the pixel coordinates; and generating the pixel distance ranking matrix based on the pixel positions in the original mapping image and the pixel distances. An image mapping module is used to perform feature-to-pixel mapping based on the distance ranking matrix to obtain converter image data. The converter image data is a set of pixels obtained by mapping converter features in the converter production parameter information to pixels according to the optimal mapping relationship. The optimal mapping relationship is the feature-to-pixel mapping relationship obtained by performing feature exchange on the converter features in the distance ranking matrix and continuously iterating according to the error reduction rate after feature exchange. Performing feature-to-pixel mapping based on the distance ranking matrix to obtain converter image data includes: initializing algorithm variables, including an iteration index, initial error, feature exchange record vector, and 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 length of the feature exchange record vector and the feature sorting vector is 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. Performing feature exchange on the converter features based on the greedy algorithm and the feature distance ranking matrix includes: according to the feature exchange record vector, from the... The process involves selecting a first converter feature from the converter features, where the first converter feature is the converter feature that has not been exchanged for the longest time recorded in the feature exchange record vector; determining a second converter feature using a greedy algorithm, where the second converter feature is the converter feature whose iteration error reaches its maximum value after being exchanged with the first converter feature; performing feature exchange on the first and second converter features; and performing feature exchange on the first and second converter features, which includes: calculating the iteration error based on the pixel distance ranking matrix and the feature distance ranking matrix after the feature exchange; calculating the error reduction rate based on the iteration error and the initial error; updating the feature exchange record vector if the error reduction rate is less than or equal to a preset minimum error reduction rate; updating the feature exchange record vector, the feature ranking vector, and the feature distance ranking matrix if the error reduction rate is greater than the preset minimum error reduction rate; calculating the error reduction rate during the feature exchange process; continuously iterating 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 mapping the converter features to pixels according to the optimal mapping relationship to obtain the converter image data. 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.

Images