Blast furnace molten iron silicon content prediction method and system, computer equipment and medium

Abstract

The invention provides a blast furnace molten iron silicon content prediction method and system, computer equipment and a medium, and the method comprises the following steps: obtaining total data in all time periods, carrying out outlier processing, and carrying out missing value interpolation based on an outlier processing result; performing correlation calculation on the molten iron silicon content based on a missing value interpolation result, sorting correlation calculation results, and performing data screening according to a sorting result to obtain selected feature total data; inputting the selected feature total data into the improved time sequence convolutional network structure for model training, and generating a prediction model; and inputting recent data of the selected characteristics into the prediction model, and predicting the silicon content of the molten iron of the blast furnace at the future moment. Through time sequence data abnormal value processing, missing value intelligent interpolation, correlation automatic analysis and molten iron silicon content on-line prediction on full-amount data, the blast furnace molten iron silicon content can be predicted in real time, and the problems existing when the blast furnace molten iron silicon content is predicted at present are solved.

Description

Technical Field

[0001] The present invention relates to the technical field of blast furnaces, and in particular to a method and system for predicting silicon content in molten iron of a blast furnace, computer equipment and a medium. Background Art

[0002] As the largest single equipment in the steel production process, the safe and long-life operation of the blast furnace is crucial to the development of steel enterprises. The temperature of the furnace hearth area is mainly expressed by the physical heat (temperature) and chemical heat (silicon content) of the molten iron. Their range and frequency of change can reflect the stability of the production process, the efficiency of energy utilization and the quality of pig iron. However, because the blast furnace is a closed high-temperature and high-pressure container, its internal operating conditions cannot be directly detected, which makes controlling the smooth operation of the blast furnace an industrial problem.

[0003] At present, the operation status of blast furnace can only be judged by multiple sensor data and molten iron inspection data results. Among them, the data that can reflect the temperature of blast furnace hearth, such as tuyere temperature, iron mouth temperature, molten iron temperature, and molten iron silicon content, are important references for blast furnace operation. Too high or too low furnace temperature will not only reduce the quality of molten iron, but also cause furnace conditions to fluctuate or even abnormal. As a characterizing variable of blast furnace chemical heat, the silicon content of molten iron plays an important role in blast furnace operation. At present, the acquisition of molten iron silicon content data is to take samples from the molten iron tank car after iron is tapped, and then send it to the laboratory for analysis. Although the laboratory analysis results are very accurate, due to objective reasons such as transportation time, there is a time lag in the molten iron inspection data, which makes the molten iron silicon content test results unable to timely reflect the objective situation of the temperature condition in the furnace. Therefore, it is necessary to propose an accurate method for predicting the silicon content of molten iron to provide operators with real-time blast furnace chemical heat reference data. Summary of the invention

[0004] In view of the above-mentioned shortcomings of the prior art, the object of the present invention is to provide a method and system for predicting the silicon content in molten iron of a blast furnace, a computer device and a medium, so as to solve the technical problems existing in the prior art.

[0005] To achieve the above object and other related objects, the present invention provides a method for predicting silicon content in molten iron of a blast furnace, comprising the following steps:

[0006] Obtaining full data in all time periods; wherein the full data includes blast furnace sensor data, blast furnace raw material data, and blast furnace inspection and testing data;

[0007] Performing outlier processing on the full amount of data, and interpolating missing values ​​based on the outlier processing results; and,

[0008] Based on the missing value interpolation results, the correlation of the silicon content in molten iron is calculated, the correlation calculation results are sorted, and the data is screened according to the sorting results to obtain the full amount of data of the selected features;

[0009] The full amount of data of the selected features is input into the improved temporal convolutional network structure for model training to generate a prediction model; and the recent data of the selected features is input into the prediction model to predict the silicon content of molten iron in the blast furnace at a future moment; wherein the recent data of the selected features includes the full amount of data within a preset time period, and the preset time period takes the current moment as the final moment and the time length is less than that of all the time periods.

[0010] In one embodiment of the present invention, the process of performing outlier processing on the full amount of data includes:

[0011] The full amount of data is processed offline and Kalman smoothing is performed to calculate the confidence interval, and the following is obtained:

[0012]

[0013] In the formula, x i is the true value of the full data, is the smoothed value of the full data after Kalman smoothing, δ i is the variance of the Kalman smoother, α is the significance level, Z α2 is the quantile position, n is the length of the full data;

[0014] Data points outside the confidence interval are removed as outliers to obtain an outlier processing result.

[0015] In one embodiment of the present invention, the process of interpolating missing values ​​based on outlier processing results includes:

[0016] The outlier processing result is concatenated with the missing value matrix, mask matrix, and mask missing matrix introduced in advance or in real time; and the concatenated result is input into a neural network interpolation model, and vectorized and positionally encoded using at least one layer of linear neural network; wherein the neural network interpolation model is generated based on a self-attention mechanism structure;

[0017] The total amount of data in all time periods is added to the position encoding results and then masked, and the masked results are input into the multi-layer self-attention mechanism structure;

[0018] At least one layer of linear neural network is used to map the output data of the self-attention mechanism structure to obtain the missing value interpolation result; wherein the output data after dimension mapping has the same data dimension as the full amount of data in all time periods.

[0019] In one embodiment of the present invention, a missing value matrix M is introduced in advance or in real time, and has:

[0020]

[0021] In the formula, represents the value of the dth row and tth column in the missing value matrix, Represents the full data value corresponding to the dth row and tth column in the missing value matrix, the missing value matrix M∈R T×D , D = max(d), T = max(t);

[0022] The mask matrix I is introduced in advance or in real time, and there is:

[0023]

[0024] In the formula, represents the mask data value corresponding to the dth row and tth column in the mask matrix, Represents the value of the dth row and tth column in the mask matrix, mask matrix I∈R T×D ;

[0025] Introduce mask missing matrix in advance or in real time have:

[0026]

[0027] In the formula, Represents the mask data value corresponding to the dth row and tth column in the mask missing matrix. The mask missing matrix

[0028] In one embodiment of the present invention, after obtaining the missing value interpolation result, the method further includes:

[0029] The loss function value is calculated based on the missing value matrix, the mask matrix and the mask missing matrix, and the loss function value is:

[0030]

[0031] In the formula, L represents the loss function value, represents the missing value interpolation result,

[0032] The calculated loss function value is used for back propagation to update the neural network interpolation model.

[0033] In one embodiment of the present invention, the full amount of data of the selected features is input into the improved temporal convolutional network structure for model training, and the process of generating a prediction model includes:

[0034] Based on the fixed hyperparameters debugged during the offline data processing, outlier processing is performed on the full amount of data of the selected features through a Kalman smoother;

[0035] Based on the fixed hyperparameters debugged during the offline data processing, the missing values ​​of the full amount of data of the selected features are interpolated online through the neural network interpolation model;

[0036] Perform correlation calculation on the missing value online interpolation results, sort the correlation calculation results, and screen the data according to the sorting results to obtain the selected variable data;

[0037] The selected variable data are input into the improved time series convolutional network structure for model training, and the model parameters with the best prediction hit rate of molten iron silicon content obtained through the test set are combined with the improved time series convolutional network structure as a prediction model for blast furnace molten iron silicon content.

[0038] In one embodiment of the present invention, the selected characteristic recent data is input into the prediction model, and the process of predicting the silicon content of molten iron in the blast furnace at a future time includes:

[0039] Taking each item of the selected characteristic recent data as an independent variable and segmenting it according to a fixed step length to obtain a plurality of one-dimensional data blocks;

[0040] Use multiple one-dimensional convolution kernels to perform convolution calculations on each data block to construct and generate multiple feature vectors;

[0041] The long sides of the multiple generated feature vectors are spliced ​​to form a two-dimensional matrix, and the two-dimensional matrix formed by splicing the feature vectors of each data block is overlapped and spliced ​​to form a three-dimensional matrix;

[0042] A one-dimensional deep convolutional network is used to mine the temporal dependency of the feature vector concatenation results, and a one-layer one-dimensional point-by-point convolutional neural network is used to mine the interaction relationship between single-variable features, and a one-layer one-dimensional point-by-point convolutional neural network is used to mine the interaction relationship between cross-variable features, and residual connections are performed;

[0043] The residual connection results are mapped to the length of the prediction time through a layer of fully connected neural network to obtain the prediction result of the silicon content in the blast furnace molten iron at the future moment.

[0044] The present invention also provides a blast furnace molten iron silicon content prediction system, the system comprising:

[0045] A data acquisition module is used to obtain the full amount of data in all time periods; wherein the full amount of data includes blast furnace sensor data, blast furnace raw material data, and blast furnace inspection and testing data;

[0046] An outlier processing module, used for performing outlier processing on the full amount of data;

[0047] Missing value interpolation module, used to interpolate missing values ​​for the full data after outlier processing;

[0048] A correlation calculation module is used to perform correlation calculation on the silicon content of molten iron according to the missing value interpolation results, sort the correlation calculation results, and screen the data according to the sorting results to obtain the full amount of data of the selected features;

[0049] A prediction module is used to input the full amount of data of the selected features into the improved temporal convolutional network structure for model training to generate a prediction model; and input the recent data of the selected features into the prediction model to predict the silicon content of molten iron in the blast furnace at a future moment; wherein the recent data of the selected features includes the full amount of data within a preset time period, and the preset time period takes the current moment as the final moment and the time length is less than that of all the time periods.

[0050] The present invention also provides a computer device, comprising a memory, a processor and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of any one of the above-mentioned methods for predicting the silicon content in molten iron in a blast furnace.

[0051] The present invention also provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the steps of any one of the above-mentioned methods for predicting the silicon content in molten iron in a blast furnace are implemented.

[0052] As described above, the present invention provides a method and system for predicting the silicon content of molten iron in a blast furnace, a computer device and a medium, which have the following beneficial effects: by obtaining the full amount of data in all time periods, the full amount of data includes blast furnace sensor data, blast furnace raw material data, and blast furnace inspection and testing data; then outlier processing is performed on the full amount of data, and missing value interpolation is performed based on the outlier processing result; and, the correlation calculation of the silicon content in the molten iron is performed based on the missing value interpolation result, and the correlation calculation results are sorted, and data is screened according to the sorting result to obtain the selected feature full amount data; then the selected feature full amount data is input into the improved temporal convolutional network structure for model training to generate a prediction model; and the selected feature recent data is input into the prediction model to predict the silicon content of the molten iron in the future. Wherein, the selected feature recent data includes the full amount of data in a preset time period, and the preset time period takes the current moment as the final moment and the time length is less than all time periods. It can be seen from this that the present invention can predict the silicon content of molten iron in blast furnace in real time by performing time series data outlier processing, missing value intelligent interpolation, automatic correlation analysis and online prediction of molten iron silicon content on all data such as blast furnace sensor data, blast furnace raw material data, blast furnace inspection and testing data, thereby improving the time lag problem currently existing in predicting the silicon content of molten iron in blast furnace and improving the prediction effect of molten iron silicon content. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 A schematic diagram of an exemplary system architecture for applying the technical solutions in one or more embodiments of the present invention;

[0054] Figure 2 A schematic diagram of a process for predicting silicon content in molten iron in a blast furnace provided by an embodiment of the present invention;

[0055] Figure 3 A schematic flow chart of a method for predicting silicon content in molten iron in a blast furnace provided in another embodiment of the present invention;

[0056] Figure 4 A schematic diagram of the hardware structure of a blast furnace iron silicon content prediction system provided by an embodiment of the present invention;

[0057] Figure 5 The figure is a schematic diagram of the hardware structure of a computer device suitable for implementing one or more embodiments of the present invention. DETAILED DESCRIPTION

[0058] The following is an explanation of the embodiments of the present invention by specific specific examples, and those skilled in the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification. The present invention can also be implemented or applied by other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present invention. It is understood that the following embodiments and the features in the embodiments can be combined with each other without conflict. In addition, it is understood that the illustrations provided in the following embodiments only illustrate the basic concept of the present invention in a schematic manner, and only the components related to the present invention are shown in the drawings instead of being drawn according to the number, shape and size of the components during actual implementation. The type, quantity and proportion of each component during actual implementation can be a random change, and the component layout type may also be more complicated.

[0059] Figure 1 FIG. 1 is a schematic diagram showing an exemplary system architecture to which the technical solution in one or more embodiments of the present invention can be applied. Figure 1 As shown, the system architecture 100 may include a terminal device 110, a network 120, and a server 130. The terminal device 110 may include various electronic devices such as a smart phone, a tablet computer, a laptop computer, and a desktop computer. The server 130 may be an independent physical server, or a server cluster or a distributed system composed of multiple physical servers, or a cloud server providing cloud computing services. The network 120 may be a communication medium of various connection types that can provide a communication link between the terminal device 110 and the server 130, such as a wired communication link or a wireless communication link.

[0060] According to the implementation requirements, the system architecture in the embodiment of the present invention may have any number of terminal devices, networks and servers. For example, the server 130 may be a server group composed of multiple server devices. In addition, the technical solution provided in the embodiment of the present invention may be applied to the terminal device 110, or to the server 130, or may be implemented by the terminal device 110 and the server 130 together, and the present invention does not make any special limitation on this.

[0061] In one embodiment of the present invention, the terminal device 110 or the server 130 of the present invention can obtain the full amount of data in all time periods, and the full amount of data includes blast furnace sensor data, blast furnace raw material data, and blast furnace inspection and testing data; then outlier processing is performed on the full amount of data, and missing value interpolation is performed based on the outlier processing result; and, the correlation calculation of the silicon content of molten iron is performed based on the missing value interpolation result, and the correlation calculation results are sorted, and data is screened according to the sorting result to obtain the selected feature full amount data; then the selected feature full amount data is input into the improved temporal convolutional network structure for model training to generate a prediction model; and the selected feature recent data is input into the prediction model to predict the silicon content of blast furnace molten iron at a future moment. Among them, the selected feature recent data includes the full amount of data within a preset time period, and the preset time period takes the current moment as the final moment and the time length is less than all time periods. By using the terminal device 110 or the server 130 to execute the method for predicting the silicon content in molten iron of a blast furnace, the silicon content in molten iron of the blast furnace can be predicted in real time by performing time series data outlier processing, intelligent interpolation of missing values, automatic correlation analysis and online prediction of the silicon content in molten iron on all data such as blast furnace sensor data, blast furnace raw material data, and blast furnace inspection and testing data, thereby improving the time lag problem currently existing in predicting the silicon content in molten iron of the blast furnace and improving the prediction effect of the silicon content in molten iron.

[0062] The above section introduces the contents of an exemplary system architecture applying the technical solution of the present invention. Next, the method for predicting silicon content in molten iron of a blast furnace of the present invention will be introduced.

[0063] Figure 2 A schematic flow chart of a method for predicting silicon content in molten iron of a blast furnace is shown. Specifically, in an exemplary embodiment, Figure 2 As shown, this embodiment provides a method for predicting silicon content in molten iron of a blast furnace, comprising the following steps:

[0064] S210, acquiring full data in all time periods; wherein the full data includes blast furnace sensor data, blast furnace raw material data, and blast furnace inspection and testing data;

[0065] S220: Perform outlier processing on the entire data, and interpolate missing values ​​based on the outlier processing results.

[0066] S230, performing correlation calculation on the silicon content of the molten iron based on the missing value interpolation result, sorting the correlation calculation results, and performing data screening according to the sorting results to obtain the full amount of data of the selected features. As an example, in this embodiment or other embodiments, when performing correlation calculation, linear and nonlinear correlations can be mined through Pearson (Pearson correlation coefficient), Spearman (Spearman correlation coefficient), MIC (maximum mutual information coefficient), Random Forest (random forest), F_regression (univariate linear regression test), MI (mutual information), Xgboost (distributed gradient enhancement) correlation analysis methods, wherein Pearson and F_regression calculate linear correlations, Spearman calculates monotonic correlations, and MI, MIC, Random Forest, and Xgboost consider linear and nonlinear correlations at the same time. At the same time, the calculation results of Pearson, Spearman, MIC, F_regression, and MI can be normalized, the calculation results of Random Forest and Xgboost can be normalized after taking the logarithm, all the results can be weighted summed and sorted, and finally, the characteristic variables for predicting the silicon content in molten iron can be obtained by combining expert experience.

[0067] S240, input the selected feature full data into the improved time series convolutional network structure for model training to generate a prediction model; and input the selected feature recent data into the prediction model to predict the silicon content of molten iron in the future. The selected feature recent data includes the full data within a preset time period, the preset time period takes the current time as the final time and the time length is less than all time periods.

[0068] According to the above records, in an exemplary embodiment, the process of performing outlier processing on the full amount of data includes: taking the full amount of data offline, performing Kalman smoothing, calculating the confidence interval, and removing the data points outside the confidence interval as outliers to obtain the outlier processing result. The formula for calculating the confidence interval includes: In the formula, x i is the true value of the full data, is the smoothed value of the full data after Kalman smoothing, δ i is the variance of the Kalman smoother, α is the significance level, Z α2is the quantile position, and n is the length of the full data. It can be seen that the sliding window Kalman smoothing process for outliers of different variable data in this embodiment includes: for different variable data, using targeted sliding window size, step size, and Smoother (smoother) hyperparameters, if necessary, using differential transformation, power transformation and Box-Cox transformation (generalized impermanence transformation) and other methods, and using a combined filtering strategy. The confidence interval of the data in the window is calculated after Kalman smoothing, and it is necessary to select hyperparameters in a targeted manner, and finally retain the data in the interval as the data after outlier processing.

[0069] According to the above records, in an exemplary embodiment, the process of interpolating missing values ​​based on the outlier processing results includes: splicing the outlier processing results with the missing value matrix, mask matrix, and mask missing matrix introduced in advance or in real time; and, inputting the splicing result into the neural network interpolation model, and using at least one layer of linear neural network for vectorization processing, and performing position encoding; wherein the neural network interpolation model is generated based on the self-attention mechanism (Self-Attention) structure; adding the full amount of data in all time periods to the position encoding results and then performing mask processing, and inputting the mask processing results into the multi-layer self-attention mechanism structure; using at least one layer of linear neural network to map the output data of the self-attention mechanism structure to obtain the missing value interpolation result; wherein the output data after dimensional mapping is the same as the data dimension of the full amount of data in all time periods. In this embodiment or other embodiments, the missing value matrix M is introduced in advance or in real time, and there are: In the formula, represents the value of the dth row and tth column in the missing value matrix, Represents the full data value corresponding to the dth row and tth column in the missing value matrix, the missing value matrix M∈R T×D , D = max(d), T = max(t). Introducing the mask matrix I in advance or in real time, we have: In the formula, represents the mask data value corresponding to the dth row and tth column in the mask matrix, Represents the value of the dth row and tth column in the mask matrix, mask matrix I∈R T×D As an example, this embodiment can randomly generate a 01 matrix as a mask matrix in proportion, multiply the mask matrix by the input data point by point, and the position of 0 in the matrix is ​​the random mask position. Introduce the mask missing matrix in advance or in real time have: In the formula, Represents the mask data value corresponding to the dth row and tth column in the mask missing matrix. The mask missing matrix

[0070] According to the above records, in an exemplary embodiment, after obtaining the missing value interpolation result, the present embodiment may further include: calculating the loss function value based on the missing value matrix, the mask matrix and the mask missing matrix, performing back propagation using the calculated loss function value, and updating the neural network interpolation model. The calculation formula of the loss function value includes: In the formula, L represents the loss function value, represents the missing value interpolation result, During the operation of the blast furnace, the data distribution may change due to changes in raw materials and operators, so the model needs to be updated to adapt to the changing furnace conditions. Therefore, this embodiment uses the calculated loss function value for back propagation to update the neural network interpolation model.

[0071] According to the above records, in an exemplary embodiment, the full amount of selected feature data is input into an improved temporal convolutional network (TCN) structure for model training, and the process of generating a prediction model includes: based on the fixed hyperparameters debugged during offline data processing, outlier processing is performed on the full amount of selected feature data through a Kalman smoother; based on the fixed hyperparameters debugged during offline data processing, online interpolation of missing values ​​is performed on the full amount of selected feature data through a neural network interpolation model; correlation calculation is performed on the online interpolation results of the missing values, and the correlation calculation results are sorted, and data is screened according to the sorting results to obtain selected variable data; the selected variable data is input into an improved temporal convolutional network structure for model training, and the model parameters with the best prediction hit rate of molten iron silicon content obtained through the test set are combined with the improved temporal convolutional network structure as a prediction model for molten iron silicon content in a blast furnace. It can be seen that the prediction process of molten iron silicon content based on the improved TCN structural model in this embodiment includes: first, the data after missing value interpolation uses the data translation variable construction method to characterize the time lag phenomenon, and then the data is input into the neural network model for training, and the hit rate of the model on the verification data is calculated in real time, and finally the hyperparameters of the model are determined according to the effect of the hit rate of the verification data. It can be seen that in this embodiment, the data input length of the interpolation model and the prediction model are first kept consistent, and at any time when prediction is required, data consistent with the input length of the model data is input, and then after the data enters the model, automatic outlier processing, missing value interpolation, correlation calculation and prediction tasks will be performed. At the same time, by regularly triggering training tasks, collecting full data for interpolation model, prediction model training and correlation calculation, the training process will be executed multiple times, and finally retaining the model parameters with the highest hit rate for future prediction tasks.

[0072] According to the above records, in an exemplary embodiment, the selected feature recent data is input into the prediction model, and the process of predicting the silicon content of molten iron in the blast furnace at a future time includes: taking each data of the selected feature recent data as an independent variable, and dividing it according to a fixed length and step size to obtain multiple one-dimensional data blocks; using multiple one-dimensional convolution kernels to perform convolution calculations on each data block to construct and generate multiple feature vectors; splicing the multiple generated feature vectors on the long side to form a two-dimensional matrix, and overlapping and splicing the two-dimensional matrix formed by splicing the feature vectors of each data block to form a three-dimensional matrix; using a one-dimensional deep convolutional network to mine the time dependence of the feature vector splicing result, and using a layer of one-dimensional point-by-point convolutional neural network to mine the interaction relationship between single variable features, and using a layer of one-dimensional point-by-point convolutional neural network to mine the interaction relationship between cross-variable features, and performing residual connection; mapping the residual connection result to the length of the prediction time through a layer of fully connected neural network to obtain the prediction result of the silicon content in the blast furnace at a future time.

[0073] In another exemplary embodiment of the present invention, Figure 3 As shown, this embodiment also provides a method for predicting silicon content in molten iron of a blast furnace, comprising the following steps:

[0074] (1) Automatic data cleaning. Automatic data cleaning refers to the use of Kalman Smoother to process outliers on input data. In linear dynamic systems, the Kalman Filtering equation is forward recursive, that is, the past and current observations are used to calculate the current state; the Kalman Smoothing equation is reverse recursive, that is, the future state and current observations are used to calculate the current state. From the calculation process, it is necessary to first perform forward recursion on Kalman Filtering, and then perform backward recursion on Kalman Smoothing. In this way, the estimation of the state at a certain moment in the process refers to all data. Compared with using only Kalman Filtering, the accuracy of the result is further improved. After performing Kalman Smoothing on the variable data, the confidence interval is calculated as: In the formula, x i is the true value of the full data, is the smoothed value of the full data after Kalman smoothing, δ i is the variance of the Kalman smoother, α is the significance level, Z α2is the quantile position, and n is the length of the full data. After obtaining the confidence interval, the points outside the confidence interval are regarded as outliers and removed. In this implementation, the Kalman Smoother hyperparameter can be selected according to different variables to achieve automatic outlier processing.

[0075] (2) Missing value interpolation. Specifically, this embodiment can interpolate missing values ​​based on a neural network model with a Self-Attention structure. The specific interpolation process is as follows:

[0076] First, in order to avoid the missing values ​​in the original data affecting the calculation of the loss function, a matrix M∈R expressing the missing values ​​is introduced. T×D , In the formula, represents the value of the dth row and tth column in the missing value matrix, Represents the full data value corresponding to the dth row and tth column in the missing value matrix, the missing value matrix M∈R T×D , D = max(d), T = max(t).

[0077] In order to train the interpolation ability of the model, it is necessary to artificially mask part of the input data as an interpolation task and introduce a matrix I∈R expressing the artificial mask T×D , In the formula, represents the mask data value corresponding to the dth row and tth column in the mask matrix, Represents the value of the dth row and tth column in the mask matrix, mask matrix I∈R T×D .

[0078] In addition, to express The missing values ​​of In the formula, Represents the mask data value corresponding to the dth row and tth column in the mask missing matrix. The mask missing matrix

[0079] The input data is concatenated with the missing value label matrix and then input into the model. A layer of linear neural network is used for embedding (vectorization), and then positional encoding is performed. The result is: Among them, PE is a two-dimensional matrix, the size of which is the same as the dimension of Embedding (vectorization processing), the rows represent time points, the columns represent position vectors; pos represents the relative position of the time point of the value in the entire sequence; d modelrepresents the dimension of the position vector; i represents the position of each value in the position vector at a certain moment. Therefore, the above formula means that the even positions of the position vector at each moment are transformed by sin and the odd positions are transformed by cos, so as to fill the entire PE matrix.

[0080] The input data is then added to the Positional Encoding result and then manually masked, and then enters the multi-layer Self-Attention structure. The output data of the multi-layer Self-Attention structure passes through a layer of linear neural network, and the mapping dimension is consistent with the input data dimension as the final result. Right now The interpolation result output by the interpolation model.

[0081] Finally, the loss function is back-propagated to update the model. The formula for calculating the mask loss function is: for: In the formula, L represents the loss function value, represents the missing value interpolation result,

[0082] (3) Correlation calculation. Specifically, the correlation between the silicon content of molten iron and other variables in this embodiment is mined by Pearson (Pearson correlation coefficient), Spearman (Spearman correlation coefficient), MIC (maximum mutual information coefficient), RandomForest (random forest), F_regression (univariate linear regression test), MI (mutual information), Xgboost (distributed gradient enhancement) correlation analysis method linear and nonlinear correlation, wherein Pearson and F_regression calculate linear correlation, Spearman calculates monotonic correlation, MI, MIC, Random Forest, and Xgboost simultaneously consider linear and nonlinear correlation. As an example, for example, it takes about 7 hours from the time when the raw materials enter the furnace to the time when the iron is discharged, then all parameter data are shifted backward by 1 to 8 hours to consider the correlation of the time delay. Finally, the correlation of all parameters and the time delay parameters is calculated using the above correlation calculation method, and weighted summation is performed, the results are sorted, and the data or variables with high correlation ranking are screened in combination with expert experience.

[0083] (4) Online training prediction model. Specifically, after the improved temporal convolutional network structure is deployed, this embodiment can perform online training according to the following process:

[0084] a) The data outlier processing module automatically processes outliers on the input data according to the fixed hyperparameter KalmanSmoother (Kalman smoother) debugged during the offline data processing process;

[0085] b) After the hyperparameters of the neural network interpolation model based on the Self-Attention structure are fixed through offline interpolation experience, the missing data are interpolated online;

[0086] c) After missing values ​​are interpolated, the data enters the integrated correlation calculation module to calculate the variable correlation and obtain the variable correlation ranking. At this time, it is necessary to select relevant variables based on expert experience and return them to the model;

[0087] d) After the model receives the selected variables, it re-retrieves the data before the missing value interpolation, interpolates the missing values ​​again, and saves the trained interpolation model;

[0088] e) After re-interpolation of missing values, the selected variable data enters the time series prediction model based on the improved TCN structure for prediction training, and finally retains the optimal model parameters based on the model's prediction hit rate of molten iron silicon content on the test set.

[0089] (5) Model online training prediction strategy. Specifically, the trained prediction model will perform online prediction according to the following process: a) After inputting recent data, the model will extract corresponding data based on the variables selected in the previous step; b) The extracted data will first pass through the outlier processing module, and then use the interpolation model saved in the previous step to interpolate missing values; c) The processed data will finally enter the prediction model saved in the previous step, and finally return the predicted data of molten iron silicon content in the future to the user.

[0090] (6) Prediction of silicon content in molten iron. The specific prediction process is as follows:

[0091] a) First, each variable of the input data needs to be independent of each other, and then patching is performed, that is, the input data is split according to a fixed length and step size; next, the embedding (vectorization processing) method based on the convolution kernel is used to construct features, that is, multiple convolution kernels are used for convolution calculation for each patch to construct multiple features; finally, the embedding (vectorization processing) results are spliced ​​front and back, and the embedding (vectorization processing) results of each patch are overlapped and spliced;

[0092] b) After Patchify Embedding, the data enters a processing module composed of multiple TCN (Temporal Convolutional Network) networks: first, it enters a one-dimensional Depthwise Convolution neural network to mine the cross-temporal dependency of a single variable; then it enters a two-layer one-dimensional Pointwise Convolution neural network, the first layer mines the interaction between single variable features, and the second layer mines the interaction across variables; the data needs to be processed in series by multiple processing modules, and the calculation results of each processing module are residually connected, and the final output dimension is consistent with the input dimension;

[0093] c) The final output result is mapped to the length of the predicted time through a layer of fully connected neural network, and then compared with the actual data to calculate the MAE (Mean Absolute Error) loss for back propagation.

[0094] It can be seen that the present embodiment uses a sliding window Kalman smoothing method for outliers of different variable data: for different variable data, a targeted sliding window size, step size, and Smoother hyperparameters are used, and when necessary, differential transformation, power transformation, Box-Cox transformation (generalized transformation) and other methods are adopted, and a combined filtering strategy is used. The data in the window is Kalman smoothed to calculate the confidence interval, and it is necessary to select hyperparameters in a targeted manner, and finally retain the data in the interval as the data after outlier processing. At the same time, the present embodiment is based on the data missing value interpolation method of the Self-Attention (self-attention mechanism) structural model: a part of the variable data related to the prediction of the silicon content of molten iron after outlier processing is randomly extracted as verification data, and then the data is input into the neural network model for training, and finally the hyperparameters of the interpolation model are determined according to the effect of the verification data interpolation result. Moreover, the present embodiment is based on the calculation of variable correlation based on multi-index fusion, and multiple correlation analysis methods such as Pearson (Pearson correlation coefficient), Spearman (Spearman correlation coefficient), MIC (maximum mutual information coefficient) can be used to calculate the variable correlation, and then weighted summation is performed, and finally the variables are sorted by importance, and the parameters with high correlation ranking are screened in combination with process knowledge. In addition, the present embodiment is based on the prediction method of molten iron silicon content based on the improved TCN (Temporal Convolutional Network) structural model: first, the data after missing value interpolation is characterized by the time lag phenomenon using the data translation variable construction method, and then the data is input into the neural network model for training, and the hit rate of the model on the verification data is calculated in real time, and finally the hyperparameters of the interpolation model are determined according to the effect of the hit rate of the verification data. At the same time, the present embodiment can also use the prediction model to predict the pipeline online, that is, the data input length of the interpolation model is consistent with that of the prediction model, and at any time when prediction is required, data consistent with the model data input length is input, and then the data enters the model. Automatic outlier processing, missing value interpolation, correlation calculation and prediction tasks will be performed. In addition, the present embodiment can also update the pipeline online. During the operation of the blast furnace, the data distribution will change due to changes in raw materials and operators, so the model needs to be updated to adapt to the changing furnace conditions. The model regularly triggers training tasks, collects all data for interpolation model and prediction model training and correlation calculation. The training process will be executed multiple times, and finally retain the model parameters with the highest hit rate for future prediction tasks.

[0095] In summary, the present invention provides a method for predicting the silicon content of molten iron in a blast furnace, by obtaining the full amount of data in all time periods, the full amount of data including blast furnace sensor data, blast furnace raw material data, and blast furnace inspection and testing data; then outlier processing is performed on the full amount of data, and missing value interpolation is performed based on the outlier processing result; and, correlation calculation is performed on the silicon content of the molten iron based on the missing value interpolation result, and the correlation calculation results are sorted, and data is screened according to the sorting result to obtain the selected feature full amount data; then the selected feature full amount data is input into the temporal convolutional network structure for model training to generate a prediction model; and the selected feature recent data is input into the prediction model to predict the silicon content of the blast furnace molten iron at a future moment. Among them, the selected feature recent data includes the full amount of data within a preset time period, and the preset time period takes the current moment as the final moment and the time length is less than all time periods. It can be seen from this that this method can predict the silicon content of molten iron in real time by performing time series data outlier processing, missing value intelligent interpolation, automatic correlation analysis and online prediction of molten iron silicon content on all data such as blast furnace sensor data, blast furnace raw material data, blast furnace inspection and testing data, thereby improving the current time lag problem in predicting the silicon content of molten iron in blast furnaces and improving the prediction effect of molten iron silicon content.

[0096] In another exemplary embodiment of the present invention, Figure 4 As shown, this embodiment also provides a blast furnace molten iron silicon content prediction system, including:

[0097] The data acquisition module 410 is used to obtain the full amount of data in all time periods; wherein the full amount of data includes blast furnace sensor data, blast furnace raw material data, and blast furnace inspection and testing data;

[0098] An outlier processing module 420, configured to perform outlier processing on the full amount of data;

[0099] A missing value interpolation module 430 is used to interpolate missing values ​​of the full amount of data after outlier processing;

[0100] The correlation calculation module 440 is used to perform correlation calculation on the silicon content of molten iron according to the missing value interpolation result, sort the correlation calculation results, and screen the data according to the sorting results to obtain the full amount of data of the selected features. As an example, in this embodiment or other embodiments, when performing correlation calculation, linear and nonlinear correlations can be mined through Pearson (Pearson correlation coefficient), Spearman (Spearman correlation coefficient), MIC (maximum mutual information coefficient), Random Forest (random forest), F_regression (univariate linear regression test), MI (mutual information), Xgboost (distributed gradient enhancement) correlation analysis methods, where Pearson and F_regression calculate linear correlations, Spearman calculates monotonic correlations, and MI, MIC, Random Forest, and Xgboost consider linear and nonlinear correlations at the same time. At the same time, the calculation results of Pearson, Spearman, MIC, F_regression, and MI can be normalized, the calculation results of RandomForest and Xgboost can be normalized after taking the logarithm, all the results can be weighted summed and sorted, and finally, the characteristic variables for predicting the silicon content in molten iron can be obtained by combining expert experience.

[0101] The prediction module 450 is used to input the selected feature full data into the improved temporal convolutional network structure for model training to generate a prediction model; and input the selected feature recent data into the prediction model to predict the silicon content of molten iron in the future. The selected feature recent data includes the full data within a preset time period, and the preset time period takes the current time as the final time and the time length is less than all time periods.

[0102] According to the above records, in an exemplary embodiment, the process of performing outlier processing on the full amount of data includes: taking the full amount of data offline, performing Kalman smoothing, calculating the confidence interval, and removing the data points outside the confidence interval as outliers to obtain the outlier processing result. The formula for calculating the confidence interval includes: In the formula, x i is the true value of the full data, is the smoothed value of the full data after Kalman smoothing, δ i is the variance of the Kalman smoother, α is the significance level, Z α2is the quantile position, and n is the length of the full data. It can be seen that the sliding window Kalman smoothing process for outliers of different variable data in this embodiment includes: for different variable data, using targeted sliding window size, step size, and Smoother (smoother) hyperparameters, if necessary, using differential transformation, power transformation and Box-Cox transformation (generalized impermanence transformation) and other methods, and using a combined filtering strategy. The confidence interval of the data in the window is calculated after Kalman smoothing, and it is necessary to select hyperparameters in a targeted manner, and finally retain the data in the interval as the data after outlier processing.

[0103] According to the above records, in an exemplary embodiment, the process of interpolating missing values ​​based on the outlier processing results includes: splicing the outlier processing results with the missing value matrix, mask matrix, and mask missing matrix introduced in advance or in real time; and, inputting the splicing result into the neural network interpolation model, and using at least one layer of linear neural network for vectorization processing, and performing position encoding; wherein the neural network interpolation model is generated based on the self-attention mechanism (Self-Attention) structure; adding the full amount of data in all time periods to the position encoding results and then performing mask processing, and inputting the mask processing results into the multi-layer self-attention mechanism structure; using at least one layer of linear neural network to map the output data of the self-attention mechanism structure to obtain the missing value interpolation result; wherein the output data after dimensional mapping is the same as the data dimension of the full amount of data in all time periods. In this embodiment or other embodiments, the missing value matrix M is introduced in advance or in real time, and there are: In the formula, represents the value of the dth row and tth column in the missing value matrix, Represents the full data value corresponding to the dth row and tth column in the missing value matrix, the missing value matrix M∈R T×D , D = max(d), T = max(t). Introducing the mask matrix I in advance or in real time, we have: In the formula, represents the mask data value corresponding to the dth row and tth column in the mask matrix, Represents the value of the dth row and tth column in the mask matrix, mask matrix I∈R T×D As an example, this embodiment can randomly generate a 01 matrix as a mask matrix in proportion, multiply the mask matrix by the input data point by point, and the position of 0 in the matrix is ​​the random mask position. Introduce the mask missing matrix in advance or in real time have: In the formula, Represents the mask data value corresponding to the dth row and tth column in the mask missing matrix. The mask missing matrix

[0104] According to the above records, in an exemplary embodiment, after obtaining the missing value interpolation result, the present embodiment may further include: calculating the loss function value based on the missing value matrix, the mask matrix and the mask missing matrix, performing back propagation using the calculated loss function value, and updating the neural network interpolation model. The calculation formula of the loss function value includes: In the formula, L represents the loss function value, represents the missing value interpolation result, During the operation of the blast furnace, the data distribution may change due to changes in raw materials and operators, so the model needs to be updated to adapt to the changing furnace conditions. Therefore, this embodiment uses the calculated loss function value for back propagation to update the neural network interpolation model.

[0105] According to the above records, in an exemplary embodiment, the full amount of selected feature data is input into an improved temporal convolutional network (TCN) structure for model training, and the process of generating a prediction model includes: based on the fixed hyperparameters debugged during offline data processing, outlier processing is performed on the full amount of selected feature data through a Kalman smoother; based on the fixed hyperparameters debugged during offline data processing, online interpolation of missing values ​​is performed on the full amount of selected feature data through a neural network interpolation model; correlation calculation is performed on the online interpolation results of the missing values, and the correlation calculation results are sorted, and data is screened according to the sorting results to obtain selected variable data; the selected variable data is input into an improved temporal convolutional network structure for model training, and the model parameters with the best prediction hit rate of molten iron silicon content obtained through the test set are combined with the improved temporal convolutional network structure as a prediction model for molten iron silicon content in a blast furnace. It can be seen that the prediction process of molten iron silicon content based on the improved TCN structural model in this embodiment includes: first, the data after missing value interpolation uses the data translation variable construction method to characterize the time lag phenomenon, and then the data is input into the neural network model for training, and the hit rate of the model on the verification data is calculated in real time, and finally the hyperparameters of the model are determined according to the effect of the hit rate of the verification data. It can be seen that in this embodiment, the data input length of the interpolation model and the prediction model are first kept consistent, and at any time when prediction is required, data consistent with the input length of the model data is input, and then after the data enters the model, automatic outlier processing, missing value interpolation, correlation calculation and prediction tasks will be performed. At the same time, by regularly triggering training tasks, collecting full data for interpolation model, prediction model training and correlation calculation, the training process will be executed multiple times, and finally retaining the model parameters with the highest hit rate for future prediction tasks.

[0106] According to the above records, in an exemplary embodiment, the selected feature recent data is input into the prediction model, and the process of predicting the silicon content of molten iron in the blast furnace at a future time includes: taking each data of the selected feature recent data as an independent variable, and dividing it according to a fixed length and step size to obtain multiple one-dimensional data blocks; using multiple one-dimensional convolution kernels to perform convolution calculations on each data block to construct and generate multiple feature vectors; splicing the multiple generated feature vectors on the long side to form a two-dimensional matrix, and overlapping and splicing the two-dimensional matrix formed by splicing the feature vectors of each data block to form a three-dimensional matrix; using a one-dimensional deep convolutional network to mine the time dependence of the feature vector splicing result, and using a layer of one-dimensional point-by-point convolutional neural network to mine the interaction relationship between single variable features, and using a layer of one-dimensional point-by-point convolutional neural network to mine the interaction relationship between cross-variable features, and performing residual connection; mapping the residual connection result to the length of the prediction time through a layer of fully connected neural network to obtain the prediction result of the silicon content in the blast furnace at a future time.

[0107] In summary, the present invention provides a blast furnace molten iron silicon content prediction system, which obtains the full amount of data in all time periods, and the full amount of data includes blast furnace sensor data, blast furnace raw material data, and blast furnace inspection and testing data; then outlier processing is performed on the full amount of data, and missing value interpolation is performed based on the outlier processing result; and, based on the missing value interpolation result, correlation calculation is performed on the molten iron silicon content, and the correlation calculation results are sorted, and data is screened according to the sorting result to obtain the selected feature full amount data; then the selected feature full amount data is input into the temporal convolutional network structure for model training to generate a prediction model; and the selected feature recent data is input into the prediction model to predict the blast furnace molten iron silicon content at a future moment. Among them, the selected feature recent data includes the full amount of data within a preset time period, and the preset time period takes the current moment as the final moment and the time length is less than all time periods. It can be seen from this that this system can predict the silicon content in molten iron in real time by performing time series data outlier processing, intelligent interpolation of missing values, automatic correlation analysis and online prediction of silicon content in molten iron on all data such as blast furnace sensor data, blast furnace raw material data, blast furnace inspection and testing data, thereby improving the current time lag problem in predicting the silicon content in molten iron and improving the prediction effect of molten iron silicon content.

[0108] It can be understood that the blast furnace iron silicon content prediction system provided in the above embodiment and the blast furnace iron silicon content prediction method provided in the above embodiment belong to the same concept, wherein the specific manner in which the blast furnace iron silicon content prediction method performs operations has been described in detail in the above method embodiment, and will not be repeated here. In actual applications, the blast furnace iron silicon content prediction system provided in the above embodiment can allocate the above functions to different functional modules as needed, that is, divide the internal structure of the blast furnace iron silicon content prediction system into different functional modules, and then implement all or part of the functions of the corresponding functional modules through the blast furnace iron silicon content prediction method described in the above embodiment, and no specific restrictions are made here.

[0109] The embodiment of the present invention further provides a computer device, which may include a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program so that the computer device executes Figure 2 The steps of the method for predicting silicon content in blast furnace molten iron. Figure 5 FIG. 1 is a schematic diagram showing a structure of a computer device 1000. Figure 5 As shown, the computer device 1000 includes: a processor 1010 , a memory 1020 , a power supply 1030 , a display unit 1040 , and an input unit 1060 .

[0110] The processor 1010 is the control center of the computer device 1000. It uses various interfaces and lines to connect various components, and executes various functions of the computer device 1000 by running or executing computer programs / instructions stored in the memory 1020, thereby monitoring the computer device 1000 as a whole. In the embodiment of the present invention, when the processor 1010 calls the computer program stored in the memory 1020, the following is executed: Figure 2 The steps of the method for predicting silicon content in molten iron of a blast furnace. Optionally, the processor 1010 may include one or more processing units; preferably, the processor 1010 may integrate an application processor and a modem processor, wherein the application processor mainly processes the operating system, user interface and application, etc., and the modem processor mainly processes wireless communication. In some embodiments, the processor and the memory may be implemented on a single chip, and in some embodiments, they may also be implemented separately on separate chips.

[0111] The memory 1020 may mainly include a program storage area and a data storage area, wherein the program storage area may store an operating system, various applications, etc.; the data storage area may store instruction data, etc. created according to the use of the computer device 1000. In addition, the memory 1020 may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other non-volatile solid-state storage devices, etc.

[0112] The computer device 1000 also includes a power source 1030 (such as a battery) for supplying power to various components. The power source can be logically connected to the processor 1010 through a power management system, so that the power management system can manage functions such as charging, discharging, and power consumption.

[0113] The display unit 1040 may be used to display information input by the user or information provided to the user and various menus of the computer device 1000, and in the embodiment of the present invention, is mainly used to display the display interface of each application in the computer device 1000 and objects such as text and pictures displayed in the display interface. The display unit 1040 may include a display panel 1050. The display panel 1050 may be configured in the form of a liquid crystal display (LCD), an organic light-emitting diode (OLED), etc.

[0114] The input unit 1060 may be used to receive information such as numbers or characters input by the user. The input unit 1060 may include a touch panel 1070 and other input devices 1080. The touch panel 1070, also known as a touch screen, may collect touch operations of the user on or near it (such as operations of the user using a finger, a stylus, or any other suitable object or accessory on or near the touch panel 1070).

[0115] Specifically, the touch panel 1070 can detect the user's touch operation, detect the signal brought by the touch operation, convert these signals into touch point coordinates, send them to the processor 1010, and receive and execute the command sent by the processor 1010. In addition, the touch panel 1070 can be implemented in various types such as resistive, capacitive, infrared and surface acoustic wave. Other input devices 1080 can include but are not limited to one or more of a physical keyboard, function keys (such as volume control keys, power on / off keys, etc.), a trackball, a mouse, a joystick, etc.

[0116] Of course, the touch panel 1070 can cover the display panel 1050. When the touch panel 1070 detects a touch operation on or near it, it is transmitted to the processor 1010 to determine the type of touch event, and then the processor 1010 provides a corresponding visual output on the display panel 1050 according to the type of touch event. Figure 5 In the embodiment, the touch panel 1070 and the display panel 1050 are two independent components to implement the input and output functions of the computer device 1000, but in some embodiments, the touch panel 1070 and the display panel 1050 can be integrated to implement the input and output functions of the computer device 1000.

[0117] The computer device 1000 may further include one or more sensors, such as a pressure sensor, a gravity acceleration sensor, a proximity light sensor, etc. Of course, according to the needs of specific applications, the computer device 1000 may also include other components such as a camera.

[0118] The embodiment of the present invention further provides a computer-readable storage medium, in which a computer program / instruction is stored. When the computer program / instruction is executed by a processor, the above-mentioned device can perform the present invention as follows: Figure 2 The steps of the method for predicting silicon content in blast furnace molten iron.

[0119] It can be understood by those skilled in the art that Figure 5 This is only an example of a computer device and does not constitute a limitation on the device. The device may include more or fewer components than shown in the figure, or combine certain components, or different components. For the convenience of description, the above parts are divided into modules (or units) according to their functions and described separately. Of course, when implementing the present invention, the functions of each module (or unit) can be implemented in the same or multiple software or hardware.

[0120] Those skilled in the art should understand that the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code. The present invention is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be applied to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the process. Figure 1 A process or multiple processes and / or boxes Figure 1 These computer program instructions can also be stored in a computer-readable memory that can guide a computer or other programmable data processing device to work in a specific way, so that the instructions stored in the computer-readable memory produce a product including an instruction device, which implements the functions specified in the process. Figure 1 A process or multiple processes and / or boxes Figure 1These computer program instructions can also be loaded onto a computer or other programmable data processing device, so that a series of operation steps are executed on the computer or other programmable device to produce a computer-implemented process, so that the instructions executed on the computer or other programmable device provide for implementing the process in the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0121] The above embodiments are merely illustrative of the principles and effects of the present invention, and are not intended to limit the present invention. Anyone familiar with the art may modify or alter the above embodiments without departing from the spirit and scope of the present invention. Therefore, all equivalent modifications or alterations made by a person of ordinary skill in the art without departing from the spirit and technical concept disclosed by the present invention shall still be covered by the claims of the present invention.

Claims

1. A method for predicting silicon content in molten iron of a blast furnace, characterized in that: The method comprises the following steps: Obtaining full data in all time periods; wherein the full data includes blast furnace sensor data, blast furnace raw material data, and blast furnace inspection and testing data; Performing outlier processing on the full amount of data, and interpolating missing values ​​based on the outlier processing results; and, Based on the missing value interpolation results, the correlation of the silicon content in molten iron is calculated, the correlation calculation results are sorted, and the data is screened according to the sorting results to obtain the full amount of data of the selected features; The full amount of data of the selected features is input into the improved temporal convolutional network structure for model training to generate a prediction model; and the recent data of the selected features is input into the prediction model to predict the silicon content of molten iron in the blast furnace at a future moment; wherein the recent data of the selected features includes the full amount of data within a preset time period, and the preset time period takes the current moment as the final moment and the time length is less than that of all the time periods.

2. The method for predicting silicon content in molten iron of blast furnace according to claim 1, characterized in that: The process of performing outlier processing on the full amount of data includes: The full amount of data is processed offline and Kalman smoothing is performed to calculate the confidence interval, and the following is obtained: In the formula, x i is the true value of the full data, is the smoothed value of the full data after Kalman smoothing, δ i is the variance of the Kalman smoother, α is the significance level, Z α2 is the quantile position, n is the length of the full data; Data points outside the confidence interval are removed as outliers to obtain an outlier processing result.

3. The method for predicting silicon content in molten iron of blast furnace according to claim 2, characterized in that: The process of missing value interpolation based on outlier processing results includes: The outlier processing result is concatenated with the missing value matrix, mask matrix, and mask missing matrix introduced in advance or in real time; and the concatenated result is input into a neural network interpolation model, and vectorized and positionally encoded using at least one layer of linear neural network; wherein the neural network interpolation model is generated based on a self-attention mechanism structure; The total amount of data in all time periods is added to the position encoding results and then masked, and the masked results are input into the multi-layer self-attention mechanism structure; At least one layer of linear neural network is used to map the output data of the self-attention mechanism structure to obtain the missing value interpolation result; wherein the output data after dimension mapping has the same data dimension as the full amount of data in all time periods.

4. The method for predicting silicon content in molten iron of blast furnace according to claim 3, characterized in that: Introduce the missing value matrix M in advance or in real time, as follows: In the formula, represents the value of the dth row and tth column in the missing value matrix, Represents the full data value corresponding to the dth row and tth column in the missing value matrix, the missing value matrix M∈R T×D , D = max(d), T = max(t); The mask matrix i is introduced in advance or in real time, and we have: In the formula, represents the mask data value corresponding to the dth row and tth column in the mask matrix, Represents the value of the dth row and tth column in the mask matrix, mask matrix I∈R T×D ; Introduce mask missing matrix in advance or in real time have: In the formula, Represents the mask data value corresponding to the dth row and tth column in the mask missing matrix. The mask missing matrix 5. The method for predicting silicon content in molten iron of blast furnace according to claim 4, characterized in that: After obtaining the missing value interpolation result, the method further includes: The loss function value is calculated based on the missing value matrix, the mask matrix and the mask missing matrix, and the loss function value is: In the formula, L represents the loss function value, represents the missing value interpolation result, The calculated loss function value is used for back propagation to update the neural network interpolation model.

6. The method for predicting silicon content in molten iron of any one of claims 1 to 5, characterized in that: The whole amount of data of the selected features is input into the improved temporal convolutional network structure for model training. The process of generating a prediction model includes: Based on the fixed hyperparameters debugged during the offline data processing, outlier processing is performed on the full amount of data of the selected features through a Kalman smoother; Based on the fixed hyperparameters debugged during the offline data processing, the missing values ​​of the full amount of data of the selected features are interpolated online through the neural network interpolation model; Perform correlation calculation on the missing value online interpolation results, sort the correlation calculation results, and screen the data according to the sorting results to obtain the selected variable data; The selected variable data are input into the improved time series convolutional network structure for model training, and the model parameters with the best prediction hit rate of molten iron silicon content obtained through the test set are combined with the improved time series convolutional network structure as a prediction model for blast furnace molten iron silicon content.

7. The method for predicting silicon content in molten iron of blast furnace according to claim 6, characterized in that: The process of inputting the selected characteristic recent data into the prediction model and predicting the silicon content of molten iron in the future comprises: Taking each item of the selected characteristic recent data as an independent variable and segmenting it according to a fixed step length to obtain a plurality of one-dimensional data blocks; Use multiple one-dimensional convolution kernels to perform convolution calculations on each data block to construct and generate multiple feature vectors; The long sides of the multiple generated feature vectors are spliced ​​to form a two-dimensional matrix, and the two-dimensional matrix formed by splicing the feature vectors of each data block is overlapped and spliced ​​to form a three-dimensional matrix; A one-dimensional deep convolutional network is used to mine the temporal dependency of the feature vector concatenation results, and a one-layer one-dimensional point-by-point convolutional neural network is used to mine the interaction relationship between single-variable features, and a one-layer one-dimensional point-by-point convolutional neural network is used to mine the interaction relationship between cross-variable features, and residual connections are performed; The residual connection results are mapped to the length of the prediction time through a layer of fully connected neural network to obtain the prediction result of the silicon content in the blast furnace molten iron at the future moment.

8. A blast furnace iron molten silicon content prediction system, characterized in that: The system comprises: A data acquisition module is used to obtain the full amount of data in all time periods; wherein the full amount of data includes blast furnace sensor data, blast furnace raw material data, and blast furnace inspection and testing data; An outlier processing module, used for performing outlier processing on the full amount of data; Missing value interpolation module, used to interpolate missing values ​​for the full data after outlier processing; A correlation calculation module is used to perform correlation calculation on the silicon content of molten iron according to the missing value interpolation results, sort the correlation calculation results, and screen the data according to the sorting results to obtain the full amount of data of the selected features; A prediction module is used to input the full amount of data of the selected features into the improved temporal convolutional network structure for model training to generate a prediction model; and input the recent data of the selected features into the prediction model to predict the silicon content of molten iron in the blast furnace at a future moment; wherein the recent data of the selected features includes the full amount of data within a preset time period, and the preset time period takes the current moment as the final moment and the time length is less than that of all the time periods.

9. A computer device, characterized in that: It comprises a memory, a processor and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the method for predicting the silicon content in molten iron in a blast furnace as claimed in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: A computer program is stored thereon, and when the computer program is executed by a processor, the steps of the method for predicting the silicon content in molten iron in a blast furnace as claimed in any one of claims 1 to 7 are implemented.

Images